Organizing Grammar: Linguistic Studies in Honor of Henk van Riemsdijk

;��jl, ' t", ..

Der Mann { den ! "'dass! '" 0 } du gesehen hast ist reich . Ich habe die Stadt { wo! *dass! *0 ) Hans wohnt besucht. Hans hat von dem Tag { wann! *dass! *0 } er nach Amerika gehen wird erzaehlt.

Generalized Local Identification. Let y= a. or SPEC(a). We say that po identifies a fea�re F of y if and only if F is present in po, where

:: �_::-r�?��£:' ':

some projections W and a! are sisters.

�[� 'i ';

All the marked canonical features of relative pronouns, including any

wme explicit syntacticon entry correspond to a canonical realisation of any ' :�;'�f' : null relative pronouns, such as those of English: f

>'(23)

English null relative pronouns: a. SPEC(CP), ANIMATE, III PERSON, -_; b. SPEC(CP), PLACE; 1t = { where/ 0 }

1t =

{ who/ 0 }

We thus have found more examples of null morphemes in the Syntacticon, . in addition and akin to the Italian nul l object clitic and Van Riemsdijk's WG null verbs of motion.

References Bresnan, Joan. 1 976. On the Form and Functioning of Transformations. Linguistic Inquiry 7, 3-40. Chomsky, Noam and Howard Lasnik. 1 977. Filters and Control. Linguistic Inquiry

8. 425-504.

Emonds, Joseph. 1979. Appositive Relatives have no Properties. Linguistic Inquiry

1 0. 2 1 1 -243.

Emonds, Joseph. 1987. The Invisible Category Princ ip le Linguistic Inquiry 1 8,

6 13-632.

.

1 26 Joseph Emonds Emonds, Joseph. 2000. Lexicon and Grammar: the English Syntacticon. BerlinINew York: Mouton de Gruyter. Emonds, Joseph. 2003 . Two Types of Lexical Access: Why Syntacticon Items can be Null. Theoretical and Applied Linguistics at Kobe Shoin 6, 1 -24. Hoekstra, Jarich. 1 997. The Syntax of Infinitives in Frisian. Doctoral dissertation. Fryske Akademy, Leeuwarden. lackendoff, Ray. 1 983 . Semantics and Cognition. Cambridge, Massachusetts: MIT Press. Lieber, Rochelle. 1 98 0. On the Organization of the Lexicon. Doctoral dissertation. Massachusetts Institute of Technology, Cambridge, Massachusetts. Riemsdijk, Henk van. 2002. The unbearable lightness of GOing. Journal of Comparative Germanic Linguistics 5 . 143-1 96. Rizzi, Luigi. 1986. Null Operators in Italian and the Theory of pro. Linguistic Inquiry 1 7. 50 1 -557.

.�•, :�

Long-distance reciprocals . . Martin Everaert 1. ,

Introduction

I

paper, giving an overview of Binding Theory, Asudeh and appear) state: (to rymple Dal

In a recent

A puzzle that has gone largely unaddressed in the literature on binding is the local nature of reciprocal binding. Though there are many examples of reflexive pronouns that need not be locally bound, there seem to be no comparable examples of long-distance reciprocals.

:,;§�'; d it other words, what you typically get is the following opposition. In �;��i'::' languages allowing ' long distance binding' ('non-local binding'), such as �a�f Kannada (cf. ( I a» and Dutch (2a), binding of the reciprocal outside the ' tri:{. ;: Jrt inimal governing category is excluded (cf. (l b) and (2b); Amritaval li ;:�§�m:2000: 67,89; Everaert 1 986: 2 1 4-2 1 8) : 2 ' "�" (I) a. [shyaama tannannuj priitisuttaane anta] raamaj heeLidanu , ...-.......,.. �,., .. ,

. ;-���;

Shyama selfacc loves that Rama said 'Rama said that Shyama loves him (=Rama).' b. *makkaLuj [naanu obbaranna obbarui baide anta] heeLidaru children I oneacc onenom scolded that said 'The children said that I scolded one another.'

"

.. :.' :�:::�� ' ,

•••_ �. , "

.� ,

y y '

n "

I

I would like to thank Alexis Dimitriadis, Eric Reuland and Anca Sevcenco for

------

their comments, help. �; Note the example in

(i), from Broekhuis ( 1 994), in which the reciprocal is non­ J()cally bound. Following Pollard and Sag (I 994) one m ight take this as a case of binding of an exempt anaphor.

(i)

Ziji )jeten mij stiekem elkaarsi dagboek lezen

'They let me read each other's diaries in secret'

1 28 Martin Everaert

(2) a. Ziji

lieten m ij voor zichi werken. They let me for themselves work 'They let me work for them.' b. *Ziji lieten mij voor elkaari werken. They let me work for each other

Are Asudeh and Dalrymple right in saying that this puzzle has gone largely unaddressed? Not quite. To begin with, the observation itself has been around for quite some time. Even without having access to something as helpful as a 'Case study on Reciprocals' (cf. Everaert and Van Riemsdijk's The Blackwell Companion to Syntax for an explanation of the notion of 'case', I was immediately able to come up with the following references.3 Yang ( 1 983) is one of the very first to discuss the fact that reciprocals are, cross- linguistically, strictly locally bound. Yang argues for a Reciprocal-Binding principle that is slightly different from that for reflexives: a reciprocal is, in essence, bound within the domain of its first c­ commanding subject. Everaert ( 1 986) examines the distributional properties of reciprocals in a number of Germanic languages. He argues ( 1 986:2 1 8) that the fact that Dutch anaphors like zichzelJ (himself/herself/etc.) and elkaar (each other) behave like quantified NPs m ight account for the fact that they are necessarily locally bound. In the outline of their questionnaire for a typology of anaphoric dependencies in South Asian languages, Lust et al. (2000:875) ask their authors to see whether reciprocals allow long-distance binding, resulting in observations as in ( 1 ) above. But there is at least one paper that I know of that precisely tries to address the issue raised by Asudeh and Dalrymple. It is Henk van Riemsdijk's 1 985 paper Why long reciprocals don 't exist in the journal

Theoretical Linguistic Research 4 And I am sure 1 am missing numerous other references. Although Van Riemsdijk quite precisely describes what he takes as instances of ' long-distance reciprocals" it is good to make clear that in the context of this discussion we do not include the phenomenon of ' long distance' or 'wide scope' reciprocals in cases such as:

3

4

(i)

John and Mary think they like each other.

Higginbotham ( 1 980) argues that this sentence has two interpretations, of which the interpretation "John thinks ' I like Mary' and Mary thinks ' I like John'" could be taken as a long-distance binding of the reciprocal. Dimitriadis (2000), a.o., shows that this interpretation could very well follow from ' local' binding given a proper analysis of the semantics of plural pronouns.

Long-distance reciprocals 1 29

It is a well-known fact that many languages have so-called long reflexives. Long reflexives are bound anaphors in that, unlike pronouns, they require an antecedent within a sentence. They are ' long' in that the antecedent, given standard assumptions, occurs outside the minimal governing category containing the reflexive. Principle A of the binding theory is a principle that deals with bound anaphors. In other words it is meant to apply not only to reflexives but also to reciprocals. In most approaches to long reflexives, modifications of principle A play an important role. One would therefore expect these modifications to extend to reciprocals as well, i.e. one would expect there to be long reciprocals under precisely those conditions under which long reflexives are permitted. But the truth of the matter appears to be long reciprocals simply do not exist. Probably due to the fact that the journal Theoretical Linguistic Research ,�eased to exist quite soon after it appeared, the paper has gone largely 'hnnoticed. In this paper I wiJI reassess the discussion of the point that Henk so poignantly sketched in the quote above.

,,,,,,, ,::,,,,.r.:. Reciprocals as 'anaphors' Crucial in Van Riemsdijk's argumentation above is the assumption that " ��ifl:\ �i·bciprocals and reflexives, both being anaphors, should have a similar ,j�,�iA distributional pattern. By and large this happens to be the case (cf. Everaert \;��n:, :2000), but there are some cases where their distribution diverges, as the , examples in (3) illustrate (Chomsky 1 98 1 , Lebeaux 1 983, Kuno 1 987): ,, �'"

,

.'

" (3)

a. TheYi bought each otheri' s/*themseIvesj' books. b. [John and Mary]i didn't think that ?each other/*themselvesj would leave early. c. TheYi think it bothered ?each other/*themselvesi that ... d. TheYi made sure it was clear to ?each other/themselvesj that ...

The most straightforward option to account for the (un)grammaticality of the examples in (3) is that the binding theory itself doesn't make a 'difference between reflexives and reciprocals but that such distributional differences are the result of independent principles interacting with the 'binding principles. Case considerations, i.e. lack of genitive case for reflexives, might, for instance, play a role in the case of (3a) (as has been argued for in Bresnan 2000, a.o.). For the example in (3b) Lebeaux ( 1 983)

1 30 Martin Everaert

and Chomsky ( 1 986) argue that reflex ives but not reciprocals move at Logical Form, resulting in the violation of another principle of grammar (the ECP). For the cases in (3c,d), however, it is not immediately clear how we could account for the distributional diff erences. So Van Riemsdijk's point is clear. Setting the ex amples in (3) apart, there is something to explain. And there are some suggestions in the literature (at least, that I know of) how to go about it. Let me first begin with Van Riemsdijk's analysis. Van Riemsdij k ( 1 985) observes that in the original formulation of the notion Governing Category (Chomsky 1981), there are two notions ' subject' relevant for determining what is a governing category: the NP serving the grammatical function subject ([N P,S]), and INFLlT ense. The former, Van Riemsdijk argues, one can take as a +Theta position, the latter as a -Theta position. He, f urthermore, notes that reciprocals (for some reason) are generally assigned a theta role (+ Theta), while refl exives are not necessarily (± Theta): Jean se lave/Jean parle de lui-meme. Subsequently he argues that the binding conditions could be made sensitive to theta-marking specification. +Theta-marked anaphors must take theta-marked antecedents, i.e. grammatical subjects, while -Theta-marked anaphors take non-theta-marked antecedents (!NFL/Tense). Suppose one f ollows the suggestion in A nderson (1982) that in certain complements (infinitives, subj unctives) !NFL/Tense is dependent (co-indexed) upon the matrix INFLl Tense, extending the govern ing category beyond the immediate domain. In this way, the differences between reflexives and reciprocals are ex plained. The configuration of long-distance binding will h e limited to anaphors that can take INFL/T ense as its accessible subject, i.e. reflexives (depending on their ±Theta-specif ication). Reciprocals, however, necessarily in need of a +theta subject will always be locally bound by the su bject of the clause containing the anaphor. Everaert ( 1 99 1 ) offers an explanation of why 'co mplex reflexives' like zichzelJ would be barred from long-distance binding, contrary to ' simplex reflexives like zich. H e argues that the f ormer type of anaphor is specifi ed <+ A,+ P> while the latter is specified <+A,-P>. D efining the not ions 'governing category' l 'm inimal governing category' relative to the A(naphor)- and P(pronominal)-f eatures, respectively, it is derived that <+ A,+P> reflexives, bound in some governing category and in their minimal govern ing category, are necessarily locally bound, while <+ A,-P> reflexives, bound in some govern ing category and not bound in their minimal govern ing category, are not. If it could be argued that reciprocals, cross-linguistically, have to be necessarily specified as <+ A,+P>, bound in

Long-distance reciprocals 1 3 I

their minimal governing category, the lack of long-distance reciprocals would follow. Whatever the precise reason might be to argue for such an analysis (I refer the reader to the discussion in Everaert 1991: I 04-6, on reciprocals), the parallel between (certain) complex reflexives and reciprocals is central to this analysis: both do not participate in long­ distance binding . There is another line of reasoning that m ight explain why reciprocals are locally bound. In Belletti (1982) it is argued that the Italian reciprocal I 'uno I 'altro (cf. 5) should be analysed as a complex element containing a floating quantifier (cf. 4):5

(4) a. I miei amici hanno parlato tutti dello stesso problema. 'My friends spoke all of the same problem.' b. [NP [tuttiJ [NP i miei amicilil hanno parlato [pp ej [pp dello stesso problema]] (5) a. I miei amici hanno parlato Puno dell'altro. 'My friends spoke of each other.' b. [NP [I' uno]j [NP i miei amici]jl hanno parlato [pp ej [pp dell' [NP altro]] It is precisely this mechanism that is also invoked by Heim, Lasnik and May (1991). They argue that in the case of English each other, each moves at logical form, leaving behind an NP-trace. The clause boundedness of this type of reciprocal m ight, thus, be triggered by the presence of an NP-trace, forcing strict locality under the assumption there is no long-distance NP­ movement.

3. Are there instances of long distance binding of reciprocals? So far we assumed that the descriptive generalization that reciprocals do not allow long-distance binding is well-established. But is it? ,S

It is tempting to assume that, ultimately, the morpho-syntactic status of reciprocals might be responsible for the fact that reciprocals are clause-bound. However, Polish examples as in (i) (Reinders-Machowska 1991) show that such a position cannot be straightforwardly upheld: (i)

a. b.

ChlOPCYi

czytali

0

The boys read

dziewcZO\tj wspomnienia the gir ls' memories

about them/themselves

ChlopcYi

dziewcUl,tj wspomnienia

o sobie*i/j

czytali

The boys read

the girls' memories

sobieilj

about each other

(= the girls;::fi the boys)

1 32 Martin Everaert

In the Lust et al. (2000) volume mentioned above, there are scarce remarks about possible instances of long-distance binding of reciprocals. In the chapter on Hindi/Urdu Davison (2000:433) gives the following example as marginally acceptable. ?[raam aUf syaam]j sooc-tee hai [ki ravi-nee kahaa (6) Ram and Shyam think-IMPF are that Ravi sayperf [ki eek duusraa-neei paagal hail that one another-M.SG crazy is 'Ram and Shyam think that Ravi said that each other is crazy. ( Ram and Shyam each think that Ravi said that the other was crazy' '

=

Given the fact that in an identical environment the reflexive apnaa cannot take an antecedent across a finite clause boundary (Davison 2000: 4 1 8), this is noteworthy. In the chapter on Marathi Wali (2000:55 5-556) shows that long­ distance bound reciprocals are excluded in finite clause embeddings (7a), but allowed in nonfinite clauses (cf. 7b): (7) a. tyaa lokaJi-naaj vaaTta ki ekamekaa-ni*j dagaD those people-OAT feel that each other-ERG stones phek-l-et. threw-AGR 'Those people feel that each other threw the stones.' b. tyaa 10ki'Hi-nii shaam-laa [ekamekaa-naai boekaaraay-laa those people-ERG Sham-OAT each other-DAT pinch-INF laavla. forced 'Those people forced Sham to pinch each other.' This, however, may be expected, since the Marathi' reflexive swatciah is also excluded in finite complements, but allowed in infinitives (Wali 2000:530, 534).

Long-distance reciprocals 133

ravij swataah-laa.ilj dosh l ilii mhaNaali ki Lil i said that Ravi self-DAT blame 'Lili said that Ravi blames self.' shaam-Iaaj [swataahlaai/j bockaaraay-laa] b. liIi-nii pinch-INF Lili-erg Sham-DAT self-DAT 'Lili forced Sham to pinch self.'

deto. gives laavla. forced

So, in this case reciprocals behave similar to, at least, one type of reflexive, as predi cted by the binding theory. Japanese might give another example of long-distance reciprocals. Japanese has several strategies to encode a reciprocal relation: the use of . the anaphoric element otagai (cf. 9a), the use of aw, a verb creating a complex predicate marking the predicate as reciprocal, simultaneously forcing the suppression of an (overt) object position (cf. 9b), or a combination of the two (cf. 9c) (Kobuchi-Phi lip p.c.):

.

;(9) a.

John to Mary ga otagai-o seme-ta. John and Mary-NOM each other blame-PAST b. John to Mary ga seme-aw-ta. John and Mary-NoM blame-RecM-PAST seme-aw-ta. c. John to Mary ga otagai-o John and Mary- NOM each other blame-RecM-PAsT 'John and Mary blame each other.'

Nishigauchi (1992) argues that the Japanese reciprocal is in fact a composite reciprocal, as argued for in Heim, Lasnik and May (1991). That 'is, the reciprocal consists of a distributor part and a reciprocator part, as exemplified in ( 10). (10)

[John and Mary] eachil like [ ej other ] distributor reciprocator

Nishigauchi argues that the reciprocal marker construction (9b) consists of aw as an element with a distributor function, licensing an empty element with the reciprocator function. In the case of the reciprocal (pro )noun construction (9a) otagai is analysed as a complex anaphor consisting of an

134

Martin Everaert

empty D-operator with the distributor function making it parallel to English each other: [[e]D otagai]/[[each] other].6 Nishigauchi (1992: 1 59) observes that otagai is strictly locally bound (cf. Ila), but that there are some exceptions, as the example in (lIb) shows, indicating that an 'animacy' condition on intervening antecedents might be responsible for the locality: (ll) a. * [John to Mary]i-ga Bill-ga otagai-o seme-ta to. John and Mary-NoM Bill-NOM each other-ACC accused-PAST that 'John and Mary thought that Bill accused each other.' b. [John to Mary]i-ga kono ziken ga otagaii-o John and Mary-NoM this incident NOM each other-ACC kizu-tuke-ta to omow-ta. that thought hurt 'John and Mary thought that this incident would hurt each other.' (12) gives a case of long-distance reciprocalization containing only the reciprocal marker:

(12)

?[John to Mary]i-ga zibun-ni sensei-tatij ga [e]i/*j home-ta John and Mary- NOM self-GEN teacher-PL NOM praise- PAST to zyasui si-aw-ta. that suspect do-RecM-PAST , John and Mary each suspected that selF s teachers praised the other.'

An interesting case is given in (13), which shows that the place of the reciprocal marker, on the matrix verb, determines the scope of reciprocal (pro)noun otogai. (13)

6

kai-ta tegami-o [John to Mary]j-ga Bill ga otagaij-ni John and Mary-NOM Bill NOM each other-DAT wrote letter-ACC yomi-aw-ta. read-RecM-PAST 'John and Mary each read the letter that Bill wrote to the other.'

Nishigauchi (1992:192) seems to suggest that this distributive operator is absent in cases where otagai is combined with the reciprocal marker, as in ( 1 3).

Long-distance reciprocals 135 4.

Concluding remarks

It is not at all clear that, on an observational level, it is true that long­ distance reciprocalization is blocked. We can only hope that systematic cross-linguistic research will clarify this issue.7 But suppose, for argument's sake, that the generalization holds. What would be a possible explanation? The above discussion gives some indication. If a language encodes a reciprocal relation via a (verbal) reciprocal marker, a non-local interpretation would be straightforwardly blocked by an independent mechanism such as the Lexical Integrity hypothesis, the reciprocal marker being part of the verb's morphology (cf. Everaert 1991 ). If the reciprocal is argumental, the semantics of reciprocity might be responsible for it being necessarily a verbal operator, as suggested (for complex reflexives) in Reinhart and Reuland ( 1 991 :291). That would mean that, in this respect, the behaviour of reciprocals is not so much different from the reflexivizers triggering strict local binding such as Icelandic sjaljur sig, Japanse zibun zisint or Dutch zichze!f . :References Asudeh, Ash and Mary Dalrymple. To appear. Binding Theory. In: Encyclopedia of language and linguistics (2nd edition), ed. by Keith Brown. Amsterdam: Elsevier. Belletti, Adriana. 1982. On the Anaphoric Status of the Reciprocal Construction in Italian. The Linguistic Review 2:101-138. Bresnan, Jo an 2000. Lexical-Functional Syntax. Oxford: Blackwell. Broekhuis, Hans. 1994. The Referential Properties of Noun Phrases I. Modern .

Grammar ofDutch, vol. I. Tilburg University. Dimitriadis, Alexis. 2000. Beyond Identity: Topics in Pronominal and Reciprocal

Anaphora. Doctoral Dissertation, University of Pennsylvania. Everaert, Martin. 1986. The Syntax of Rejlexivization, BerlinlNew York: Mouton de Gruyter.. Everaert, Martin. 1991. Contextual Detennination of the Anaphor/Pronominal Distinction, In Long-distance Anaphora, ed. by Jan Koster and Eric Reuland, 77-118. Cambridge: Cambridge University Press.

7

I know of two projects that aim at such a typology: the DFGINWO-funded project A Typology of Reciprocal Markers: Analysis and Documentation (Konig, Everaert. Gast and Dimitriadis) and the Australian Research Council-funded project Reciprocals across Languages (Evans, Nordlinger).

1 36

Martin Everaert

Everaert, Martin (2000). Types of Anaphoric Expressions: Reflexives and Reciprocals, In Reciprocals. Forms andjunctions, ed. by Zygmunt Frajzyngier and Traci Curl, 63-83. Amsterdam: John Benjamins. Everaert, Martin and Henk van Riemsdijk, eds. To appear. The Blackwell

Companion to Syntax, Volume I-V, Oxford: Blackwel l.

Heim, Irene, Howard Lasnik, and Robert May. 1 991. Reciprocity and P lurality. Linguistic Inquiry, 22 :63 - 1 0 1 . Higginbotham, James. 1 98 1 . Reciprocal Interpretation. Theoretical Linguistic Research 1 :97- 1 1 7. Lust, Barbara C., Kashi Wali, James W. Gair, and K.V. Subbarao, eds. 2000.

Lexical Anaphors and Pronouns in Selected South Asian Languages. A Principled Typology. BerlinlNew York: Mouton de Gruyter. Nishigauchi, Taisuke 1 992. Syntax of Reciprocals in Japanese. Journal of East Asian Linguistics 1: 157- 1 96. Pollard, Carl and Ivan Sag. 1 994. Head-Driven Phrase Structure Grammar.

Stanford: CSLI, and Chicago: The University of Chicago Press. Reinders-Machowska, Eva. 199 1 . Binding in Polish. In Long-distance Anaphora, ed. by Jan Koster and Eric Reuland, 1 3 7- 1 50. Cambridge: Cambridge University Press. Reinhart, Tanya, and Eric Reuland. 1 99 1 . Anaphors and Logophors: An Argument Structure Perspective. In Long-distance Anaphora, ed. by Jan Koster and Eric Reuland, 283-3 34. Cambridge: Cambridge University Press. Riemsdijk, Henk van. 1 985. Why Long Reciprocals Don't Exist. Theoretical

LingUIstic Research 2:37- 45 . Yang, Dong-Whee. 1 983 . The Extended Binding Theory of Anaphors, Research 1 9: 1 69- 1 92.

Language

:

The notion of topic and the problem of ''''>::''';'--a ntifica tion in Hungarian

'1��1i1:;i��1�!;alllna Gecseg and Ferenc Kiefer :1:, Introduction . is a commonly accepted view that the syntactic structure of the ;;;�:�C" Hungarian sentence is determined by its information structure (E. Kiss 20 02, Killman 2001, Szabolcsi 1997). The following scheme is assumed to ,. 'account for word order in the preverbal domain of Hungarian sentences: . Topic* - distributive quantifier* - focusNM - verb - postverbal (1) material It

""."" ,

".".

"

.,"

. The left-most position of the scheme is occupied by the topic(s) (identified '"'''' with the logical subject) of the sentence which is usually defined as an f§�j;,expression denoting an individual (or set of individuals) which the ��;i:',predicate of the sentence asserts something about. The referent of the topic must be given in the discourse. The predicate (or the comment) of the sentence comprises everything that follows the topic in the scheme ( 1 ). " The second position in the sentence - which is the first position of the comment/predicate - is occupied by quantifiers with distributive reading: '\;.,

'

i�lj�,!�

A ' gyerekeket "minden kutya megugatta. the childrenacc all/every dog barked at 'Every dog barked at the children' (3)

"SokJlegalilbb "negy dhik 'megbukott a vizsgan. . many/at least four student failed the exam-at 'Many/at least four students failed at the exam'

Sentence (2) has a topic-comment structure: the topic is a gyerekeket 'the children' and the comment is minden kutya megugatta 'were barked at by every dog' containing the universal quantifier (with distributive reading) �minden kutya 'every dog' In contrast� (3) contains the distributive ' :quantifier sok ditikllegaltibb negy ditik ' many students/at least four students' in the preverbal field: the sentence is claimed to be topicless and _

1 3 8 Zsuzsanna Gecseg and Ferenc Kieler

to constitute a complex predicate as a whole. (VM stands for 'verbal modifier', which may alternate with focus .) In order to identify the topic(s) and the comment in a sentence, various phonological and/or syntactic criteria/tests have been used. For example, in Hungarian the first constituent of the predicate is claimed to bear the first obligatory full stress of the sentence. Applying this criterion to (2)-(3), the result is as follows : in (2) the sentence-initial definite NP can be stressed or unstressed, while the universal quantifier bears the first full stress of the sentence. By contrast, (3) contains a sentence-initial quantifier which is always stressed, hence - according to this criterion - it must be the first element of the predicate. Another test frequently applied is to insert adverbials such as tegnap 'yesterday', az utcan 'in the street', etc. into the sentence. The claim underlying th is test is that only topics may precede such adverbials : (4) a. A gyerekeket tegnap minden kutya megugatta. the childrenacc yesterday all/every dog barked-at 'Every dog barked at the children yesterday' megugatta. b. * A gyerekeket minden kutya tegnap yesterday barked-at the childrenacc every dog

(5)

*Soklt5bb mint negy dia k tegnap megbukott a vizsgan. many more than four student yesterday failed the exam-at 'Many/more than four students failed at the exam yesterday '

In the following sections we are going to show that the claim that sentences containing sentence-initial quantifiers are topiciess is rather problematic. We are going to propose an alternative analysis based on th e hypothes is that the notions ' logical subject' and 'topic' must be kept apart: the former is definable in terms of the logico-syntactic structure of the sentence, whereas the latter is a pragmatic notion. 2.

Some problems with current analyses of Hungarian quantifiers

2.1. A quantified .ftentence doesn't always express a thetic judgment Sentences which have a topic-predicate structure are also referred to as sentences express ing a categorical judgment, in s hort as categorical sentences . Sentences without such an articulation are sentences expressing a thetic judgm ent, in short thetic sentences (Kuroda 1 972, Ladusaw 1 994). The fact that thetic sentences serve to report an event (they are also called

The notion oftopic and the problem of quantification 139

event-reporting sentences, cf. Lambrecht 1994) explains why such .. sentences can function as answers to questions such as What happened?, . What is the matter? and What's up? We saw in the Introduction that sentences containing a sentence-initial are generally considered to be topicless. But if they are topicless, ntifier qua are thetic sentences, hence they must be able to function as possible to questions of the kind mentioned above. Yet this is not always wers ans . e: .•. the cas ( 6) Q: Mi tortent? 'What happened?' AI: Minden diak megbukott a vizsgan. 'Every student failed at the exam' A2: ??Sok diak megbukott a vizsgan. 'Many students failed at the exam' Mi ujsag? 'What's up?' AI :#Minden gyerek 'Every child A2:#Sok gyerek 'Many children

szereti like szereti like

a csokoladet. chocolate' a csokohidet. chocolate'

In (6) the answer AI, which contains a universal quantifier, is fully . acceptable whereas the answer A 2 with the indefinite quantifier sDk diak 'many students' is questionable, to say the least. On the other hand, the incorrectness of the dialogue (7) shows that the semantics of the predicate, too, may have an effect on the information structure of the sentence: as • -Ladusaw (I994) has pointed out, an individual-level predicate (for instance the two-place predicate szeret 'likes' can only occur in a categorical sentence.. Consequently, the answers AI and A2 in (7) must have a topic. By contrast, AI in (6) is topicless since it may be a correct answer to the . question 'What happened?' .

1 40 Zsuzsanna Gecseg and Ferenc Kiefer .. 2.2.

Sok N 'many N' doesn 'I behave uniformly with respect 10 the prosodic criterion and sentence adverbial test

Comp are now (8) and (9): (8)

"Sok gyerek 'elke sett. many child arrived-late 'Many children arrived late'

(9)

'Sok gyerek ma r "reggel megerkezett. many child already morning arrived 'Many children arr ived already in the morning'

Note that in their most natural intonation, the first fully stressed e lement is the quantifier sok ' many' in (8) and the adverbial reggel 'in the morning' in (9). We saw above that in Hungarian - according to the prosodic criterion the first element of the p redicate is claimed to bear the first obligatory main stress of the sentence. If this claim is correct, the quantifi er sok 'many ' belongs sometimes to the top ic, as in (9), and sometimes to the predicate, as in (8). However, both sentences, (8) and (9), can app ear in a context where something is asserted about children. A s for the sentence adverbial test, it shows, too, that the sentence­ initial NP sok gyerek has not an equal status in ( l 0) and ( 1 1 ): ( 1 0)

(11)

*Sok gyerek tegnap elke sett. many child yesterday arrived-late 'Many children arrived late yes� erday' Sok gyerek tegnap ma r reggel mege rkez ett. ' many chi ld yesterday already morning arr ived ' Many children arrived already in the morning y esterday'

In sum, then, whether the qu' anti fier sok N belongs to th� top ic or not seems to dep end on the context, as well. 2.3. NPs ofthe type a legtobb N 'most Ns ' have an unclear status concerning topicality There is an NP, namely a iegt6bb N 'most Ns' which is treated in two different ways by standard app roaches. Szabolcsi ( 1997) considers it as a topic expression whi le E. Kiss ( 1999) claims that it is a distributiv e

The notion oftopic and the problem ofquantification 141

,.allan1t11ll�r and as such, it is included in the predicative part of the sentence. :;;Ar�lme�nts can be advanced in favor of each of the two approaches. First, a legtobb N is an inherently distributive quantifier, similarly to . universal quantifier, which cannot have a collective reading: A legtobb gyerek felemelte az asztalt. (only the distributive reading is available) 'Most of the children lifted the table' * A legtobb gyerek korbeallta az asztalt. 'Most of the children surrounded the table'

:;Ii;im;:; Second, according to E. Kiss (1999), a legtobb N doesn't serve to refer to :,�D: '3n individual but to realize a counting operation over a set of individuals, figl�;: ':similarly to the universal quantifier. ��;fE :., However, in view of its syntactic and prosodic properties, this NP must '�;�(�;j,e a topical expression. In fact, a legtobb N can be followed by a definite :<,i':"""'NP in the preverbal domain, which is considered to be a topic constituent: A legtobb ember Petert orultnek tartja. holds the most man Peteracc mad 'Most people take Peter to be a fool' :l� addition, the prosodic criterion and the sentence adverbial test, too, show �at /egiObb N behaves in this respect like a topical expression: A legtobb di..k tegnap Marira szavazott. o d n jfj�::;;;;:' !�o�t :���:� !o��S�: �a ��ste����

f[�,;t15)

a

�

'T:IT}m�.3. Our proposal: topicality is a pragmatic notion

It!;". ��; �:�:�cc�::!��;e�:�;���:�:;!���!�fe�:�E���: i

o

current theories of Hungarian syntax as referring to an individual present in the discourse situation such that something is predicated about it in the sentence (Szabolcsi 1997, Eo Kiss 1999, 2002). Yet if one claims that a sentence introduced by a distributive quantifier is topicless, one seems to forget the fact that per definitionem any assertive sentence must assert something about something - by virtue of the fact that it is always uttered

142 Zsuzsanna Gecseg and Ferenc Kiefer --

in a discourse context. To put it differently, the coherence of the discourse requires a certain continuity of information from one utterance to another. This continuity is in fact guaranteed by the (i mplicit or explicit) topic of the sentence, denoti ng the individual about which something is asserted. Even a thetic sentence has a topic - thoug h a very abstract one, corresponding to the implicit spatio-temporal indications i denti fying a situation recoverabl e fr om the context. For instance, the thetic utterance It is raining, too, has a topic-comment structure. The (implicit) topic denotes a particular situation, which can be identified by means of the default spatia-temporal indications 'here and now' and the comment corresponds to the string it is raining denoting the property of ' being rainy' . The notion of topic should thus be treated as a pragmati c categ ory, which means that the identification of the topic of a particular sentence shouldn't be based on its syntactic (and log ical) structure but on the properties of the possi ble contexts in which it can appear. The prosodic criterion and the adverbial test, which are used to identify the topic of a sentence, are based on the assumption that the topic must be an explicit (phrasal) constituent of the sentence, where thi s latter is taken to be an abstract unit, which does not depend on the context. Yet the notion of aboutness implies that the information structure of the sentence does depend on its di scourse context. The distinction between categorical and thetic sentences, often referred to in Hungarian syntax and semanti cs, orig inates in a distinction between two types of context corresponding to two types of question-answer pairs. In the same way, the idea underlying the well-known topic-test of questioning (Hatcher, 1 953 ) i s to find an appropriate elementary context for the tested sentence, where the topic is identifi ed as the common (nominal) part of the sentence and the wh­ question it precedes. If we apply the question-test to quantified sentences, the result is as follows': a gyerekek? ( 16) Q: H ova tii ntek where disappeared the children 'Where have the children gone?' AI: Minden gyerek hazament. home-went all child 'All the children went home' I Note that AI can be uttered without any context, whereas be preceded by another utterance.

A2 and A3 must always

The notion oftopic and the problem of quantification 143

A2: A legtobb gyerek setalni ment. the most child to walk went 'Most of the children went out for a walk' A3: Legalabb harom gyerek a kertben jatszik. at least three child the garden-in plays 'At least three children are playing in the garden' In (16) the common part of the question-answer pair is not the whole NP ' but the noun gyerek 'child'. Therefore, only the nominal part of the NP (denoting a contextually determined set of children) belongs to the topic, which can even be replaced by a phonetically null pronominal element: ( 1 7) AI: Mind hazament(ek). home-went(pl) all 'All went home' A2: A legtobbUiik) setalni ment. the most(-of-them) to walk went 'Most of them went out for a walk' A3: Legahibb harman a kertben jatszanak. at least three the garden-in play , At least three of them are playing in the garden' confirms Lambrecht's hypothesis that preferred topic expressions are unaccented pronominals. In this sense null pronominals are the best 'candidates for topicality. . We saw in the previous section that NPs of the type sDk N 'many Ns' are sometimes claimed to display the properties of so-called topic expressions, in other cases they are considered to be part of the predicate of the sentence. The question-test applied to (9), where sok N was claimed to be a topic, reveals that it is impossible to find a natural context in which the common part of the question and the answer is sok N: (18) Q:??Mikor erkezett sok gyerek? when arrived many child 'When did many children arrive?' A:#Sok gyerek mar reggel megerkezett. many child already morning arrived 'Many children arrived already in the morning'

144 Zsuzsanna Gecseg and Ferenc Kiefer .. The problem with this question is that it is a 'pseudo-question': it presupposes somehow that the speaker of the question in (18) knows that many children arrived at a certain moment. Moreover, the dialogue violates the rule of coreference: once a new referent is introduced into the discourse by means of an indefinite NP, any subsequent mention of the same referent must be expressed by a definite NP, since two indefinite NPs denoting the same individual cannot appear in succession in a coherent discourse. The strange character of the dialogue (18) doesn't mean that there is no appropriate question which could be answered by a sentence such as A in (18); the same is true for sentence (8): (19)

Hany gyerek kesett el? 'How many children were late?'

(20)

Mikor erkeztek a gyerekek? 'When did the children arrive?'

The sentence (8) can be used to answer the question ( 1 9) and the sentence (9) to answer the question (20). The common property of these question­ answer pairs is that their topic is a contextually determined set of children. This set of individuals is denoted by the nominal part of the quantifier. Notice, however, that (19) and (20) are different questions in the sense that the former asks about the number of children sharing a certain property and the latter is a question about a property of a certain group of children. This reflects the fact that (8) and (9) have different logical structures - though they can have the same topic in a given context.

4. Conclusion In this paper we were looking more closely at the information structure of Hungarian sentences containing a distributive quantifier. We argued in favor of the hypothesis that these sentences, too, have a topic though their topic-comment structure is not reflected in their syntactic structure. Topicality is essentially a pragmatic notion, which necessarily presupposes a discourse context. As a consequence, the topic of a particular sentence should be identified on the basis of available contexts, and not on the basis of the syntactic properties of the sentence. The scheme (1), however, remains valid if 'topic' is replaced by 'logical subject'.

The notion o/topic and the problem ofquantification 145

rJ.j

Kiss, Katalin. 1 999. Mondattan. [Syntax] In magyar nyelvtan. [New Hungarian grammar], ed. by Katalin E. Kiss, Ferenc Kiefer and Peter Siptar, ,

1 7- 1 84.

Budapest: Osiris

Kiss, Katalin. Press. Hatch er, Anna O.

Kiado.

2002. Syntax of Hungarian.

Cambridge: Cambridge University

1 95 3 . Syntax and the Sentence. Word 12: 234-250. 200 1 . Magyar leiro nyelvtan 1. Mondattan. [Hungarian descriptive U'�'��'i:,,:Kalman, gramm ar. Syntax] Budapest: Tinta Kiado. :,;;2������kuroda, Sige-Yuki. 1 972. The categoric and thetic judgment: evidence from

!��!;;;,"":'

Laszlo.

:]r�f:;i�i:,,: Japanese syntax. Foundations o / Language 9: 153- 1 85. ':' ';;l�;i� :Ladusaw, William. 1 994. Thetic and categorical. stage and "'0:",'" '

,,

individual, weak and

strong. In Proceedings from Semantics and L inguistic Theory 4, ed. by Mandy "H : Harvey and L1ynn Santelmann, 220-229. Ithaca: Cornell University Press. ; ,i�L ;jl�]�!T,ij�ambrecht, Knut. 1994. Information Structure and Sentence Fo rm. Cambridge: ,

7��iE,,,

'

Cambridge University Press.

,��;�EF1;,i:$zabolcsi, Anna. 1997. Strategies for scope taking. In Ways of Scope Taking, ed. by Anna Szabolcsi, 1 09-1 54. Dordrecht: Kluwer.

: �:s::�� �''-' ;�' ;"�fg

���£��;:

��"; ij�J ; , �;I��g' "

.'

!" " , , ,"';:-;�;�""'.

Questions of complexity Casper de Groot

1.

Summary

In this paper I will raise two interrelated questions concerning morphosyntactic differences that hold between Hungarian spoken inside Hungary and Hungarian spoken outside Hungary. The first question relates to the structure of words: why tend the varieties of Hungarian outside Hungary to prefer less articulated words than Standard Hungarian? The second question relates to the order of constituents: why do pre-nominal non-finite relative clauses in Standard Hungarian become post-nominal adpositional modifiers in the varieties outside Hungary? 2.

Morphology

Studies on Hungarian spoken outside Hungary (Austria, Slovakia, Ukraine, Romania, Vojvodina, Prekmurje, Slovenia, United States and Australia as presented in Fenyvesi ed. (2005)) reveal interesting information on the change of Hungarian as a minority language. Most differences - if not all between standard Hungarian (HH) and Hungarian outside Hungary (HO) can be explained in terms of language contact where HO takes over features from adjacent languages, which all happ�n to belong to the Indo-European ·language family. De Groot (2005) argues that a number of the differences observed follow linguistic universals and implicational hierarchies, and that the co-occurrences of changes can actually be· explained in terms of universals or hierarchies. However, for the systematic change froffi,sYnthetiC:.to .analytical expressions in the varieties of HO there are no typological studies available that could form the basis for an explanation. Where HH uses morphologically complex, synthetic forms, HO shows a preference for the use of morphologically simplex, analytical forms, as in table 1. Note that the preference for analytical forms applies in all three major components of

Questions of complexity 147

i.e. inflection (modality), derivation (reflexive and causative), d· and compoun mg. I J. Synthetic versus analytical expressions in Hungarian inside :';;','Olltsl(le Hungary (De Groot 2005: 365)

", . .,.",nrnh.nlngy, .

inside Hungary

t!ulnganan outside Hungary

Ki tud-ok

out be.abie-ISG

Szepit-kez-ett

Szepit-ette

beautify-REFL-PAST.3SG .INOEF 'She beautified herself.'

'

men-ni? go-INF

maga-t beautify -PA ST.3SG .OEF oneself-ACC

She b eautifie d hers elf.

'

a

szek-et

Meg-rajzol-tat-ta PVB-draw-CAUS-PAST.3SG.OEF

permit-PAST.3SG.DEF the chair-ACC

the chair-ACC 'S/he had the chair

rajzol-ni. draw-INF 'S/he had the chair

a

szek-el.

tag-ietszam member-number

'number of members'

Hagy-ta

ietszam-a tag-ok member-PL number-3SG.POSS num ber of members' '

As for a general view on morphological change, Hock and Joseph (1996: 183) posit the following: The fate of morphology from Sanskrit to its modern descendants gives credence to the common belief that languages tend to develop in cycles: from isolating to agglutinating, from agglutinating to inflectional (through amalgamation of different affixes into one), from inflectional 'to isolating (through sound change and analogy), and so on. " Schematically: (1) agglutinating -+ inflectional -+ isolating I Abbreviations: 1 SO: fi st person singular; 380: third person singular; ACe: accusative; CAUS: causative; DEF: definite conjugation; HH: Hungarian inside Hungary; HO: Hungarian outside Hungary; INDEF: indefinite conj ugation; INF: infinitive; INSTR: instrument; MOD: modality; Nder: derived noun; PL: plural; PRES. PART: present participle; POSS: possessor; REFL: verbal reflexive; REL : relative marker; SUP: superessive case ending; V der: derived verb

r

148

Casper de Groot

When we consider the varieties of Hungarian outside Hungary, these varieties still contain all morphological properties of an agglutinating language (cf. szepitette, szeket, and letszama in Table 1). The change from HH to HO then does not follow the chain of gradual changes in (1) as attested in many languages, but rather seems to be a change from morphologically complex to morphologically simplex, i.e. from an agglutinating language with highly articulated words to an agglutinating language with less articulated words. If we take complexity to be a typological parameter, the following subtypes should be distinguished: ( 2)

complex a. articulated h. more lexemes

--4-

simplex -+ less articulated -+ less lexemes

An illustration of ( 2a) is example (3) which shows that the forms in HH are more articulated than the preferred forms in equivalent expressions in HO. Compare: (3)

HH . a rajzol-tat-t-a 1 2 34 draw-CA US-PAST-3SO '(S)he had it designed' b. busz-oz-as I 2 3 bus- Vder-Nder 'bus trip'

HO a'. hagy-t-a rajzol-ni 2 1 23 1 let"PAST-3SG draW-INF '(S)he had it designed. ' b'. utaz-as bus-szal 1 2 1 2 travel-Nder bUS-lNSTR 'bus trip'

An illustration of ( 2b) is example (4) which shows that compounds in HH, consisting of two lexemes, have counterparts in HO where the two lexemes form the basis of two words. Compare:

(4)

HH tag-Ietszam 1 2 member-number 'Number of members'

HO tag-ok letszam-a 1 1 member-PL number"3so.poss 'number of members'

Questions of complexity

149

Interestingly, while HO prefers to use forms which are morphologically . . complex than the counterparts in HH, the syntax of the varieties of HO more complex. Note the infinitival construction in (H03a), which is to all sorts of syntactic operations relevant to verbal clusters in the of E. Kiss and Van Riemsdijk ( 2004), where (HH3a) does not have options, because it is no instance of a verbal cluster. Example ( H03b) ;.;;:,�:;;gllrO""S a construction of a nominal with a post nominal modifier, whereas b) consists of just one nominal. Finally, where HH in (4) combines ' :\;�::/�;: "two lexemes in a compound, HO employs two constituents, united in a �;,j:possessive construction. ,� In conclusion of this section, one may wonder why language change by ,;.;:;;;;:[> . i.e. a ::t���\� ';',language contact may involve morphological of words, which at the same time ; r ; p reference to reduce the more complexity in syntax? ·

=.'

Jess

.-

...-�-

�;::�

•

::� ��H ecessitates :�:w:',;:n ;'�I�;\n,:i3. Syntax

articulatedness

simplification,

'''',:'';�Y'-''''''':'', � � -----,�

.

...

�,

·The second question I will raise here, concerns the preference of certain �:ill.f'B�'ii ;" post nominal modification in the varieties of Hungarian outside Hungary. ?:{R�h;;.;£.n,· equivalents of English the picture on the wall or the letter from Mary :j��=:=��tfr.' �dard Hungarian employs non-finite pre-nominal constructions to �express the modification of the nominal head. For instance:

'i�_

!!���i:

a fal-on levo kep the wa -sup COP.PRES.PART picture 'the picture on the wall' [lit. ."the on the wall being picture"]

ll

;::�r:T::' "'.,"

\

Examples such as (5) pattern along with non-finite a :

s

az eneklo t he 'the singing girl'

participle constructions

lany

sing.PRES.PART girl

Constructions as (5), like those as in (6), can be taken to be non-finite relative constructions ( cf. De Groot 1989: 191). They both have finite ·90unterparts, which take the form of finite relative constructions: .(7) a. a kep amely a fal-on van the picture REL the wall-sup coP.3SG 'the picture which is on the wall'

150 Casper de Groot

b. a hiny aid enekel the girl REL sing.3SG 'the girl who sings' When we have a look at data from the varieties of Hungarian outside Hungary, we see that in HO there is a preference for using another construction which widely differs from the one in standard Hungarian. Instead of a non-finite pre-nominal relative construction, we find a post­ nominal adpositional phrase, as in (8): (8)

a kep a fal-on the picture the wall-SuP 'the picture on the wall'

The difference between the form used in HH (cf.(S» and HO (cf. (8» may be captured in the following fashion:

(9)

HH

Sentential modifier + Noun

HO Noun + Non-sentential modifier

Interestingly, both differences, i.e. the application of the copula in HH and the non-application of the copula in HO, and the order of the constituents, i.e. Mod N in HH and N Mod in HO, must apply at the same time. The non­ application of the copula without the change of the position of the modifier yield an ungrammatical output in HO (cf. (lOa». The same holds if there is a change in the position of the modifier together with the application of the copula (cf. ( lOb»:

(10)

HO a. *a fal-on kep the wall-,sup picture 'the picture on the wall' leva b. *a kep a fal-on the picture the wall-suP COP.PRES.PART 'the picture on the wall'

have not found any theoretical or typological studies in the linguistic literature that explains the change in the usage of constructions of type (5) into constructions of type (8) by language contact, nor the co-occurrence of the distribution of the copula and the position of the modifier by language

Questions of complexity 151

'contact. It is known that relative constructions in languages with basic SOY . word order can take two positions in the noun phrase, before or after the nominal head. The position is almost without exception conditioned by the nature of the relative clause, being non-finite or finite. Non-finite relative . precede the head, whereas finite relative clauses follow the head. This typo logical observation may explain the non-occurrence of (lOb) in HO, but is by far not sufficient to explain all differences.

4. Conclusions The data presented in this paper suggest that Hungarian in contact with other languages becomes less complex from a morphological point of view and more complex from a syntactical one. This may be an instance of 'trade orr, a notion often heard in discussions concerning language complexity in the following sense. If a component in the grammar of a language is relatively simple, i.e. few distinctions and few rules, some other component will be relatively complex. I doubt this view, at least in the case of Hungarian, because Hungarian in contact does not lose much of its morphological complexity whereas the increase of syntactic complexity is significant. A second observation concerns the co-occurrence of two henomena, the usage of the copula and the order of constituents in the noun phrase, for which no straightfotward explanation is available. I conclude that data from Hungarian in contact with other languages offer interesting new material for the study of language contact, morphology and syntax and the relation between these fields.

p

References E.

Kiss, Katalin and Henk van R i ems dijk. 2004. 'Verb clusters. Some basis

notions'. In: Katalin E. Kiss and Henk van Riemsdijk (eds), Verb Clu�ters. A Study ofHungarian, German and Dutch, 1-40. Amsterdam: Benjamins. Groot, Casper de. 1 989. Predicate Structure in a Functional Grammar of Hungarian. Dordrecht: Foris. Groot, Casper de. 2005. 'The granunars of Hungarian outside Hungary from a linguistic-typological perspective.' In: Anna Fenyvesi (ed.),

contact outside Hungary: Studies on Hungarian

as

370. Amsterdam/New York: Benjamins. Hockt Hans H. and Brian D. Joseph. 1996. Language

and Language Relationship: BerlinlNew York:

Hungarian language a minority language, 35 1-

History, Language Change,

Mouton de Gruyter.

Functional heads, lexical heads and hybrid categories Liliane Haegeman

Introduction !

1.

It is a standard assumption that there exists an opposition between functional heads and lexical heads. However, as Van Riemsdijk has shown (see Corver and Van Riemsdijk 200 1 for discussion and references), the question arises whether a clear-cut dichotomy is tenable and, if it is not, how to handle 'hybrid' categories which are semi-lexical/semi-functional. This paper offers a case study of the dichotomy between functional and lexical categories. 2.

Lexical split: The case of Italian sembrare

Some speakers of Italian allow clitic climbing with sembrare 'seem ' (Burzia 1 986: 73 7; Beninca and Poletto 1 994: 40: (59a-c); Zushi 2005 ; Cinque 2004: 1 5 7, 1 7 1 , note 27). For those speakers, sembrare in (1) is a restructuring verb like modal, aspectual and motion verbs. (1)

Non 10 sembra capire. not it Seem3sg understand ' He doesn't seem to understand it. '

Cinque (2004) proposes that the monoclausal structure associated with restructuring is base-generated: That is to say: so-called 'restructuring' verbs are merged as functional heads in the extended projection (Grimshaw 1 99 1 ) of the 'main' lexical verb. Hence there . is simply one c lause. I

I dedicate this paper to Henk van Riemsdijk. In his o�n research he has set the

standards for all work in formal syntax . His work on Swiss German has initiated

the interest in dialect studies that has now become a salient feature of syntactic research. H enk has also created a world-wide network of formal linguists, which has become a 'safe haven' which enables syntacticians to pursue generative linguistics regardless of the sometimes hostile attitudes of their home institutions. I owe him thanks for having created this stimulating environment and also for

having encouraged me in my first steps in formal syntax, an encouragement he has maintained throughout.

Functional heads, lexical heads and hybrid categories

1 53

ing to Cinque, restructuring verbs are always functional (2004: 1 53 ) • Cinque' s h ierarchy of the functional heads that constitute IP is given in (2).

.

M ood Pspeech act > MoodPevaluative > MoodPevidential > ModPepistemic > TP (Past) > TP (Future) > MoodPirrealis > ModPalethic > ASPPhabitual > AspPrepetitive > AspPfrequentative > ModPvolitional > AspPcelerative > TP (Anterior) > AspPtenninative > AspPcontinuative > ASPPretrospective > Asp Pproximalive > AspPdurative > Asp Pgeneric/progressive > Asp Pprospective > M odPobligation > ModPpennission/ability > AspPcompletive > VoiceP >

ASpPcelerative > ASpPrepetitive > ASpPfrcquentative

(Cinque 2004: 1 3 3, his [3])

. In (3), potere 'may' triggers restructuring, as shown by the fact that the clitic /0, an object of non-finite incontrare 'meet', ends up on the finite . modal potresti. In C inque's approach, potere in (3) is merged as an Alethic Modal head (for arguments see Cinque 1 999: 78-8 1 ) in the extended . projection of incontrare 'meet' . The clitic 10 'it' targets a landing site in the rE2E�::';ex.terlae:a projection of incontrare, the verb which selects it. incontrare domani. him maYcond-2sg meet tomorrow ' You might meet him tomorrow.' Lo

potresti

'Restructuring' semhrare in ( I ) is merged in Evidential Mood (cf. Rooryck 2001 on evidentiality). Since functional heads do not have any argument ktructure, functional sembrare will lack an EXPERIENCER argument. However, as shown in (4a), for the relevant speakers, sembrare may also co-occur with an EXPERIENCER argument. Given the presence of the EXPERIENCER, semhrare cannot be functional in (4a) and must be analysed as a lexical verb. Not being functional, lexical sembrare in (4a) does not give rise to restructuring and hence clitic climbing is no longer possible, witness the ungrammaticality of (4b): (4) a. G ianni mi sembra capirlo. Gianni me Seem3sg understand it ' Gianni seems to me to understand it.' b. *G ianni me 10 sembra capire. Cinque assumes a strict dichotomy with:

1 54

Liliane Haegeman

. . . two mutually exclusive options: the obligatory presence of transparency effects in a monoclausal structure [cf (1)] and the obligatory absence of transparency effects in the biclausal structure [cf (4)] (Cinque 2004: 133). A verb is either functional (with restructuring) or lexical (without restructu ring). According to Cinque, then, for the relevant speakers, Italian has two verbs sembrare: One is functional, with ' restructuring' effects, the second is lexical, without 'restructuring' . Cinque detects a semantic difference between these verbs: As a lexical verb sembrare "literally means that a certain state of affairs seems to be true (to someone)". As a functional verb, sembrare is a marker of Evidentiality "which (mildly) commits the speaker to a certain state of affairs" (see Cinque 2004: 157). The different readings are illustrated in (5). (5) a. Gianni sembra a tutti apprezzarlo molto, Gianni Seem3sg to everybody appreciate-it much, ma io non credo che 10 apprezzi. but I don 't believe that he it appreciates 'To everyone Gianni seems to appreciate it a lot, but I don't think that he does. ' b. #Gianni 10 sembra apprezzare molto, rna io non credo . . . . Gianni it Seem3sg appreciate much, but I don't believe . . This squib explores some consequences of Cinque's proposal. Section 3 provides support for his analysis of semhrare. Section 4 examines French sembler, for which Cinque (2002: 634) allows for an analysis as a functional head. The French data provided here challenge this hypothesis. Section 5 introduces Dutch data which raise problems for the hypothesis that verbs are either functional, with restructuring, or lexical, without restructuring. 3. Functional sembrare as an Evidential head If sembrare with clitic climbing is an Evidential Mood head, we predict that it will not be found in the complement domain of functional heads located lower in the functional hierarchy (2). This prediction is correct. Given the presence of clitic climbing, (6a) must contain functional sembrare. But, being merged in Moodevidentiah sembrare could not be embedded under Modalethic, the head that hosts po/ere. In contrast, (6b) is

Functional heads, lexical heads and hybrid categories 155

:jn'amlm�ltlc:al because the absence of clitic climbing2 allows us to assume that lexical sembrare has been selected: · sembrare capire a. ?Lo potrebbe 10 maYcond-3sg seem understand (rna io sono sicura che non abbia capito niente). but I am sure he has not understood a thing Potrebbe sembrare capirlo (ma io sono sicura che non abbia capito b. mente) .3 'He might seem to understand-it, (but I am sure that he has not understood a thing.' •

Second, if sembrare is merged in Moodevidentiah it should be incompatible with perfective aspect, in T(Anterior). This prediction is borne out. (For complications see Haegeman 2005). *Non 10 sana sembrati capire. Non it be3pl seempart-pi understand �Bc�caluse functional sembrare encodes Evidential Mood, we also expect it to be ungrammatical in contexts in which Evidential Mood is excluded. Two such contexts are conditional clauses (8a) and clauses associated with fac1tlve predicates (9a) (see Haegeman 2005 for discussion of such contexts and for a syntactic account). In both these contexts, sembrare is incompatible with clitic climbing. (8b) and (9b) are grammatical because they can be taken to instantiate lexical sembrare. (8) a. * Se 10 sembrano trovare troppo difficile, if it seem3pl find too difficult, faremo il secondo capitolo. dOfut-l pl the second chapter b. Se sembrano trovarlo troppo difficiIe, faremo it secondo capitolo. 'If they seem to find-it too difficult, we' ll do the second chapter.'

i Cinque shows that absence of clitic climbing

may, but need not, correlate with absence of restructuri ng (2004: 1 49- 156). 3 Than ks to Susi Wunnbrand for pointing out the relevance of these data.

1 56 Liliane Haegeman

(9) a. nChe 10 sembrino b.

trovare troppo difficile, non mi sorprende. too difficult, non me surprise3SG that it seemsubj-3pl find Che sembrino trovarlo troppo difficile, non mi sorprende. that seemsubj-3p l find-it too difficult, non me surpriSe3sg 'It doesn't surprise me that they seem to find it too difficult. '

4. French sembler 'seem' is not a restructuring verb Cinque

(2002:634) suggests that, like Italian sembrare, French sembler

' seem' is merged as an Evidential functional head. However, the data below do not confirm this hypothesis. Because clitic c limbing is generally not available with French modals ( l Oa), it cannot serve as a diagnostic for restructuring. However, the clitics

en and y can very marginally undergo climbing in high literary styles ( l Ob,c) (cf. Cinque 2000: 6 1 9, note 4). However, none of my informants accept such climbing of en and of y with

sembler (1 1 ):

( l O) a. *ll l'a voulu voir. he it-have3sg wanted see b. J' en voudrais voir beaucoup. I-en wanlcond- lsg see many 'I would like to see many of them .' c. J'y voudrais aller demain. I -y wantcond- l sg go tomorrow.

' I would l ike to go there tomorrow. '

(1 I ) a. *11

he

h. *11

en a

semble

avoir parle_

ya

semble

aller. go

en have3sg seempart

he y have3Sg seempart

have talked

As pointed out by C inque (2002: 63 3), in French leftward floating of the quantifier tout 'tout a gauche' is possible with sembler ( 1 2).

( 1 2)

II

a tout sem ble comprendre. He have3sg all seem part understand 'He seemed to understand everything_ '

However, Cinque signals that this phenomenon is not a

bona fide (2002: 632 and 633). So ( 1 2) does not necessarily show that sembler is functional. I ndeed, in ( 1 2) sembler is used

d iagnostic for restructuring

Functional heads, lexical heads and hybrid categories

1 57

the perfect aspect. In this respect it differs from functional sembrare, which is not compatible with the perfect (cf. [7]). ( 1 1 ) and, contrary to ::Cilnqllle 's suggestion, ( 1 2) suggest that sembler is not merged in the ; E,ride�ntl;al head. The conclusion is confirmed by ( 1 3), which also illustrates gauche' with sembler. In ( 1 3), sembler is embedded under the Modal pouvoir 'may/might'. Since Modalethic is lower in the iLhilerarcl1lV than Moodevidcntial (cf. [2]), the fact that sembler embeds under 1V1(JUalethic implies that sembler cannot have been merged in Moodevidential . lIs peuvent tout sembler comprendre they maY3pl al l seem understand (mais est-ce que cela va durer?) (but will this last?) 'They may seem to understand everything (but will this last?)'

sembler is not a functional marker of Evidentiality is reinforced by the that my informants have no difficulty in using 'tout it gauche' with itse'mtJ'ler in conditional clauses (1 4a), and in clauses associated with factive icates ( 1 4b), contexts which are incompatible with Evidentiality:4

.

S1;; . .�:��;

.

S ' i ls avaient tout semble comprendre, If they havepast_3pl all seempart understand, je n'aurais pas du eprendre mon cours. I would not have had to repeat my c1ass ' If they had seemed to understand everything, I would not have had to repeat the class. ' b. Je suis contente qu' its ont tout semble com prendre. e a nders and t Ia he ap t t �d :::�iliing�

�

�";. :; �;� �:���� : ::!;

From these data we conclude that unlike Italian sembrare, French sembler l;[��m:iis not merged as an Evidential functional head in a monoclausal structure: �f;���m�Rather, it remains a lexical verb in a biclausal structure. This conclusion <;iY�ili;�!potentially challenges Cinque's sharp dichotomy between functional verbs, ' ; " , i n iWbich allow restructuring, and lexical verbs, which don't. If 'tout it gauche' '!:,t'iflMwere to be taken to be an indication of some kind of restructuring, then our , }i;:;:, ;'i���cbnclusion should be that sembler is a lexical verb which gives rise to

i Wl\

.

v.;

.

:�<

"

':�

' . : .. . Thanks to Jonathan Carre, Georgette Oal, Philip

for judgements.

Miller and Daniele van de Velde

Liliane Haegeman

1 58

restructuring (cf. Wunnbrand 2004): Sembler would belong to a ' hybrid' category with both lexical (merger in V) and functional (restructuring) properties. This issue does not arise, of course, if 'tout a gauche' is not s taken to be evidence for restructuring, which Cinque seems to assume. 5. Dutch scllijnen

and Lijken (cf. Van Bruggeo 1980, Wurmbrand 2004)

At first sight, standard Dutch has two analogues for sembrarelsemblerl seem: schijnen and lijken. For many speakers schijnen cannot, and lijken can, take an EXPERIENCER argument: (15) a. *Hij schijnt me een beetje onhandig te zijn in de omgang. he Seem3sg me a bit clumsy to be in social interaction b. Hij lijkt me een beetje onhandig te zijn in de omgang. to be in social interaction he Seem3sg me a bit clumsy 'He seems to me a bit clumsy in social interaction. '

Van Bruggen ( 1 980: 6 1 ) proposes that schijnen is more modal while lijken has more of a perceptual quality. Thus, schijnen would be similar to functional sembrare, and lijken is like lexical sembrare (cf. also Wurmbrand 2004: 1 002). If, following Cinque's analysis of functional sembrare, Dutch schijnen is a marker of Evidential Mood and lijken is a lexical verb, we correctly predict that schijnen will, and lijken won't, lead to a degradation in a conditional clause or in a complement of factive 6 pred lcates. · ( 16)

Je moet de dokter roepen :.. you must the doctor call a. ?? . . als zijn toestand schijnt te verslechteren. b. '" als zijn toestand lijkt te verslechteren. if his condition seems to get worse . . 'You must call th� doctor if his condition seems to .worsen.'

( 1 7) a. ??Ik vind het jammer dat hij dat niet schijnt te begrijpen. b. Ik vind het jammer dat hij dat niet lijkt te begrijpen. I find it a pity that he that not seems to understand ' I regret that he does not seem to understand that.' 5

For

Spanish see Torrego ( 1 996) and Ausin and Depiante (2000). For Catalan see Picallo ( 1 990) 6 For variation in judgements with factive predicates see Haegeman (2005).

Functional heads, lexical heads and hybrid categories 1 59

schijnen is merged as a functional head is confirmed by the observation that, like German scheinen in ( 1 8), it is incompatible with 7 ". 've aspect ( 1 9a). That speakers anow the verb lijken with the jje;rtec�t (I 9b) can then be seen as confinnation that for them lijken is a !fe){iciaJ verb, and thus should not trigger restructuring.

*Sie hat zu Hause zu arbeiten gescheintlgeschienen. seempart she has at home to work 'She has seemed to work at home.' (Wurmbrand 2004: 1 00 1 ( 1 4b» a. *Het postmodemisme heeft de grond onder de Postmodemism has the foundation under the wetenschappelijke traditie schijnen te willen wegvagen . scientific tradition 'seem' to want away-sweep b. Het postmodernisme heeft de grond onder de Postmodernism has the foundation under the wetenschappelijke traditie lijken te willen wegvagen. under scientific tradition ' seem' to want awaysweep 'Postmodernism seemed to want to remove the foundations of the scientific tradition. ' •

''

, at this point a complication arises. In ( l 9b) lijken displays the . .. . . ,=� :
. . . . . .

''t' �;�j:::: ' seemed' would not be possible.

�li?f��tt�_�: .

;::-�i�:��,:

:

;;i;:(!: ' . 7 My discussion is based on the judgements of Hans Broekhuis, Roland Noske and .• ,i i ·.�' · {\.Ibert Oosterhof. There is considerable variation among speakers. Arnold Evers, .. .

. ,. ..

for instance, apparently admits schijnen with the perfect and in conditional clauses (Evers 2004: 1 04, his [26a]):

(i) I

als h ij de vogels weg had schijnen te jagen . . . when he the birds away had ' seem' to chase

intend to explore speaker variation in future collaboration with Hans Broekhui� .

1 60 Liliane Haegeman

IPP typically arises in restructuring contexts (cf. Haegeman 1 998 and references cited there). Thus, in spite of it being merged as a lexical verb, the IPP effect with lijken suggests that this verb g ives rise to restructuring. This conclusion raises further doubts about the strict functional/lexical dichotomy: Apparently, in l ine with Wunnbrand (2004), lexical verbs may also give rise to restructuring. Lexical verbs giving rise to restructuring are ' semi-lexical' (Corver and Van Riemsdijk 2000) and are a hybrid category between purely lexical verbs and purely functional verbs. References Ausin, Alonso and Marcella Depiante. 2000. On the syntax of parecer 'to seem ' with and without an Experiencer. In Papers from the 3rd Hispanic Linguistics Symposium. Hispanic Linguistics at the Turn of the Millennium. Hector Campos, Elena Herburger, Alfonso Morales-Front and Thomas-J. Walsh (eds.). Somerville, MA :Cascadilla. Beninca, Paola and Cecilia Poletto. 1 994. Bisogna and its companions : The verbs of necessity. In Paths towards Universal Grammar. Guglielmo Cinque, Koster, Jean-Yves Pollock, Luigi and Raffaella Zanuttini (eds.), 3 5-57. Georgetown University Press. Bruggen. Fanny van 1 980. Schijnen, lijken, blijken. Tabu I I (3): 54-63 . Burzio, Luigi 1 986. Italian Syntax. Dordrecht: Kluwer Cinque, Guglielmo 1 999. Adverbs and Functional heads. Oxford /New York: Oxford University Press. Cinque, Guglielmo 2002. A note on ' restructuring' and quantifier climbing French. Linguistic Inquiry 33: 6 1 7-636. Cinque, Guglielmo 2004. ' Restructuring' and functional structure. In Structures and Beyond. The Cartography o/Syntac!ic Structures (Vol 3). Adriana Belletti (ed.), 1 32-] 9 1 . Oxford/ New York: Oxford University Press: Corver, Norbert and Henk van Riemsdijk (eds.). 200 1 . Semi-lexical Categories. BerIinlNew York: Mouton de Gruyter. Evers, Arnold. 2004. Move Head. In Yearbook 2004. Annemarie Kerkhoff, Joke de Lange, and Oren Sadeh Leicht (eds.), 93- 1 07. Utrecht I nstitute Linguistics OTS. Grimshaw, Jane 1 99 1 . Extended projection. Ms Brandeis University. Haegeman, Liliane 1998. V-positions and the middle field in West Flemish. Syntax I : 259-299. Haegeman, Liliane. 2005. Evidential mood, restructuring, and the status of sembrare. Silex. 1 990. verbs in Catalan. Natural Language and Linguistic

Jan

Rizzi,

in

of

Picallo, Carme.Ms. Modal Rooryck, No. 5,lohan.61 1 Theory 8: 285-3 1 2. 1

200 1 . Evidentiality. Glot International Vol. 5, No. 4: 1 25- 1 33 - 68.

and

Functional heads, lexical heads and hybrid categories 1 6 1 . Torrego, Esther. 1996. Experiencers and Raising verbs. In Current Issues in Comparative Grammar. Robert Freidin (ed.), 1 0 1 -120. Dordrecht: Kluwer. Wunn brand, Susi. 2004. Two types of restructuring - Lexical vs. functional.

Lingua 1 1 4 : 99 1 - 1 0 1 4.

i, Mihoko. 2005. Deriving the similarities between Japanese and Italian: a case study in comparative syntax. Lingua 1 1 5 : 7 1 1-752.

Concatenation and interpretation Martin Haiden

1.

Introduction 1

In the minimalist framework (Chomsky 1 99 5 , 200 la, 2004), any operation beyond bare concatenation must be virtually conceptually necessary in the sense that it is either an interface condition, or a general condition of computational efficiency (Chomsky, 2004). Phenomena that cannot be explained in this way are considered to be imperfections. Movement (the disp lacement property) is a well-discussed example of this reasoning. In earlier versions of the m inimalist program, it was considered an imperfection. In more recent implementations, it follows directly from the way concatenation is formalized. This is a major step towards, and Chomsky claims even beyond explanatory adequacy. Much less attention has been attracted by imperfections in the mapping to the interface components. This squib considers the structure-building formalism from this perspective. 2. The derivation of SEM

Two interpretational operations are generally assumed to exist: one of them, call it I, maps obj ects from syntax. into semantics. Another operation, call it c[J, maps syntactic objects into phonology. Chomsky observes that neither of these operations is trivial. Apart from transferring the output of the computational system to the respective interface, each operation must eliminate those features that belong to the other interface, respectively. Semantic features must be eliminated by c[J, phonological features by E. Although the elimination of un-interpretable features can 'be motivated as an interface condition, general conditions of computational efficiency favor a theory that delivers pairs directly, without deletion

I This squib is dedicated to Henk van Riemsdijk in recognition of his ability to address the most notorious empirical problems in an elegant theory of grammar. Acknowledging his superiority in combining the two goals, it humbly attempts to meet just one of them.

Concatenation and interpretation

1 63

' . . of mutually un-interpretable features. The formalism below is designed to :9.? that. 1.

Disentangling merge and head-of

Chc:>m!SKY ( 1 995) assumes that the concatenating operation merge takes two ��;:o t' lectS x, y, it forms a pair (or set), and it assigns a label (either x or y) to newly formed object. In effect then, merge is a complex operation. It sts of at least two components: pair/set formation and labeling. The components are defined independently in Bendjaballah and Haiden :2EE"" ;r'"n l''l ) , Haiden (2005). Bare merge takes two variables ranging over the . · ·SaIJ1e finite domain E, x,y E E2, and delivers the product xy E E:

,

,

Merge (Haiden 2005, p. 162) E2 � E

. (x,y) xy

p'

1-7

By definition, the output of p is in E, and it can merge with another ��!:vaflalble in E: p(p(x, y), z)) = p(xy, z) = xyz. As a product, p creates flat [��gstJiuctun;�s. The order of subsequent merging operations does not determine hierarchical relation among variables. Asymmetrical relations are introduced by a fi Itering operation that tdE�sl��nates the head by eliminating its dependent. This operation can be ��;;;:Ji1lodele:d as a sum. Head-of(Haiden 2005, p. 162)

LP(X,Y)== ex

y€ E

In (2), summation applies with respect to y, and y is thus elim inated, It replaced by a constant C, which corresponds to the sum of the values of '''': ':''' '';':'''' -:''. This operation models the designation of the head x ofxy by interpreting complement y. In standard minimalism, the interpretation of features requires an ,.. .. . . . . . ' independent operation r. In the present formalism, r delivers . interpretations as a by-product of the generation of asymmetrical relations: ' -. . : � � : -A given variable is designated as a complement by being interpreted. Only < :unvalued features (variables) remain derivationally active. The last variable to be interpreted corresponds to the root node. Its interpretation closes the . derivation.

1 64 Martin Haiden 2.2.

Complements, specifiers. andfeature checking

Let us consider an initial array of lexical items, represented as variables x,y,z. Assume x merges with y to form xy, and then head-of interprets y as a dependent of x. Next, z merges with the head x of xy, to form zx. (3)

�

z

x

by merge

x

by head-of

I

�

x

y

by merge

Now there are two options to proceed. Head-of can interpret x as a dependent of z. This yields a complementation structure (4). By contrast, if z is interpreted as a dependent of x, the result is a shell- or specifier­ structure (5). (4)

complement-structure z

by head-of

� z

x

by merge

x

by head-of

I

�

x

(5 )

y

by merge

specifier/shell structure

Both confi gurations are bare structures. They do not incorporate derivationally introduced objects like bar levels, or light heads. This is particu larly evident in the shell configuration. Multiple specifiers/shells can

Concatenation and interpretation 1 65

merged as long as the interpreted configuration is valid at the semantic

Merge may apply successively without the intervention of head-of If a e is not interpreted immediately, it ' incorporates' and 'moves' as of a complex head. A ssume an initial array of variables x,y,z,g. For :�?�r��'f�c,nclret�em::ss, let z be a verb, y a noun, and x a feature F of N that needs to ",==o:-,' hp. checked against g, a functional head G. For expository reasons, both the derivation, and an informal illustration with more familiar labels will given. First xy, and then zxy, are derived by merge. Then head-of interprets y . ... .... . Ft�jj(i\as a complement, while leaving x unaffected. This designates zx (not z ;!�j,g!L;�lone) as the head of zxy (6a). Informally speaking, we would say that an ' of feature F has percolated from N to V (6b) . 5���§k; index . .i:Eilii��i�.-. _,;,"

b.

zx

�

z

xy

.

Next, zx merges with g, forming gzx. Head-of interprets z as a lement of gx (7a). Informally, the index of feature F has percolated to next higher head G (7b). b.

gx

�

g

zx

�

z

xy

Finally, head-of interprets x as a dependent of g (Sa). This is similar to h t e deletion of a checked feature (Sb). Elimination of checked features is thus a further consequence of the generation of hierarchical relations. It 'does not need to be stipulated as a separate operation, i .e., an imperfection in Chomsky's terminology.

1 66

Martin Haiden

(8) a.

b.

g

I

z

xy

3. The derivation of PHON

Three observations come to bear on the formalization of the phonological interface. First, phonology concatenates objects to form asymmetrical prosodic structures (Halle and Vergnaud 1 987, Nespor and Vogel 1 986, Vergnaud 2004). This is true even' if we maintain that phonology is not recursive in any meaningful sense (Pinker and Jackendoff 2005). Second, prosodic structures are not identical with syntactic structures, although they tend to be similar in certain respects (cf. Cinque 1 993, Kiparsky 1 966, Zubizarreta 1 998). Third, the concatenation of lexical items X, Y in phonology must be identifiable as a concatenation of X and Y in syntax, even though their hierarchical relation might well be the opposite in the two domains. Observations one and two indicate that phonology has independent access to the structure-building algorithm of the computational system . In other words, there must be a component of phonological derivation. The postulation of a phonological component in this sense might appear to constitute what Chomsky (2004) calls the ''worst case" (p. 4): 2 the existence of independent cycles. The "best case", according to Chomsky, "is that there is a single cycle only" (p. 4), and that the "components of the derivation of proceed cyclically in parallel (p. 4). It will now be shown how representations PHON are generated, and then, in which way the derivation of can be . considered to constitute a single cycle.

2

Page numbers in citations of Chomsky (2004) refer to the earlier manuscript Chomsky (200 1 b).

Concatenation and interpretation 1 67 3. 1.

Bare prosodic structure

In the preceding section, it was argued that the mapping to SEM is a natural ence of the structure building algorithm itself. The same is now . con sequ attempted for the mapping to PHON : merge concatenates timing units, bead-of introduces asymmetrical prosodic structure by interpreting dependent units as segments. . Let us assume with Lowenstamm ( 1996) that the primitive unit of . prosodic structure is a single consonant position C followed by a single : vocalic position V. Now consider a complex syllabic constituent as exemplified by the English word aim. This constituent must be represented py three CV units, C V/, CV2, C V3. These CV units are manipulated by the structure building algorithm exactly like syntactic objects. In the particular example, merge applies twice to create the sequence CV,CV2CV3 • Next, head-of interprets C V3 as the complement (i.e., coda) of the complex nucleus CV1CV2 and spells it out as a segment [m] . Then, head-of applies . , :again. It interprets CV2 as the complement (i.e., off-glide) of the nucleus Hl-=r;g.} GV/, and it spells it out as [j]. CV, is thus the head, which can merge with ;;t" �d)ther CV units to form complex prosodic structures. it does not undergo ,,IF-��;;: : any further merger, it is itself spelled out as [e] by head-of (9) illustrates . . . " th e derivation. \. Notice that this is a bare prosodic structure. It does not contain any Al.erivationally created object like coda, onset, or syllable. It is thus possible I�;;.: , to generalize the inclusiveness condition (Chomsky 1 995) to the derivation PHON, an unexpected, but welcome result (cf. Chomsky 200 1 b/2004,

If

'

.

'

';'�t�:: ()f

�[ii,�..: �.4).

,�"�"§:�" "�';�L' ·(.··'..9).

Bare prosodic structure: aim [ejm] C V C V C V

c v c v c v

I I

e

j

I

m

�.2. Mapping. and the Projection Principle The phonological and syntactic components are not independent. Let us assume that their relation i � constrained by a generalized version of the Projection Principle adapted from Haiden (2005, p. 1 58):

1 68 Martin Haiden ( 1 0)

Projection Principle Lexically

stipulated

pairs

throughout a derivation.


SEM>

remain

unaltered

The Projection Principle can only apply to complex objects,

concatenation of two lexical items

X, Y

if the

in phonology is recognizable as a

concatenation of X, Y in syntax. This is the case, if the mapping from phono logy into syntax is structure-preserving for

that this is the case. (1 1)

a.

Mapping PF

fA�B where

---4

merge.

3

Let us assume

syntax

A n B=0

b. f A � B is structure preserving for p 4. Global Economy The formalism sketched above honors the

inclusiveness condition

10

Its

most general form, and it delivers bare output representations PHON, SEM

without redundancy. The same is not true of theories that relegate all morpho-phonological processes to a shallow, post-syntactic component.

Proponents of the latter position must concede that at least "the features of

certain Roots are visible in the syntactic derivation" (Embick 2000, p. 226).

The component of morpho-phonological processing is therefore assumed to be split (Halle and Marantz

1993),

or, more appropriately, doubled. In fact,

more or less every stage in a derivation triggers certain phonological

processes (Kiparsky 1 982, Mohanan 1986). On the minimalist assumption

that the narrow syntactic component does not manipulate phonological features, a parallel

approach (Borer

1 99 1 ,

1997)

is ; therefore more

appropriate than a split one. "[T]he best case is that' there ,is a single cycle only.

[ . . .] and the

[

..

.] components of the d�rivation, : ofi <:PHON , SEM>

proceed cycl ically in parallel" (Chomsky 200 1 b/2004; j>.:,4). · With ( 1 0), ( 1 1 ), the present formalism takes us a considerable way towards th is goal.

3

Notice that structure preservation for merge does not entail the preservation of hierarchical structures under the mapping into syntax.

Concatenation and interpretation

169

2003. Templatic Architecture. Recherches Li n gui s t i q ues de Vi n cennes 32: 157-168. , Hagit. 1991. The Causative-Is ncchoative8: 119-158. L :�ii�lf]�orler., Ha t 1997. ed. Wolfgang Advances John 5-30. ," " 1995. Press. :j��!§l�; ;C£homsky, Noam. Noam. 2001a. Deri v ati o n Ph . Mic a Kenstowicz, 1-52. Noam. 2001b. s., ky, homs i::!i}i,:;,p Ch m ky (2004). 2004. St Beyond. �":::n" ;Chomsky, e Adri a na 104-131. ,:�';¥:\" : ;�:< n v r Pr s. )������j'i'. i m 1993. Gg ss. f�:- C s 24: 2 39-297. ";:���Ji;' t��:;Einbick, avi . 2000. ry 3 185-230. :;:;i��i'��,!i::' �( 2005. and Vergnaud, Jean-Roger. 1987. l�� "'J� Ha 1993. Morphology 1' ;!IW�i' ��:;� � �:r:: ::,e::' � S;::�di��I�O�n�s:n�:IL�:;U���::� ����;/f. .i Cam g Press. ���l � ���i p arsky, Paul . 1966. 7:69-98. I. ";;C', kiparsky, Paul. 982. L ca l Yang, 3-91, Hanshin. In ! 4. J a n 1996. CV h r Laks, 419-441. Puthanvettil. 1986. {:;I;:" "':;Mohanan, Karavannur Lancaster, ! ,,;�ii/�i Mari n a, and 1986. i W �: N . N. J . : Fori s . ! ' I ; a k ntioffon 95:Ray.201-236. 2005. Language: What's ia About It?JCogni �, Sabrina,

gi

,

R. Rennison,

by

U. Dressler, Martin Prinzhom, and

BerlinlNew York: Mouton de Gruyter.

The Minimalist Program. Cambridge, Mass.: M IT by ase In Ken Hale: A Life in Language,

"�' :? homsky, ed. by """" " ,

h el

Cambridge, MA: M IT Press.

Beyond Explanatory Adequacy.

, ",""r "

Beyond Explanatory Adequacy. In

Noam.

';

;

D

d

I:

Theta Theory.

:;lialle, Morris, ;;�����r Mass. : MIT Press. ' £" ne, Morris, and Marantz, Alec.

.

o

r

�;),; ::

g

brid e MA: MIT Ober den

t

"" H

exi

Morning Ca m , ed. e . Phonology, ed. by

H

,

,

It,

i<­ �

"

BerlinlNew York: Mouton d e Gruyter.

An Essay on Stress.

Distributed

and the Pieces of

deutschen Akzent. Studia Grammatica Linguistics in the

l Morphology and Phonology. In

by I. S.

Seoul:

as t e Only Syllable Type.

Current Trends in

Jacques Durand and Bernard

Manchester:

ESRI.

The Theory of Lexical Phonology:

Dordrecht, Boston,

espor,

Riverton,

Pinker, Steven, and Spec l

Cambridge,

° b

;;i�:i'f'{ owenstamm,

.•

Belletti,

Features, Syntax, and Categories in the Latin Perfect.

lt8iden, Martin.

: :� : . :� i

ructures and

ed. by

Lingui tic Inquiry

Linguistic Inqui

"

M IT. Published as

es Oxford: Oxford U i e sity l e l o. A Null Theory of Phrase and Compound Stre u inque,

:

;i,

M

s

o

Th Cartography of Syntactic Structures,

I

I

The Morphology-Syntax Interface: A Study of Autonomy. In

.

';�

£

I

Martin.

The ingui ti Review

in Morphology,

;:

[

�. � i itt

Haiden,

Alternation : A Case Study in

Parallel Morphology.

!

�

and

Tokyo: Kluwer.

Vogel, Irene. c e d

,

Prosodic Phonology.

Dordrecht, Holland;

The Faculty of

1 70 Martin Haiden Vergnaud, Jean-Roger. 2004. On a certain notion of ·occurrence' : The source of metrical structure, and of much more. In Living on the Edge. 28 Essays in

Honour of Jonathan Kaye, ed. by Stefan Ploch, 599-632. BerlinlNew York:

Mouton de Gruyter. Zubizarreta, Maria Luisa. 1 998. Mass.: MIT Press.

Prosody. Focus, and Word Order. Cambridge,

:;��£",, �1.e3

time goes by: . digressive discourse ert Haider, Masyuki Oishi and Shigeo Tonoike

''':' :'I�· ' t3' 1m van Henksdijk: Why, in your opinion, has Henk kindly refrained until . :

from removing all major syntactic problems (relating to linearization) : from the agenda by solving them?

r;�0�:: " Hubert Haider: First, he needs something to keep in reserve for the. time :]IT;;l:: 'while waiting for the olives to get ripe. Second, he is kind enough to avoid :;it���:�,: H embarrassing the grand master of the field by depriving him of the lead. ";�ti;":' rhird, it is wise to leave some work for the young generation or at least to :1��; ; take this as an excuse for spending time on more mundane matters l ike

":�: travelling to opera performances, grooming one's stock of wines in the .. . , and checking its process of maturation regularly by testing one or the J),ther specimen in the company of fellow syntacticians in order to inspire , with break-through lightnings of syntactic inspiration.

.

,;="" ; , .. .... .

Masayuki Oishi: I may be wrong, but he might not like things like, say, :��::;�Hriearization. Recall those well-organized piles of the bottles of wine stored �k=':�:1ih the wine cellar within his house in Tilburgseweg. He is not interested in "�;;;:L:Jinearizing those bottles to form a mere long linear sequence by linking the ;§!:!� !! �thead of a bottle with the bottom of another bottle, like a long chain. In ;-� �JrC' , l;l()ing so, one m ight ' strand,' leaving a good bottle of wine with a label 'R' 'f"sr0" behind somewhere. Henk never strands in his l ife. Furthermore, one of his ,\��7j�" Joudspeakers has malfunctioned, meaning that he does not need any ,��tdii1earized PF outputs until he gets a new pair of loudspeakers. Another ;:;;;�" probability is that he has been too busy in grafting in his garden and woods with Elisabeth as we all know. ·.�,��%�m" .. .-. .

,, ,

;_�f��£!���::;" '

,

���'F ShigeO Tonoike: You must ask Henk for his real reason, but I would think :ml;�?'?:it is because he

is sensible enough to realize that it is almost impossible. It ::,:;' ;: . . is almost impossible because each major problem has several possibilities of solution and each possibility depends on the choice of such possibilities in other major problems. So suppose you have ten such problems and each has 4 different possibilities that must be considered. Then you have 4 to the lOth power possibilities.

1 72 Hubert Haider, Masyuki Oishi and Shigeo Tonoike RvH: Why are there different languages at all and not just a single, UG­ determined one? HH: Imagine a situation in which the whole world would just speak English. Maybe this is not too unrealistic a prognosis for the future, given the ongoing loss of small languages and the consequences thereof in the future. Eventually, there will be j ust one language, namely a kind of heavily creolized English. So, linguistics will reduce to the study of dialectal variation and theories on micro-parametric differences. And our colleagues in the future will be likely to be mislead in thinking that the grammar of this creolized English is identical with UG. Maybe, this will be a deja vu experience for historians of linguistics in the future, looking back into our days . We could risk a bet without much risk that the dialectal variation in the world-wide post-modem English would be significant and that dialect splits would tum the dialects into different languages, unless the world government would be able to guarantee a world wide schooling system with rigorous training in a norm variety. Evidently, ua leaves room for variants of grammars within a defined system space. Why is it not a closed programme that allows only a single way of assembling the rules for a grammar? Presumably because ua is not the program for language processing at all, but a meta-program, namely a program that guarantees processible linguistic expressions given the wide range of brain resources we have for pattern computing. Second, the abstract problem - what structure should a l inguistic expression have in order to be effectively and efficiently processible by the given brain resources? - seems to have more than one solution, . and the alternative solutions are each optimal or sub-optimal, depending on the point of view. At least, there is no particular system that could be singled out as THE superior solution. And finally, people hate to speak the same language as those who they despise. So they push for changes. MO: Why do you ask such a difficult question, Riem? The question seems to have confused me a lot. But to go on toward the end, let me clarify something. Suppose that ua in our mind/brain has not changed since our ancestors appeared somewhere in Africa, as someone says "men were deceivers ever," and that it will not change as one of the defining properties of humans. Then the range of what we can acquire is predetermined in advance. In this sense, the question of whether we see more than one language is a matter of abstraction. If an I-language is just a name given to

..�

As time goes by 1 73

of properties and values of parameters (putting aside a further question what the truth of 'parameters' is) and the lexicon, the differences we ;"'kc!pn,rp. are just superficial not real, and no more than a result of the intc�rac�t1cm between our UG and your or my Umwelt; the more cases of fln:zwe'll the more 'languages.' It's much like asking how many colors you identify in the rainbow. .

. Interesting point, but what was first? UG or the linguistic �environment'? If it was UG, wouldn't it determine the linguistic , <, �Ilvironment? Why do we all walk on feet rather than on hands, irrespective 2���� dfthe walking environments? - But our 'tongues' and their grammars are less uniform as the current model would have it. c '·"

.

find Hubert's remark that UG is not a program for language at all interesting. It is an interesting idea, and it has serious implications to linguistic research currently conceived of and carried out. But the #;�ans�ver to the question of why there are different languages will be the under the more usual conception of UG, namely a system devoted to 'Ian.gu�lge and independent of other modules of mind. The answer will be UG is underspecified and hence allows for a certain range within :: :sE"""'i'!;rutii�h language can vary. The question is a bit like: Why are there different of physical fighting among humans like boxing, wrestling, kung fu, i,§§§:§f : etc.? The answer obviously is that human physique allows some in the way humans hurt or kill each other barehanded. I

:.

r�

J"

n " ""

;if���ff:?�V�:J.. Why - in particular - should languages differ with respect to word ii;��i�� ?"liIJi: This question is a specialization of the previous one, isn't it?

: Do you like us to put a period to this, asking so tough a question, Riem? The differences are just out there. Let's suppose that linearization is required by one of our performance systems, and that the notion of word . " order is a way of pronouncing a syntactic object. Furthermore suppose that any linguistic phenomenon is syntactic in a broad sense, with phonology, narrow syntax, and semantics included. Then we may restate the question .as: asking how we manage to pronounce in certain different ways one and the same feature set generated through Merge, as I have claimed elsewhere. Imay fail, but I am inclined to suppose that ordering is assigned just before we get into phonology; in fact ordering assignment, or directionality

1 74 Hubert Haider, Masyuki Oishi and Shigeo Tonoike

assignment, can be a part of what Transfer is to define. Transfer proceeds as determined by the values of directionality. Along the general l ine of the min imalist program, then, we may assume that Transfer has as its part a simple recursive function which converts a syntactic object of the form ( 1 a), where a., 13 nonnull, to ( l b), in accordance with directionality: =

(1 )

Convert: a. a set {a., 13} b. onto an ordered pair
I3>

This algorithm can only look at an object ( 1 a). Why before entering phonology? The reason is: Otherwise, phonology, which inevitably makes use of structural information in one way or another, cannot be fed; word ordering is just ordering of what can be pronounced, hence ordering of pronunciation; and we can get one and the same LF representation cross­ linguistically. I f the difference in word order is much like something that Hubert has been proposing, i.e. if word order is something that syntax, in fact Merge, has to take care of, then that's another story. I understand you have a different picture of word order. I may reconci le my view to yours, if Merge can be sensitive to parameters responsible to ordering at no extra cost. HH: Word order refers to the properties of the linearization patterns of terminal elements. The linearizations are different because there is more than one possible and practicable solution to the general task of grammar. A given grammar is a particular solution of the general task, namely an effective algorithm for mapping linear sequences on hierarchically structured, box-in-box-type, objects, and vice versa. Grammars are algorithms for a dimension management problem , for our brain. Natural languages use a linear interface (phonetics-phonology) and an interface with a complex box-in-box structure organization (syntaJ:(-semantics interface). Syntax is the algorithm for this domain . man agement. So, basically syntax is dimension mapping by either unfolding a complex structure on a linear array, or compressing a hierarchically organized structure into a linear array. This task allows more than one implementation, each of which has advantages and disadvantages, but there is no single solution with no disadvantages, for principled reasons. Let me give you a concrete example. Assume that in the cognitive evolution of grammars (UG), there is a choice between two grammars for a given set of linearizations, one of

As time goes by 1 75

is m ore ' friendly' to the mental parser. The outcome is easy to ;�iDl·ealct. The friendly grammar wins, the less friendly one is ignored. The ',,�,,)stmg grammars are the result of cognitive mutation and selection, driven the optimization of the grammar & parser fit. What is a parser-friendly grammar? The mental parser has two limited, Jri1ter<1e):lenaelrlt resources, namely processing time and storage size. It starts the first incoming terminal and it simultaneously starts with its topresources (gramm atical knowledge). n dow . Here is the criterion for 'parser friendliness' : A friendly grammar is a that provides the mental parser with structures that are optimal :::==':::: n4'}cn in the top-down and the bottom-up respect. Top-down means early ,�upport by the grammar, bottom-up means early decisions in the string-to­ ;i��,",� :structure mapping process of an utterance. There are just two possible ways to organize a uniform ly binary 'E�J ':bl'all�chlng structure, namely as a right- or as a left-associative structure, see and (2b) respectively: The phrase can be layered by merging with a iihrase that precedes as in (2a) - right associative (namely, the nodes on projection line of the phrase that results from merger follow their sister that is, they are on the right hand side of a phrase merged with a node . the projection line of a phrase) - or with a phrase that follows, as in - left-associative. Is one of these two options more friendly to the parser than the other? Yes, the friendly one is (2a). ntal " ' ;J[4:1fu��fm,e =;;""

,', � '__

:T§�ill .>.��r:� �

[x' YP [x' xo ... ]]] [xp [x' [x' ... xo . ] YP] ZP] [Xp ZP

...

..

unfriendly property of (2b) is the fact that in a complex projection, the node never coincides with the top node, except for the last step. So, is never an immediate convergence of top-down and bottom-up �:V�;;mill ��jkiJnform4ati'cm. But this is the decisive criterion. A continuous supp ly of top­ ·' down information presupposes a structure that allows the parser to 10,;"";:;,:,, ,'; : ' : ' ; 'Jinmediately identify the top node of a projection once it reaches the ;;:" :.,�",",,,· , . ;;:+, :;�oundary of this very projection. This is the case in (2a). . The superiority of (2a) rests on the fact that (2a) has itself a top-down . • structure. But it has a disadvantage� too. The head comes late. The early presentation of the head of a phrase to the parser is also an act of " . friendliness because it provides information for predictions which element to expect in the rest of the phrase. This aspect would per se favor head­ initial projections over head-final ones. Unfortunately, a right-associative

1 76 Hubert Haider, Masyuki Oishi and Shigeo Tonoike

projection (2a), with the head in the foot position, is necessarily a head final projection with the head at the right edge. So, if the head is to precede, it has to be higher in the structure. This higher position is not the foot position. So this structure is not admitted as a �base' structure. But it is admissible as a 'derived' structure. This is the source of complex head-initial structures, usually referred to as vo­ structures. The shell structure of complex VPs is an immediate result of combining the three factors that determine a head initial complex projection. VO is the result of implementing two apparently conflicting demands, namely the necessity for the verb to function a the head in the foot position of the projection, and the parametric option of the head, namely, progressive licensing (i.e. licensing to the right, rather than to the left as in head-final projections). If it is to license the higher positions in the projection, it cannot remain in the foot-position but will have to move to a position close to the top of the structure. So, a head cannot be both to the left and in a foot position of a complex projection. The foot position is the lowest position, the phrase initial position, however, is a high position in the phrase. In sum, the head-initial and the head-final organization is optimal to a certain extent in strictly right-branching projections and suboptimal to another, and there is no way to have the cake and eat it: If the head should come first, the price to be paid is the complication of structure by VP shells. The price for leaving the head in its foot position, on the other hand, is the disadvantage of having the head as the last element in the phrase. Given this set of circumstances, the d,ifferent word order systems among languages reflect the different syntactic solutions of a fundamental problem for grammars, namely to provide structures that can be parsed effectively, efficiently with an immediate contact of top-down and bottom-up information. MO: You're claiming that universal directionality of merging is to the left, and that once we have merged something to the right, we will inevitably make up a shell? HH: No. There is - hopefully - no merge to the right at all. Shells are necessary if the canonical directionality of licensing is opposite to the directionality of merge, namely as licensing to the right.

As time goes by 177 : So, Merge is sensitive to where to put your dependent� in your system. we put it this way? Merge is always sensitive to the label of a head in cases whereas, in VO cases, the operation is only 'allowed ' or 'have' to at the outmost boundary, or edge, of a set that you have already bui lt or the set itself - an additional burden for VO in a sense. Then this be one way to look at a parameter that we are l ooking for: The in what Merge is sensitive to. Or Transfer can be another which such directionalities as you assume are processed. It by a'[e (1110 ctiln what syntax can do. I ncidentally, I just wonder why you can 'dCIPeJIOS on " .. ;oftiu:.,rlrp. an external argument to the left even if this particular merging is in 'violation of directionality requirement. ,ST: Wh ile there is no doubt that linearity is imposed by PF,

I am not yet

convinced about the basic assumption that seems to underlie Hubert' s

':'��i;; theOl� of word order. It seems that Hubert's theory assumes that there is a

that exists independently of language or UO. That is certainly a ,' . D()SSllbllllty , but we don't know yet that it is in fact true. An equally ible possibility is that a parser is tailored to a particular form of UG . for example, that contrary to Hubert's theory, UG al lows for :;§§�/;' . lai1igUlilge to be either right-branching or left-branching, as I happen to i�i;benf:ve. In other words, UG has a parameter of directionality. Then when =,C;'2Eiiijithe parameter is set, UG can choose/create a UG-friendly parser. One can of UP (universal parser) with a parameter that has to be set according way the directionality parameter is set for UG. the '.§��·� .;Io: There are two things that I am not very comfortable with Hubert's of word order. One is that OV languages and YO languages have radically d ifferent derivation and structure: VO languages have lots of while OV languages have complex verb clusters. Would it not be er to say that the two types of l anguages have exactly the same \;;j;iTm: :hi��raJrchical structure but differ in the d irectionality of linearization and other respects such as the possibility of verb raising etc.? The other thing ' :has to do with how to arrive at the same or similar LF representations �etween two expressions of the same meaning across the OV-VO �i:!!Ij§�H[( distinction. Take John seems to love Mary and its Japanese counterpart frlary-o John-ga aisiteiru rasii. (The object-subject word order will be . ie.J.'plained shortly.) Since the two expressions mean basically the same thing they should have more or less (or exactly) the same LF representations. It is reasonable to assume that the LF representation of the English example is something like [seem [John (to) love Mary]], which can be easily derivable as an LF representation. B ut the LF representation of the Japanese

1 78 Hubert Haider, Masyuki Oishi and Shigeo Tonoike

counterpart requires extra LF operation(s) of breaking the verb cluster to give a comparable representation like [[Mary aisiteiru John] rasii], if the two languages are exactly the same except for directionality and verb raising. Then the LF representation of the Japanese counterpart comes at (almost) no cost, once we recognize that linearity plays no role at LF and make certain assumptions about the effect of raising operations like NP raising and verb raising. The problem becomes even more serious if we fol low people like Kayne in trying to eliminate LF (Le. covert) operations. RvH: Why is syntactic structuring asymmetric? - In other words: why do we not find systematic mirror image relations between expressions in one language and expressions in the other language, as in: Fragenask wirwe Henk! (= let us ask Henk); Henk-n ida' wareware-gawe kikooask'

HH: The asymmetry factor in language is its production and perception.

Linguistic expressions are produced in time and perceived in time. The linear order of elements reflect their utterance timing. It is this external asymmetry that indirectly produces the internal asymmetry. The symmetric counterparts of right-branching phrases are left-branch ing ones (merger to the right). And these structure are parser-unfriendly (see above). So, grammars viewed as symbol processing algorithms could be organized in either way (as for instance Kayne's LeA-implementation allows), but the grammars with left-branching structures would not survive the cognitive competition. ST: This brings us back to the issue of whether there is a parser independently of UG. Parser-friendl iness or unfriendliness makes sense only when we assume that a parser exists independently of UG. It is an interesting assumption, but not yet substantiated. In my recent paper (read at the l 3 0th Meeting of the Japan Linguistic Society but not yet published in a written form), I claimed Japanese has a left-branching structure, which is a mirror image of the structure of YO languages like English. The argument is based on the following wel l-known fact about scope phenomena in Japanese (as well as certain general assumptions about grammar the most important of which, in this connection, is that there is no covert operation and that elements take scope at their surface positions rather than at their trace positions).

As time goes by 1 79

a. Dare- mo-Oeverybody-acc dareka-gasomebody-nOm aisiteirulove (eve�>some, some>eve�) b. Dareka-gasomebody-nom dare-mo-oeverybody-acc aisiteiru love (*eve�>some, some>eve�) two sentences are basically the same as their English counterpart fiimreblJety loves everybody except that (3b) is not scopally ambiguous and is. The standard treatment of this phenomenon is to say that (3b) is the word order and hence is unambiguous, while (3a) is derived from (3b) "8lE5Sig'iU¥ scrambling and hence ambiguous due to the existence of the trace. I-In'U/P'vpr, such a treatment runs into a number of difficulties, the most ... ",.-. nlI C of which is of explaining why (3a) is not amb i guous while its ;�i�§,t I.tillgllSn counterpart is: if there is a covert operation like QR, then it should " 'T:�:'� /:�"" ;Qh lp to apply to (3a), making it ambiguous. Another problem is the use trace in accounting for scope interaction. If the inherent function of §t;st1:ramb linglQR is to give wider scope the element to which they apply, then . . .. ,.,'<•... " " .. traces to account for narrow scope goes against Economy Principles. If we assume that (3a) is the basic word order with the (simplified) ;ttfi'ctulre [TP [ VP [v' dare-mo-o ei] dareka-ga] aisitej iru] where verb raising ;���;;S�?]:m!:: : taken place but not NP raising of the subject, accounting for the ' ��A�"Pl'"v>��onle" reading because the subject asymmetrically c-commands the i.;O.lJflec:t. The same string can have the structure in which the sentence-initial dare-mo-o is raised and adjoined to VP, accounting for the if§ii,:::::;onlle:e:" v'elJr- reading because the object now asymmetrically c-commands subject. Under this conception, (3b) is derivable only by scrambling the O:i lj�i.'.�u'bjelct across the object, adjoining it to VP, [TP [vp dareka-gaj [vP [v' darern o-o ej ejJ] aisitejiru]. If this line of argumentation is correct, then at least : ' :�f��r;:;J ;]apanese is a good candidate of left-branch ing language. ,, But why is the relative order of adverbials in preverbal positions a �! �ig�Jmirror image of the postverbal order? - Is this a symmetry effect?

��

>

;:!�!iili;g� ,?i��m': ;';�VH:

';:l:{�E�mIt:(�: aoj <

�

:'

:::;ax ' 'i�:�.m" . ;-

�

It would be a symmetry effect if its source is the respective side of unction. Adjunction in the same relative order to the right or the left of a phrase (let us say VP) would produce a mirror image order. So, does adverb adjunction defy the restriction against merger to the right? The answer is clearly not a clear yes. Postverbal adverbial phrases do not behave as if they were adjoined if we consider binding relations. As many syntacticians realized, the evidence is puzzling since it reflects conflicting structural demands. Topicalizability of the VP seems to call for adjunction of

1 80 Hubert Haider. Masyuki Oishi and Shigeo Tonoike adjuncts to the right of the VP, but binding requires the PPs to be embedded on the right edge of the VP in order to arrive at the adequate c-command relations between an object as a binder and an element in a postverbal adverbial PP (see Pesetsky's proposal of a dual structuring with layered and cascading structures). So, there seems to be no standard solution to this problem . The solution seems to require 'backtracking' in our theoretical assumptions. If binding evidence is in contlict with movement evidence, and there is no single structure in which this contlict can be resolved, we have to rethink either our concepts of binding or our concepts of movement or both. At least, the alleged case of a mirror symmetry is a murky case. RvH : Why is the order of suffixes that code for adverbial relations (in a given language) the mirror image of the order of the correspondin g adverbials in preverbal positions (in another given language)?

HH: First, the source of the relative ordering among adverbials is a mat er

t of the different semantic functions. Adverbials that specify a process (manner) are modifying a more embedded semantic domain than adverbials that modify phrases that denote eventualities (e.g. a VP), and the domain of propositional attitude adverbials is an even h igher one. For adjuncts, the domain inclusion properties of semantics yield the relative order. For suffixes, the fact that the word structure is right-branching is the reason for the mirror symmetry with respect to the phrasal order. Take the following word structure. Each of the affixes is a separate bound word element that selects a word element .as its left-hand complement. Their selection domains are the same kind of nested domains as for the phrasal adverbials in terms of their semantic properties.

The mirror image symmetry is j ust a retlex of implem€mting a given hierarchy e ither by means of adj unction to the left (adverb is the adj unct) or by affixal selection in a word structure with suffixing. In th is case, the modifier relation is coded as a relation between the suffix and the word structure it is suffixed to.

MO: Then, are you assuming that the ordering between a verb and its

argument(s) is regulated in a different fashion from what determines the ordering among adjuncts? We have no symmetric ordering in the syntax of

As time goes by

18I

whereas we do get symmetry in the case of adj uncts and 1i ':;�§�L $.ujttj�:atJ,on. The difference results from the way how these two types of �i?:'Obiects are composed, aligned and interpreted. Right? If ARGUMENTS would be coded by suffixation, I would expect a mirror too, but for a different reason since an argumental affix could not the verb in the way a modifier could. The argument-related affixes only be agreement affixes. The relation to the DPs they agree with ld be a nested-dependency, and hence the mirror image order.

In this connection it is interesting to note that Ernst gives a convincing

,�i�ji" :;� ; ��':�� that contrary to the claims made by people like Kayne, elements

the end of the sentence can have wide scope over elements to their left. instance, Henk didn 'f answer my e-mail again is ambiguous between 'i::�t�� j':tl1le "not>again" reading and the "again>not" reading. So while it is no correct that adjuncts are licensed by different elements, they do at the end of a sentence with wide scope over elements to their left, U l!fl�estl' l119 that rigid right-branching structure cannot be correct for Why does precedence coincide with c-command for binding or Is it because binding is a precedence-based relation or is it because FGO,mn[}aI1ld entails precedence because the structures are asymmetrically The latter is what I would prefer to be true. Precedence is a PF-based c-command is the structure based notion. In a perfect fit, c­ ���E�?;;::&'()Rlmand and precedence are just two ways of describing the same f:���i;�:, situation. t:�,;�\ In fact, there are cases of scope inversion, that is, cases in which the rface precedence/c-command relation does not correspond to the j��jIL :�§p A ,���;;�';� :preferred scope relation. But, in this case, there is a PF-signal for ; Um1�f;L: reconstruction, and crucially it is a signal for reconstruction (in an :, tg1:" . antecedent gap configuration) and not just a signal for scope inversion. So, semantics accepts PF cues for mapping syntactic structures but PF does not directly communicate with the semantic construction. Here is an example: . i

1 82 Hubert Haider, Masyuki Oishi and Shigeo Tonoike

( 5)

Fast/aIle GLOW-Mitgliederj hat Henk mindestens almost all GLOW-members has Henk at-least zweimal\ej fotografiert. twice photographed

With a rise fall intonation (indicated by 1 \), the universal quantifier is interpreted in the scope of the frequency quantifier. Without the Topic­ focus intonation, the relative scope corresponds to the surface order. Note that in (6), a rise-fall intonation does not trigger scope inversion. The reason is this: The trace of the first phrase is not in the c-command domain of the second quantifier. ...

(6)

Fast aile GLOW-Mitgliederj haben ej Henk mindestens Henk at-least almost all GLOW-members have zweimal fotografiert. twice photographed

So, PF has no direct contact with LF. However, it may be used for 'flagging' the syntactic structure. ST: Again I would like to point out the possibility that c-command determ ines linear order, but linear order can be precedence or anti­ precedence, which is the prediction of Kayne's LeA, if you read it carefully. HH: In Kayne's system you would have to switch in the middle of a structure in order to get the result. First, you never switch with arguments (always merged to the left). Second, adverbials can appear in between arguments in OV but not in VO. RvH: What role does PF play for the linearization of terminals in syntax propcr? HH: PF is the linear interface. Its linearity dimension requires that grammar function as a dimension converting algorithm. It allows us to unfold hierarchical structures on linear representations, and on the other hand grammar allows us to compress hierarchical representations (needed for satisfying the demands for the semantics interface). Second, certain working memory saving operations might take place at PF . Heavy NP-shift, for instance, seems to be a case of delayed PF-

As time goes by 1 83

,,, r(;al i.zatlon. Si milarly, the positioning of parenthesis seems to be a PF­ ohen()mc;:ncm. A parenthesis 'starts' as juxtaposed comment to a fol lowing preceding utterance. If it appears in sentence internal positions� it is not SVlrltacW'�al l) integrated. Its linear position is the result of delayed or nn�m,ltUJ:e phonetic realization. It seems that my tongue has been really stunned and paralyzed by a i;���m�tinlgra.y. Phonetics is incapable of moving its parts to realize that 'hard' :O.t�lec;t, as I am now� which a clever system called syntax generates, unless (pseudo-syntax, in a sense) is kind enough to mitigate the 1'Iru-dness of the object. It is one th ing to say that Heavy NP Shift and Extraposition for that have the function of alleviating the burden on working memory, I think is a fair evaluation of their function, but quite another to say ;;!§§!;!:'tbat they are PF phenomena. [ think it is wise to avoid using PF operations :0':"' :.,, ,, 10 ,.,0 it is phonetically motivated. Otherwise, there would be no way to ,+ri Ic�tnl such a claim . : It is an empirical question whether a heavy-NP shifted phrase behaves a phrase shifted by grammar or by a system on the other side of the It could be a phenomenon of production proper, and not a t�JU'am.maltlc:al detachment like Extraposition : A sentence can be shifted to . " ."'''';;..;: '''''' ;. right in extraposing languages, no matter how short or long it is. A DP �shitte:d to the right is ungrammatical unless it is consum ing very much ' vv()rkmg memory capacities. : .RvH: Would you agree that there is no OV language with an obligatory '���l:j;:; eJ{J)I�eth,e for the subject position?

§§�i;/l'I1?:1: Definitely! No need for an expletive, if the head of the VP is final.

. Expletives are an VO phenomenon: the VP internal subj ect precedes the verb and thus requires an external licensing head with canonical directional ity. Our present theories are still tailored more tightly to VO . languages than are justified empirically. The issue of expletives seems to be a peripheral one (although the MP devotes much attention to the .exceptional English subject expletive constructions) but it is a core issue since expletives are indicators of purely syntactically motivated positions in the structure. If true, the general absence of obligatory subject expletives in

1 84 Hubert Haider, Masyuki Oishi and Shigeo Tonoike

OV is the tip of the iceberg of a lot more systematic differences between VO and OV . MO: In your story, Hubert, an expletive is one of the standard issues, or gadgets, that YO, or non-OY, languages must devise, not something that they get 'on demand,' as you suggest. Incidentally, there is so strange, unlike it. ST: Japanese bears out Hubert's prediction about expletives in OV languages. Japanese lacks both it-type expletives and there-type expletives. RvH: Do you think that linguistics can reach significant insights into some aspects of the 'true' nature of UG in a human's lifetime? HH: This is primarily a question of the interpretation of ' significance.' I think a successful development depends not only on scientific creativity but even more on a solid empirical basis. Without Tycho Brahe's detailed and patiently assembled data, Kepler could not have sorted out his theory from the pool of 'descriptively adequate' modellings. Our field suffers from too much theorizing and too little patient, systematic and methodologically thorough analysis of a wider range of languages. In other words, too many little Chomskys, but too few little or big Brahes. I think Henk was the first to take action (SynCom) to stimulate again a more evidence-based approach to sorting out the hidden UG-path through the maze of grammars and the even more inflated maze of theory construction. MO: The question may be one of the things which Chomsky calls mysteries. Some say Torricelli didn't do as many experiments as he has been rumoured to. This sounds easy on the ear since I am so lazy a man that I always welcome what comes easily. But I must admit that it does not entail at all that we don't need any piece of evidence, and remember what "easy come easy go" tells us, to get and put on such a new thinking cap. HH: Right! Einstein did no experiments except thought experiments, it is said. But he could build on a huge corpus of experimentally well-checked data and a differentiated body of theoretic modellings. Our Einsteins cannot yet stand on the shoulders of giants. They can only prepare their own shoulders, it seems.

As time goes by 1 85 ST:

At this stage we say we are doing empirical science, but in fact the discipline is still at its infancy and we cannot do experiments to determine ..: :whet.ner a given linguistic hypothesis is correct or not. I certainly hope to long enough to see a day when we have some indication of whether U.·.h �rt is correct about the non-existence of left-branching language or I correct that Japanese is such a language from neurological or genetic ::;)"LlIU.'·�'oJ. It doesn't matter who is correct so long as we know who is correct. .

' . ':

·:· Chomsky. Noam. 1 995 . The Minimalist Program. Cambridge, MA: MIT Press.

: �§§T; ChomsKY. Noam. 2005. On Phases. MS. MIT.

:��j:i3ti;alaler, Hubert. 2004. Pre-and Postverbal Adverbials in va and OV. Lingua 1 1 4 (6): 779-807 . ........ ........ , Hubert. 2005. How to Turn German into Icelandic and Derive the VO-OV Contrasts. The Journal 0/ Comparative Germanic L inguistics 8: 1 -53. , Hubert and Inger Rosengren. 2003. Scrambling - Non-triggered Chain Formation in OV languages. Journal a/Germanic Linguistics 1 5 .3 : 203-267. '�" II"'. Richard S. 1 994. The Antisymmetry of Syntax. Cambridge, MA: MIT Press.

§fsiZ':: l<.ilku<:hl·, Akira, Masayuki Oishi, and Noriaki Yusa. 1 994. Scrambling and Relativized L-relatedness. MIT Working Papers in Linguistics 24: FDrmal

Approaches tD Japanese Linguistics I, pp. 1 4 1 - 1 58. Masayuki. 2003. When Linearity Meets Bare Phrase Structure.

::���n;()istli,

Current Issues in English L inguistics. Special Publications ,Of the English Linguistics

Society ,Of Japan 2, 1 8-4 1 . Riemsdijk, Henk van. 1 978. A Case Study in Syntactic Markedness: The B inding �>���f';" Nature of Prep,Os itionaI Phrases. Foris.

�;ii��;j."':R.iennsdlijk. Henk van. 2004. Graft is the Logically Missing Case of Merge. MS. TD appear in Volume 7:2, Visnyk of the Kiev NatiDnal Linguistic University. }T()no:ike, Shigeo. 1 996. Japanese as an OVS language. In Minimalism and

Hitsuzi

Linguistic Theory, S. Haraguchi and M. Funaki (eds.). 1 05- 1 33 . Shobo, TokYD. .}§§§:: : 'T.OJli1d(jaKle�. ShigeD. 2003 . Overt QR. Thought Currents in English Literature 76:7396. Aoyama Gakuin University. ".; . onoike, ShigeD. 2005. ScramblinglQR, Quantifier Scope and the Clausal ......,. . Structures 0/ Japanese and English. MS. Paper read at the 1 30th Meeting ,Of the Japan Linguistic SDciety.

,!

There's that: unifying existential and list readings Jutta M. Hartmann 1.

Introduction )

Syntactic research dealing with the there-construction in English mostly puts aside the list reading exemplified in ( 1 ) as a separate reading not relevant for the questions investigated. ( 1 ) A: Did we call everyone? B: No, there is still John and Mary.

Yet, there are some notable exceptions as e.g. Belletti ( 1 988) and Safir ( 1 985). I will argue here (along with Safir 1 985) that if we look at the there-construction as a copula construction, it is possible to unify the existential and list reading. However, as will become clear in the final section, both Belletti' s ( 1 988) and Safir's ( 1 985) analyses are not entirely adequate, because not every definite DP that occurs in the there­ construction gives rise to the list reading. In order to understand the proposal, we have to look into the research concerning copula sentences, which has established that we have to distinguish (at least) two types of predication occurring with the copula. Consider the examples in (2). (2) a. Peter is ill. b. Mary' s husband is Peter. The first type is the 'real', predicative type as seen in (2a). Here, the predicate ' ill' is applied to 'John' , thus, expressing that 'John' belongs to the set of individuals who are ill. I will refer to this type of reading as predicational. The second type in (2b) is different. Instead of predicating 'Peter' over 'Mary's husband', it rather specifies 'Peter' as the one who is 'Mary's husband' . 1 will refer to this type of predication as spectficationai. Putting aside the discussions about the correct analysis of these two types I

This research made possible bywhich is hereby gratefully acknowledged. (NWO),is grant

Research

276-70-00 I ,

the Netherlands Organisation of Scientific

Unifying existential and list readings 1 87

copula constructions, 2 I will argue here, that we find the same two types the there-construction: the existential reading is an instance of a copula sentence and the list reading is on a par with the Dec:itl(�atl:on.al reading . Types of Predication in Copula Sentences

literature on classifying the types of predication in copula �onsmuctIO[lS is rich both in content as well as in terminology, resulting in a nm,ewnaI confusing picture of what is relevant for a classification. Without , g to review the literature and terms used, I will partly follow Huber 21illE: ' ttvm " ��,;: '}";'nn'�' who suggests categories of copula structures by cross-classifying g';�§':'lJktc vs. individuals as subjects vs. predicates. For this paper and possibly in 1iell1e�al, the relevant semantic distinction seems to be the element in 'Un�dl,catlve position. Thus, I will stick to a two-fold distinction between ���!l;at I call predicational and specificational copula clauses. I n this way, we at the classification of examples in (3) to (6) as given in table 1 (the terms circling the literature added): Peter is ill Whales are mammals. The evening star is the morning star. What I don't like about Peter is his tie.

main discussions in th e literature centre around the question of whether the §jjf:�lmt)igllity arises from a lexical amb iguity of the copula 'be' or from a s tructural '(1ltterence. On the side of the structural ambigu ity camp, an influential analysis has the one where the specificational predication is seen as an inverted structure predicat ion al (cf. among others Moro 1 997). More recent analyses have r�ued against this view. (cf. among others Heycock and Kroch 1998, Rothstein Pi;;: :� assigning 1) the ambiguities to a difference in the small clause structure that the copul a embeds. (cf. especially Adger and Ramchand for a proposal along this line). .

. '

Another line of research is additionally occupied with the question of whether two types are enough or whether we need to have an additional structure for identity clauses like The morning star is the evening star. ,

1 88 Julia M Hartmann

Table 1: Classification of Predication (along the lines of Huber 2000)

Individual Set Individual Set Subject Position

cnalational, Set predicf. (2),cati(3)onal predicidatientive,fipredi cati o Generic Set Individual specicf. fi(4),cational identity, equative specificational, equative Individual Predicate Position

Terms used here

Terms in the literature

(5)

The main argument for keeping the two classes of specification vs. predication apart comes from coordination 'data: It is possible to coordinate two sets or two individuals in predicate position, but not individuals with sets (cf. Huber 2002):

(7) a. Whales are mammals and h ighly intelligent. b.

Venus is the morning star and the evening star.

c. *The woman in the comer is a fool and Mary. What complicates the issue is that whether a predicate is a set or an individual in the semantic sense does not cut along syntactic categories: DPs and CPs (mainly wh-clefts) can be both individuals and sets (of properties) creating ambiguities between two readings that sometimes cannot be kept apart easily. Generally, DPs can be proper names, definite DPs or indefinite DPS.3 Proper names (as e.g. 'John' ) are in most cases referential expressions, thus fall under the class of individuals. In contrast to these, indefinite DPs (e.g. ' a fool ') express a set

in the default case, but

not in all cases: they can also have a specific reading in which they are interpreted as referring to individuals. Definite DPs have a strong tendency to refer to individuals, but still� they can also be interpr.eted

as

denoting a

set of properties: thus� 'th� teacher' can . be used ',to" refer : to a specific person, but it can also express a (singleton)· set of the' property of being a

teacher.4 Thus, there is a mismatch between the syntactic category and the

ofasthetheytypemiinght be or DPs and their interpretareatioaln saso possi bcatlee. s putisBare tnothemdinouns asi d e, predi smyputed.approach differs significantly from Safir (1985), who assumes that Here, definite cannot be predicates in general (Safir 1985: 118).

semantic distinction between sets (properties) and individuals. One result of 3

Mary is president of the committee

NPs

4

DPs

I

UnifYing existential and list readings 1 89

this mismatch is that especially when we have a structure of the form DP be DP we deal either with a predicational or a specificational structure. Consider (8): Mary is the chief editor of the TIME magazine. can mean that there is a person, 'Mary', and another person, 'the chief of the TIME magazine', and what we state is that they refer to the same individual. In this sense, we are dealing with a specificational reading the copula construction in my classification (or in other terms, the . �¥:2::j;:: .Jdentity, equative reading). Additionally, there is a different reading 'j��::':" available as well: ' Mary' has the property of being the chief editor of the :l;�sl" TIME magazine, a property that is a singleton set. If we invert the clause, ;:��::':: the second, predicational reading is not (readily) available and it is hard to relevant i:����" imagine a reading in which 'Mary' is a set of properties that ere:5 ��J:\E � The chief editor of the TIME magazine is Mary.

:;�:;;;;'�; ;: :kA;:+nr

are

�

,;I 'C�)

j�� i:�"putting aside these important issues, the main point that I want to make is ;t::-::iitlitiat it matters whether the predicate position is filled with phrase ' Fekpressing a set or an individual (skipping over the possibly complicating �iH l;: '" fa.ct of the nature of the subject of predication). Taking this step, we expect :f:,:S�:::;:�� . 'pula constructions to cluster in two pairs: predicational and U,gg;,: '�Pecificational sentences. On the assumption that the verb be in there­ 'l;=��;!:sentences is the copula (and not an auxiliary),6 the prediction is that we :.:L!f' ,,' should find both readings here as welL The main claim of this paper is that �1�*f::iWhen the post-copular position is filled with a DP denoting a set (the �!��ig:�,p:tedicational reading), the existential meaning arises. Whenever the post­ ,jm�nL;#)pular position is fiIIed with a DP referring to an individual (the ::,;;�'�" V�pecificational reading), we end up with the list reading.?

a

..

·:IIF' t o e g . e �;: ��� ��:��:�!��i�� !�: ���e ��fe�e�7:! �:i:�; ��";:� ;;: ::: ;(,!I.:;'.i,:'.,·· ::r:eo�;:J� magazine has the property of being Hamlet', a paraphrase which suggests .,.....----

..

..:; ';' , , :rIME

;�t�W, ::�' �; '''- ,

•

!be 'role' reading that the sentence has, cf. Doron ( 1 988:295).

There in English also occurs with unaccusative verbs restricted contexts. I eXclude these here as they seem to be different from the existential construction important respects, cf. Hartmann (2005). 'lh doing this, I want to remain neutral about the question what difference it makes that there occupies the subject position in contrast to general copula constructions in

in

190

Jutta M Hartmann

3. The existential reading: an instance of a predicational structure

The main arguments to treat the existential as a predicational structure come from parallelism arguments by McNally (1 992) as summarised by Zamparelli (2000). First of all, both DPs in predicational structures as well as existential constructions cannot take wide scope: ( 1 0) a. There weren't two people drunk. b. John and Mary aren't two students I know.

Neg > 2, *2>Neg Neg > 2, *2>Neg

Secondly, McNally shows that in both constructions, strong quantifiers can only occur if they range over kinds: ( 1 1 ) a. There was every kind of wine available for tasting. ?? b . . There was every worker ready. ( 1 2) a. John has been every kind of doctor. h. * John has been every doctor. Thirdly, for both types of predicates, it is not possible (for most speakers) to relativise a predicate nominal with a wh-relative pronoun. s ( 1 3 ) a. The people * who / that / 0 there were at the party were drunk. b. They dressed like the eccentric women * who / that / 0 they were. Furthermore, Zamparelli (2000) argues that we can grasp the similar semantics if we understand the meaning of the existential as the post­ copular property being asserted over a location meaning "that that space 'is' or 'contains' [e.g.] a man. The only case in which this assertion is ever going to be true is when a man is in existence." (Zamparelli 2000: 1 88). On the syntactic side, Williams ( 1 994) argues that the DP patterns with predicates with respect to wh-extraction. The contrast of ( 1 4) versus ( 1 5)

in which the subject is another DP. Another possibility that I do not want exclude a priori is that we might have to distinguish between the predicative heads that lead to general copula constructions and the predicative heads that derive existentials. There is only one problem with the postcopular DP in the there-construction: contrary to garden variety predicate nominals, pronouns can refer back to it: There is a god and he sometimes says no. (cf. Doron 1988 for this test). For a possible solution, see McNally (1992:62f). to

8

Unifying existential and list readings 1 9 I

shows that predicates tend to be less extractable from wh-islands than ,arguments (the examples are adjusted from Williams ( 1 994» : ? . 4) a. ?What do you wonder who fixed? b. Who do you wonder why Bill likes?

, 5) a. *How tall do you wonder who became? b. *How foolish do you wonder why Bill considers anyone t? postverbal-DP of the there-construction patterns with predicative DPs �s seen in (16). Thus, Williams argues, it must be a predicate as well. "

6) a. * Who do you wonder why there was at the party? b. *How many people do you wonder why there were?

.L��,;·j�k The list reading: an instance of specificational predication far, we have seen that the existential reading clusters with predicational P Q C UJla clauses. The next step to take is to show that list readings pattern the specificational predication. There are two arguments in favor of move. First of all, Higgins ( 1 973 :2139) pointed out with respect to constructions that specification opens a list of which the :n; !:;�:l�qmsltcolDul,ar DP is a (unique) member (see also Safir 1 985). Thus, in ( 1 7), specificational pseudocleft opens the list of items ('those things that I ,:::"':"""�V, not like about John') and then, we state that the list contains the item in ,' n(JlsI-�[;o[mlar position, 'John 's tie' . What I don 't like about John is his tie. , 'o"'.":.,' .:."' :'U

similar reading is induced by specificational copula constructions. Here n the precopular DP opens the l ist, here ' Angie' s friends' , and then, we the items for this list, the relevant individuals.

'l�t.j.,���� ti ::�

Angie's friends are Peter, Paul and Catherine (among others). there-sentences matters are slightly different: the preverbal constituent is there and as such is not able to open a list. However, as soon ' as the list is sal ient in the context, there can pick up this list and the postcopular DP fills it in. 9

Page numbers refer to the 1 979 Garland publication of Higgins' dissertation.

192

Jutta M Hartmann

( 1 9) A: B:

Whom shall we invite to our party? Well, there's John, Mary and David.

Secondly, it is not possible to coordinate the two types of predications in the there-sentences, just as much as we have seen it for copula structures in general. (20) a. #Oabi's husband is Peter but not a lucky person. b. #There's a god and 10hn. 10 5. On definite DPs in there-constructions

So far, we have seen that it makes sense to analyse the there-construction in the frame of copula clauses. Now the question is whether my way of looking at there-sentences improves on approaches already present. The testing ground are definite DPs occurring with the construction (in limited circumstances). On my account, we would expect that definites occur in the construction under two circumstances: either these definites are interpreted as sets giving the existential reading, or they are interpreted as individuals in which case they need a salient list in the context. On Safir's ( 1 985) and also on Belletti's ( 1 988) analysis definite OPs should give rise to a list reading. In essence, Safir argues that whenever a definite DP occurs in the structure, it needs case and the specificational copula (but not the predicational copula) can assign case. Therefore, the two have to co-occur, predicting that every definite OP has to instantiate the list reading. Belletti ( 1 988) even more closely links definites with the list reading. She argues that a DP that receives partitive case is interpreted as an item of the Ii St. 11 However, none of these predictions are true. In particular, with respect to my account, not every definite OP that can occur as a predicate can occur in the there-construction, as we see in (2 1 ). 1 2

Example (20b) is not as awkward as example (20a). This effect is due to the availability of the list reading with 'a. (specific) god' and John' on the list. For both approaches this is essential as they otherwise lose their account for the definiteness restriction. Due to the limited space, I cannot further elaborate on their approaches, and the reader is referred to the originals for the in depth analyses. Sentence (21 b) of course is acceptable if we have a deictic there or we read the sentence in a list context. However, under the existential reading this sentence is not possible. 10 II

12

•

if

Unifying existential and list readings 1 93

'

(21 } a. John is the president of the comm ittee. ' .' b. "There is the president of the committee.

in my analysis, I still have to take into account some formulation of definiteness restriction for the existential reading (possibly along the Ifnes of McNal ly 1 992 ). However, contra Safir's ( 1 985) and Belletti ' s 988 , ) prediction, there are also definites occurring with there that cannot be subsumed under the list reading. In order to show this, I will :idi�;cu:)s the data given in the seminal work by Birner & Ward (1998). First of all, I think we can exclude the so-called false definites as seen ' In' the examples in (22) and (23). In these examples, a definite occurs, but 's"","",;;:::w ' Lt: definite article can be replaced by an indefinite article without a change , the truth conditions of the sentences, thus, we are not dealing with real :'��;Efge1tlmtes and I put these examples aside. .

What can happen is a hangup such as Rocky Smith ran into, as the independent hauler was traversing Chicago with a load of machinery that just had to get to a factory by morning. "There was this truck in front ofme carrying giant steel coils, andpotholes all over the place," he remembers, "This guy swerves all of a sudden to avoid a big hole." He hit it anyway. One day last year on a cold, clear, crisp afternoon, there was this huge sheet of ice in the street. second class of definites that occur with there are the ones discussed 'list readings' . Thus in (24), the salient list is the list of the two in Division H; in (25), it is the list of possible birthday

'� l r.�a{1v·

And there's two components in [Division H], which is the operations division: the people that do the flight activity planning procedures work, [ . J And then, there is the payload supportfolk, who provide for customer operations integration and support [ . . . ] .

.

What could I give my sister for her birthday There's John' s book on birdwatching.

19.

contrast to Birner and Ward (1 998), however, I think, more of their 'examples can be grouped under the list reading context. Thus, in (26) and

1 94 Jutta M Hartmann

(27), the beginn ing of the sentence with ' in addition' suggests, that a I ist is salient in the context: (26)

In addition to interest-rate risk, there is the added risk that when interest rates fall, mortgages will be prepaid, thereby reducing the Portfolio's future income stream.

(27)

In addition, as the review continues, there is always the chance that we'll uncover something additional that is significant.

Furthermore, to my eye, sentences that Simer and Ward group under the reminder context, can be treated as list reading as well. The salient list is exactly the list of things to remember:

(28)

Caspian's hand had gone to his sword hilt, when Lucy said, "And you've almost promised Ramandu's daughter to go back." Caspian paused. "Well, yes. There 's that," he said. He stood irresolute for a moment and then shouted out to the ship in general .

(29) A: I cannot imagine what I'm going to make for dinner tonight. B : Well, there's that leftover meatloaf. Nevertheless, it is not possible to subsume all definites occurring in the there-construction under the list reading. First of all, there is the group of sentences in which the existence of the definite DP is asserted. C learly, these definites do not fall under the specificational readin but under the existential (predicational) reading of the -there-construction. l

�

(30)

But there is not any evidence, but there was this 10 thousandths.

(3 1 )

I think there was one flight where we had one problem . It wasn't ours, but there was that one flight. Other than that, I believe the answer to the remaining flight is yes.

Furthermore, there is a group of examples from two of Bimer and Ward's classes. (the hearer-new tokens of hearer-old types, cf. (32), and hearer­ new entities with individuating descriptions, cf. (33» that are not easily subsumed under the list reading. All of these examples can be uttered out of 1 3 This type of example seems to occur typically with stress on the copula, which is another sign of the strong assertion of existence they need.

Unifying existential and list readings 1 95

the b lue, suggesting that they need not rely on a salient list in the context. An interesting and possibly important fact about these examples is that they . occur with a PP. There was the usual crowd at the beach today. There were the same people at both conferences. There was that deaf comedienne I was telling you about on TV today. b. There was the stupidest article on the reading list. There is something strange about the definites in (32): the definite article seems to be required internally. A lso, one m ight argue that the examples in (33) need not be considered under a definite restriction, because they have a .flavour of expressing an opinion about the postverbal DP. But even if we exclude (32) and (33) we are still left with the examples in (30), which '�:s::': 1th my eye are neither easily put aside nor can be subsumed under the list read ing. Finally, an informal search in the British National Corpus provides "fil1r1"hl"!r examples where a list reading seems not to be available: Almost every nation in central and Latin America, Africa and South East Asia seems to be suffering from environmental decline, and although the basis of quantification might be questioned, at least there is the framework ofa sub-regional, quantified data base on environmental decline. (BNC, text: APN, 38) There isn't the accommodation to cram in those extra people! (BNC, text: AMD, 593). , ', As

.

my approach allows for definites occurring with the existential reading,

:;::;i!,;,:,\this suggests that my approach is on the right track, although I have to leave

·'i;" · ;' · the question of an appropriate definition of the definiteness effect to further research.

t�,i:, .

References

�dger

David, and Gillian Ramchand. 2003. Predication and Equation. Linguistic BelIetti, Adriana. 1 988. The Case of Unaccusatives. Linguistic Inquiry 1 9: 1 -34. Bimer, Betty, and Gregory Ward. 1 998. Information Status and Noncanonical Word Order in English. Amsterdam! Philadelphia: John Benjamins. ,

,'

.. Inquiry 34: 325-359.

1 96 JuUa M Hartmann

Doron, Edit. 1 988. The Semantics of Predicate Nominals. Linguistics 26: 2 8 1 -303. En�, MUvet. 1 99 1 . The Semantics of Specificity. Linguistic Inquiry 22: 1 -25. Hartmann, Jutta M. 2005. Why there is(n't) wh-movement in there-constructions. In Linguistics in the Netherlands, ed. by Jenny Doetjes and Jeroen van de Weijer, 87-98. Amsterdam: Benjamins. Heycock, Caroline, and Anthony Kroch. 1 998. Equation and Inversion in Copula Clauses. In Papers in Linguistics 10, 7 1 -87. Berlin: ZAS. Higgins, Francis R. 1 973 . The Pseudocleft Construction in English. PhD thesis , MIT. [Published by Garland in 1 979]. Huber, Stefan. 2002. Es-Clefts and Det-Clefts. Zur Syntax, Semantik und Informationsstruktur von Spaltstitzen im Deutschen und Schwedischen. PhD Dissertation, University of Lund. Keenan, Edward L. 1 987. A Semantic Definition of ' Indefinite NP'. In The Representation of (In)dejiniteness, ed. by Eric Reuland and Alice ter Meulen, Alice, 286-3 1 7 . Cambridge, MA: MIT Press. McNally, Louise. 1 992. An Interpretation for the English Existential Construction. Ph.D.thesis, University of California, Santa Cruz. Moro, Andrea. 1 997. The Raising ofPredicates: Predicative Noun Phrases and the Theory o/Clause Structure. Cambridgel New York: CUP. Rothstein, Susan. 200 ) . Subjects and Predicates. Dordrecht: Kluwer. Safir, Kenneth. 1 985. Syntactic Chains. Cambridge/ New York: CUP. Williams, Edwin. 1 994. Thematic Structure in Syntax. Cambridge, MA.: MIT Press. Zamparelli, Roberto. 2000. Layers in the Determiner Phrase. New Garland.

York/London:

• Extended projections - extended analogues: · a note on Hungarian PPs Veronika Hegedus

:1

.

.

Introduction

Riemsdijk ( 1 990) is an early work on the lexical and functional , structure of adpositional phrases (from now on abbreviated as PPs, t,: Ptcluding pre-, post-, and circumpositionai phrases), with special regard to ·" , their headedness properties and to their categorial status. Taking German " ::i dircumpositional phrases as his main data, he argues that in the case of ' Ge Ji';; rmanic PPs, we find a functional projection on top of the lexical j�:. projection that is categorially identical with the projecting lexical category. ;jh his discussion of headedness properties, he proposes a Hungarian :'� jOalogue to the German data, drawing a parallel between the structures of ·:-" :; :'iPP s in the two languages by identifYing a functional layer in the Hungarian

>;' :�: PP. j:

My aim, in this paper is to consider this parallel between German and ungarian PPs, and extend it to another construction, which, to my eye, ;'" ��;:S�eins to be an even more transparent equivalent of the German PPs that t --, .'Van Riemsdijk ( 1 990) deals with. The Hungarian constructions under C:::!ifpi,a.iscussion contain not only what are traditionally regarded as PPs, but i!;!;';z:::::i '§tructures extended with what are categorized as directional particles or , :2il�cative adverbs. Without providing a' detailed analysis of the data, I will �= :,j;,fiighlight some points where I believe the structures of functional PPs in :=;:; ' , and Hungarian to be on a par. ,�

. , �=" ,

.

. . .

, ..

' German functional prepositions and a Hungarian analogue

l3esides having prepositional phrases and postpositional phrases, German J�lso has constructions that are best regarded as circum positional phrases ;;itsee the examples in ( 1 » . Van Riemsdijk ( 1 990) takes these phrases to . illustr ate his arguments about the functional projections of PPs. He assumes that there is a projection level, which he names pP, on top of the lexical PP; eo· ,:o.""':: " · a> functional projection on a par with vP or DP above the lexical VP and NP.' The functional layer is instantiated by the postposition that contains a dEHctic morpheme encoding semantic features that are not expressed by the

1 98 Veronika Hegediis

preposition. These features, in the case of German, are direction, orientation and/or proximity (Van Riemsdijk, 1 990: 239). I auf den Berg herauf auf mich zu unter der Brucke durch

(1)

(on the mountain up(+P» ace (towards me to) acc (under the bridge through) dat (Van Riemsdijk, 1 990: 233)

The postpositional element is separable from the rest of the PP. Both Van Riemsdijk ( 1 990) and Den Dikken (2003) assume that the postposition, that is the functional P, gets incorporated into the verb in some cases. The postposition behaves as a verbal particle when it moves to the preverbal position, as is illustrated in (2). Van Riemsdijk ( 1 990), however, provides several constituency tests to show that it actually does belong to the PP: they can be complements to both nouns and prepositions and they can move together. In the surface order then, the lexical layer of the PP is head­ initial in German, while pP is head-final . (2) a. b.

. . . dass . . . that . . . dass . . . that

er he er he

auf den Berg hinauf hat steigen wollen onto the mountain up has climb wanted auf den Berg hat hinaufsteigen wollen onto the mountain has up-climb wanted (Van Riemsdijk, 1 990: 234)

Van Riemsdijk (1 990) assumes the surface order to be the base-generated order; that is the pP is head-final throughout the derivation. Den Dikken (2003), while maintaining the idea that the postposition occupies a higher functional projection above the prepositional layer, derives the surface order from a head initial base-generated order with movement to a specifier position along the functional layers. For the purposes of this paper, it does not really matter which way we go, and for the time being I am mostly concerned with the surface order in the case of Hungarian as well. 1 Later analyses, like Den Dikken's (2003) discussion of German PPs, take features similar to those suggested by Van Riemsdijk ( 1 990) to instantiate functional proj ections above PP, each having its own semantic contribution. Also, Van Riemsdijk and Huijbregts (to appear) hypothesize separate functional projections corresponding to location and direction in the structure of PPs. I will use the original formulation for the time being and remain agnostic about what extended projections to assume and what their nature is.

Extended projections - extended analogues 1 99

In his more general discussion of headedness properties of languages, Van . Riemsdij k ( 1 990) mentions that Hungarian might be a language that shows exactly the mirror image of the German order. He briefly discusses a . property of the Hungarian PP, namely, that in demonstrative i;::cc)nstructl()nS certain Ps - the so-called 'dressed' Ps - are duplicated on the C "tlerrlon.strlatnle element, as can be seen in (3). a(z) mogott a hitz mogott that behind the house behind ' behind that house' ·' He suggests that this structure might be analogous to the Geiman circum positional phrases. The difference is that the Hungarian pP is the -mirror image of the German one, with the lexical P being postpositional 'and the functional projection prepositional. Also, there is a semantic parallelism: the Hungarian preposition (i.e. the reduplicated postposition) ';:e'mn�ssc�s a demonstrative feature marked by the a(z) morpheme, in the same way as the deictic morpheme on the German postposition expresses . proximity (Van Riemsdijk, 1 990). Maracz ( 1 9 89) introduced the name 'dressed' postpositions to refer to class of postpositions that exhibit certain properties, among others, this lication in determiner phrases. The most important characteristics of '1t�,�,Jh�ese postpositions are that, having developed from possessive s (and just like possessives), they take uninflected [i�;��!f'complements (cf. (4a)). Furthermore, they can be inflected for person­ Jhtimber agreement when their complement is a pronoun (see (4b), or when �the complement extracted, although this latter possibility is for many ; ;speakers no longer available.2

is 'M.essed' a.racz preposihastions,the butsameseejudgement about extractabiforlityaofdithefferentcomplvieement of Ki s s w. The i!N:�i['!��!H;�;iITHm!;!r�i';;i,">jkiexinn ples whose acceptability is under discussion are like the following: -nak. melbesildee iisatlt Mari Janos �2i!:::':!:';:i'(O John Mary-oAT 'John sat (down) beside Mary.' Kiss n assumes that this structure is derived by extraction of the o e!��s!����oc:::�:;���;���:� is��#g£:�:!:�:� �!1�� .,

, '=

,, r��:};

-,

( 1 989)

!ii

' ,:', . '

1:3 I�c-

'

E�

•.

( 1 998, 2002)

E.

( 1 998, 2002)

200 Veronika Hegediis

(4) a.

a hilz mogott the house behind 'behind the house' b. (en) mogott-em behind- l SG I 'beh ind me'

The class is defined in opposition to 'naked' postpositions, which take complements in the oblique case, and which cannot be inflected (a property that may be explained by their different historical origin). 'Naked' postpositions form the demonstrative construction without reduplication of the P, but reduplication of the oblique case marker on the complement, parallel to other case marked DPs. As we can see in examples (5) and (6), the marker of superessive appears on the head of the nominal and on the demonstrative both in the case of PPs containing a 'naked' P and suffixed nouns. (5)

az-on a foly6-n till that-sup the river-SuP over 'over that river'

(6)

az-on a pad-on that-SUP the bench-SUP 'on that bench'

Thus, the property of having a non-inf}ected complement and reduplicating the postposition are correlated. I will assume that (3) actually is like (6), that is 'dressed' Ps are like case markers. I will consider them to represent the same category. This is in line with E. Kiss (2002), who suggests that postpositions are like case endings, and PPs are like K(ase)Ps on the extended projection line of the NP. E. Kiss (2002) also argues that the class of ' naked' Ps, that is, Ps that take oblique case marked cbmplements must · be much smaller than is generally assumed, thus making the class of Ps more homogeneous. This means that the structure proposed by Van Riemsdijk (1 990) need not be restricted to a subclass of Hungarian Ps, but is applicable to all of them. This, however, is not the only construction where a parallel can be drawn between Hungarian and German PPs containing both pre- and postpositions. In my view the case of locative PPs combined with locative

Extended projections - extended analogues 201

'

particles and adverbs is another instantiation of Van Riemsdijk's ( 1 990) . structure and may be an even closer parallel. Another Hungarian analogue

'. .

Hun garian, besides the usual PPs, we also find sentences where the PP or the oblique case marked OP is somehow associated with what are generally .c. alled locative particles3 or adverbs. These two categories either express or location and they are in the preverbal position in neutral that these elements start out forming a pP with the (most assume I clauses. postverbal PP and are moved to their surface position during the I propose to treat these structures as parallel to the German ��;�;' ;,circ:ul1npcJsitional phrases discussed in the previous .- section, with the being the functional extension of the PP, thus indeed fifid:ho'WlJU! the mirror image of the German order. The meanings expressed by functional elements are orientation (as in (7a) and (7b » , directionality ,;t;'3'i'E" '�'I'�D'" (7b) and (7c» and proximity (cf. (7c» , similarly to the German ones. lent a volgy-ben down the valleY-INE 'down in the valley' be az agy ala b. into the bed under 'under the bed' c.

:j��,i� �� ���� :a�r��e

Hungarian examples are also similar to the German ones in that the ' �St ;;0 ;�functional P is often separated from the rest of the phrase, as is illustrated in The functional preposition moves to the preverbal position after the rest the pP has been moved out. Az eger be szaladt az agy ala. the mouse into ran the bed under 'The mouse ran under the bed.'

�.� The discussion only extends· to locative particles and not to the whole class of ; �tingarian particles. The fact that particles are very often secondary predicates (cf. . ; E;�Kiss (2004)) is perfectly compatible with my proposal, with the modification that it is the pP containing the particle that is often predicative.

202

Veronika Hegediis

This can be either an instance of incorporation, or phrasal movement to a specifier position . E. Kiss (2002) analyzes the only cases where she assumes P-movement to the preverbal position - moving 'dressed' Ps after extraction of their complements from the PP - as phrasal movement to the specifier of an operator projection above VP, so it m ight be applied here as well, but for the purposes of this paper, it is not relevant what the nature of this movement is and how the derivation proceeds. All that is relevant here is that this movement seems to take place in all neutral sentences, especially when the pP is predicative; that is why it hardly ever seems to be the case that we find pPs forming a constituent. That we are really dealing with a phrase can be shown by constituency tests: just like the German pPs, their Hungarian counterparts can also be complements to nouns (as is shown in (9» , they can be contrastively topicalized (see ( 1 0» or focalized (as in ( 1 1 » , and they can be used as answers to questions (cf. ( 1 2» . Furthermore, as can be seen in ( 1 3), they can appear in the PP-with-NP construction that Van Riemsdijk and Huijbregts (to appear) take to be a property of directional PPs. a varos a1att] (9) a. a csatoma [lent the tunnel down the city under 'the tunnel under the city' b. menekiiles [vissza a fedezek mage] escape back the dugout behind 'escape back behind the dugout' ( 1 0)

[Lent a pince-ben] nem maradt mar semmi. down the cellar-INE not remained yet nothing 'In the cellar, nothing is left. '

(1 1)

Ezt a torpe-t [KINT A KERT-BEN] talalt-am. this the dwarf-ACe out the garden-INE found- l SG ' It was outside in the garden that I found this dwarf? " ,

( 1 2)

Hova szaladt az eger? where ran the mouse ' Where did the mouse run?'

( 1 3)

[Le a pince-be] az tires iivegek-kel ! down the cellar-ILL the empty bottles-INSTR 'Down in the cellar with the empty bottles! '

[Be az agy ala.] into the bed under 'Under the bed.'

Extended projections

-

extended analogues 203

When the pP is not predicative in the sentence, the movement of the functional P to the preverbal position does not take place. ( 1 1 ) is an . example l ike that: the whole pP is moved to the focus position of the e. A further illustration is ( 1 4), where the adjunct pPs are postverbal .' ,claus act' showing that they do form a constituent. 'int an d '

Janos gyakorolta a tolatas-t [ki a garazs-b61] John practised the reversing-ACC out the garage-ELA es [be a haz moge]. and into the house behind 'John practised reversing out of the garage and behind the house. ' '' . Thus, the functional preposition expressing either direction or location • Jorms a constituent with the PP initially; but they are often separated during the derivation of the clause, especially when the pP is predicative, because 'then the functional preposition moves to the preverbal position. The is analogous to the German examples under ( 1 ), with the that the functional layer is prepositional and the lexical is ·���J.t':.;!· Dc.strJ,osiltional in Hungarian, thus showing the mirror image of the German �: Stru(�tuI'e, the same way as Van Riemsdijk ( 1 990) was already suggesting.

fact, there is another construction that seems to be related to the examples discussed in the previous section. This also involves a preverbal particle and a PP, where, probably importantly, the P is realized by a Lsuffixal element. In these examples, the preverbal element seems to be the . ;if:eduplication of the suffix, the form being completely or almost identical to . ::�����:.�:, :the suffix, as can be seen in the examples in ( 1 5) and (1 6). _ _;

1 :=]ljg;'ip s)

Janos be-ugrott a medence-be. John into-jumped the POol-ILL 'John jumped into the pool.' Mari hozza-vagta a tanyer-t a fal-hoz. Mary to - threw the plate-ACC the wall-ALL 'Mary threw the plate at the wall.'

204

Veronika Hegediis

position, both Manicz ( 1 989) and E. Kiss ( 1 998, 2002) analyze them to form a complex predicate with the verb and not to have a derivational relationship with the PP . E. Kiss (2002) considers the possibility that we are dealing with extraction of the complement of P followed by movement of P to the preverbal position, but she rejects it because the P appears in two copies, so it does not seem to be the result of derivation of the type she assumes in the case of dative marked complements (see footnote 2)4. In her analysis, the preverbal element and the postposition are co-indexed, so that they can have an agreement relationship, but the preverbal element is a full PP containing a pro and incorporated into the V, and the postverbal PP is an adjunct . 5 I assume that we are faced with the same phenomenon in these examples as in the previous section. The preverbal element originates inside a pP and is separated from it during the derivation when it moves to the preverbal position. It is not so easy, however, to prove the constituency of the assumed pP. It might have to do with the fact that the functional preposition is the same as the suffix, but we hardly ever find both of them in the structure when they form a phrase. The only exception I know of is the so-called PP-with-NP construction that can only be used with directional PPs (Van Riemsdijk and Huijbregts, to appear). ( 1 7)

[Be a medence-be] a gyerekek-kel! into the pool-ILL the children-INSTR ' Into the pool with the children ! '

se

. Whatever the reason i s why we do not seem to find many instances of this type of pP with the prepositional element and the postpositional phra forming a constituent, I assume that it must be related to the fact that the structure is a kind of reduplication structure. To the extent that the meanings encoded in the phrases and the relationship between the preverbal element and the postverbal PP are similar to the constructions discussed in the previous section, they should have a similar analysis as well . I propose

Kiblses i(1998) alcases, so mentiweonsarethatdealsiniceng nomi nala izleatixiocnal ofprocess. the In compl ex n these wi t h possi her (2002) analysis, thitos theargument is noposilonger present, probably because she assumes phrasal movement preverbal t i o n there. Sheis a proposes thibuts analin ythose sis oncases analothere gy toisexampl es whereadjunct. the complForement thde pro(noun), no postverbal detai l s an the analysis see Kiss (2002). 4 E.

S

P

is

P + V

of

E.

Extended projections - extended analogues 205

the analysis is in terms of extended PPs along the lines of Van Riemsdijk ( 1 990), but the details are left for future research.

?;Dilkke.n, Marcel den. 2003. On the syntax of locative and directional adpositional phrases. Ms. Kiss, Katalin. 1 998. Verbal Prefixes or Postpositions? Postpositional Aspectualizers in Hungarian. In Approaches to Hungarian Volume 6: Papers from the Amsterdam Conference, ed. by Casper de Groot and Istvan Kenesei, 123-148. Szeged: JATE Kiss, Katalin. 2002. The Syntax of Hungarian. Cambridge: Cambridge University Press Kiss, Katalin. 2004. Egy igek<>toelmelet vazlata. [Outline of a theory of the verbal particle] Magyar nyelv [Hungarian Language] 50: 1 5-42 0;MlanlLCZ, Laszlo. 1 989. Asymmetries in Hungarian. PhD dissertation. Groningen :RiiemlsdiYk. Henk C. van. 1 990. Functional prepositions. In Unity in diversity: Papers presented to Simon C. Dik on his 50th birthday, ed. by Harm Pinkster and Inge Genee, 229-24 1 . Dordrecht: Foris. ?Rieml:;dlJk. Henk C. van and Riny Huijbregts. to appear. Festschrift for Joseph . Emonds. Mouton de Gruyter.

Against the sonority scale : evidence from Frankish tones Ben Hermans and Marc van Oostendorp

1. Introduction

While there is widespread agreement that sonority is a major force driving a wide variety of phonological phenomena, there are at least two very different opinions as to the implementation of this concept into the grammar. One view holds that sonority is to be defined in terms of a scale; it is an independent theoretical notion, in the sense that a segment's degree of sonority is read off from the scale, which functions as an external look­ up table. The scale may have an independent phonetic motivation, as in Wright (2004). The opposing view claims that sonority does not deserve the status of an independent theoretical notion, but is defined solely in terms of representations that can be motivated on independent grounds. It is not so easy to solve this issue on conceptual grounds, because, obviously, deriving sonority from representations goes at the cost of complicating these representations, while, conversely, relying on an external scale implies that the representations can be simplified. Fortunately, however, it is possible to decide this question on empirical grounds, and this is what we set out to do in this article. Our argument runs as follows. It is an essential characteristic of the sonority scale that rules, or constrai1).ts for that matter, always refer to a contiguous range of positions defined on it. We show that Contiguity of Reference, as we will call it, does not always hold; phenomena do exist that refer to discontinuous positions on the scale. This is a fundamental problem for all approaches of sonority that are based on a scale, but, interestingly, this problem does not exist for the representational approach, which precisely predicts that certain discontinuity effects should exist. This, then, is an important argument against the sonority scale and in favour of a representational approach, which holds that sonority is an epiphenomenon. The phenomenon we will study comes from the historical phonology of Frankish tone. At the point in time when the tones came into existence in the Frankish dialects, falling tones could only occur on vowels of relatively high sonority (mid vowels and low vowels) and also on certain sonorant consonants (velar nasal and rhotic liquid). Crucially, high vowels could not

Against the sonority scale 207

a falling tone. This, then, constitutes a case where Contiguity of :�11:ofp.r.p.n(=e is not respected. Far from being a problem, the distribution of the tone in Frankish dialects is exactly what is predicted by a ;e'DJ�esfmt�ltI·lonal approach towards sonority, which claims that the IAI­ Helnelll adds to sonority. Here, we follow Scheer ( 1 999) and particularly der Torre (2003) in claiming that segments carrying the IAI-element a relatively high degree of sonority. Interestingly, the latter author also following Sm ith (2000), that the velar nasal and the rhotic liquid e:sts, SUJ�g specified for this element. Adopting this proposal we are able to define vowels, mid vowels, lfJI and Irl as a natural class. Approaches toward sonority

is;;:�Rel�ar�a l· ng the representation of sonority two fundamentally different ssb�t>SlltlOIlS exist: there are those claiming that sonority is an epiphenomenon, that it can be read off from the representations, and there are those lairninlg that sonority deserves the status of an independent theoretical tool, u� �:onon·1tv scale. fAJlnOl1lg the representational approaches we can make a further division . .,AC:COI'dmg to one view, sonority is determined by absolute complexity. means that a segment's sonority increases with every individual ,������fbtllre it contains. This view has been popular among some supporters of rQV�2mlmelrlt Phonology. Particularly interesting proposals in this tradition Harri s (1 994) and B rockhaus ( 1 992), to name but a few. It has been :j;Cf,itJ:u:d both within (Scheer 2004) and outside of Government Phonology 1 999), and it is fair to say that it is too restrictive. (This sfPC,sltllon actually entails Contiguity of Reference: since all segments can be on a scale ranging from the least complex to the most complex, we sonority effects to be sensitive to whole classes of segments of some of complexity. Our arguments against sonority scales will thus also tranlSlalte to this approach.) The second subtype holds that only certain properties of a segment's �;!�'��l!m�ctwie determine sonority. In this view sonority is determined by relative ;; ' �qQnilpl�eXI1ty. Influential proposals in this tradition are Clements (1 990), in a traditional feature theory, and Rice ( 1 992), based on the Spontaneous Voicing.·Element-based versions of this approach have developed by Scheer ( 1 999), and, independently, Van der Torre (�003). Both authors claim that the IAI-element adds to sonority. Formally,

208 Ben Hermans and Marc van Ooslendorp

this is implemented by stating that segments containing this element favour prominent positions (like the rhyme and the nucleus, or the stressed syllable), and avoid non-prominent positions (like the onset of a syllable, or unstressed syllables). Since this proposal will turn out to be crucial in our account of the Frankish facts in section 3, it deserves some extra attention. In Element Theory (Harris 1 994, Harris and Lindsey 1 995) mid vowels and low vowels carry the IAI-element, whereas high vowels lack it. To illustrate we give the relevant parts of the five cardinal vowels.

( 1)

high vowels

1 = [i]

u = [u]

mid vowels

{I,A} = [e]

{U,A} = [0]

low vowel

A = [a]

It is well known that in many languages mid and low vowels attract stress from high vowels (cf. De Lacy 2002 for a typology of stress attraction). This shows, according to De Lacy, that these vowels favour prominent positions. In tenns of Element Theory this is tantamount to saying that the IAI-element favors prominent positions. In this interpretation stress moves until it reaches a segment carrying IAI, thereby satisfying IAI-' s desire to occupy a prominent position. Another widespread phenomenon is vowel reduction. Very frequently, vowel reduction takes the fonn of raising of mid and low vowels to high vowels (cf. Crosswhite 1 999 for a typology of reduction). In tenns of elements, this type of reduction can be understood as IAI-'s desire to avoid an unstressed (non-prominent) position. In this interpretation a segment loses its IAI when it happens to occur in an unstressed, i.e� non-prominent, position. What is crucial to us here is the observation by Van der . Torre (2003) that Irjl and Irl behave as if they are more sonorous than the other segments of their manner class. To account for this, he proposes that both segments contain the element IAI (cf. Smith 2000). Interestingly, this uncovers a parallel with the behaviour of the IAI-element in stress patterns; what is true for IAI at the suprasyllabic level (where foot structure or stress is located) also holds at the subsyllabic level; Igl and Irl favour prominent positions in the syllable (like the rhyme) and shun non-prominent positions )like the onset). .

Against the sonority scale 209

Evidence that l'fjl is favoured in the rhyme comes from a wealth of data that in many languages nasals tend to velarize in the rhyme wing ,sho iL . DCJs itlon. A dialect where this happens, at least under certain conditions, is Dutch. In (2) we contrast Standard Dutch with this dialect ;i;{1�aeldentlan 1 980, Van Oostendorp 200 1 , Van der Torre 2003 : 1 07). Standard Dutch [yrun] [sxun] [zo:n]

Antwerp 'green' [yryl] ] [sxul]] [za1J]

'shoe' 'son'

Velarization of nasals in the Antwerp dialect can be seen as a close relative , stress attraction. If there is no IAI in the stressed syllable then the ;�DIIOnlOI()� corrects this by moving the stress. Likewise, if there is no IAI in rhyme then the phonology corrects this by velarizing a nasal. Evidence that the velar nasal avoids subsyllabic non-prominent ji!PJDSl1tI0T1S (the onset in particular) is very easy to find. There are many ifleUl�'uaJ�es were IfJl is not allowed in onsets (Anderson 2005). This is ,'p..l1'ujv.alelnt to IAI-'s behaviour at the suprasyllabic level, where it typically the fonn of vowel reduction, as we have seen. The rhotic exhibits the same behaviour; it favours prominent positions shuns non-prominent ones. Vocalization of Ir/ to an a-like vowel can be as an instance of the former: IAI moves from a marginal coda position a syllabic peak. This process applies in Standard German (Wiese 2000), also in the Dutch dialect of The Hague (cf. Elias and Goeman 2003 ; der Torre 2003 : 1 60). In (3) we compare this dialect with Standard Standard Dutch [:lnd�r] [dlxt�r] rver]

The Hague [�nda] 'underneath' [dlxta] 'poet' [veal 'far'

Corroboration of the hypothesis that Irl shuns non-prominent positions at the subsy llabic level is a bit harder to find. Indicative, perhaps, is the fact that in Campidanian Sardinian Irl is avoided at least in word initial onsets, 'tIt · ough not in word internal ones (Bolognesi 1 998). . , In this subsection we have reviewed one particular version of a representational approach towards sonority. According to this theory the

2 1 0 Ben Hermans and Marc van Oostendorp

IAI-element contributes to sonority. Obviously, there is no scale in this theory. Instead of this we have constraints expressing that IAI favours prominent positions and disfavours non-prominent positions� both at the suprasyllabic and the subsyllabic level. The former accounts for frequently attested patterns of stress attraction and vowel reduction; the latter explains the rather striking behaviour of Iql and Irl with respect to syllable structure. 2. 2. The scalar approach From a quantitative bibliographical point of view, representational theories of sonority are somewhat marginal in the literature. By far the most dominant position is that segments are 'extrinsically' ordered according to their degree of sonority (see Wright 2004, for a recent overview). The segment classes we are interested in here are usually ordered in the following way : (4)

The sonority scale: obstruents < nasals < liquids < high vowels < mid vowels < low vowels

The segment's internal structure is immaterial as far as sonority is concerned in this approach. (Of course, a segment must be identifiable in terms of the sonority scale, and for this purpose we presumably need some representational residue.) Here, we are interested in a particularly important general property of scalar approaches. This · is the property we have termed Contiguity of Reference: suppose we have a constraint stating that a segment of a certain degree of sonority avoids a stressed syllable because its sonority is too low. Then, according to Contiguity of Reference, it is the case that any segment of lower sonority is not favoured in that position either. This would mean, for instance, that if a mid vowel avoids a stressed syllable, then surely, high vowels and (syllabic) nasals would also avoid stressed syllables, and a situation where mid vowels and nasals avoid stress but high vowels do not, is unimaginable. The converse is also true: suppose that a segment avoids a non-prominent position because its sonority degree is too high. Then it should be the case that any segment of greater sonority also avoids that position. We do not know of any work within the scalar paradigm that puts Contiguity of Reference into question and it is presumably correct as far as stress-related phenomena are concerned. In other areas, however, there is some reason to be sceptical about it. In the next section we will study the distribution of falling tone in Frankish dialects where Contiguity of

Against the sonority scale 2 1 1

Reference breaks down, causing a problem for the scalar approach and supporting the representational approach relying on the particular status of IAI-element.

dialects of the border area of the Netherlands, Belgium, Luxembourg Germany are polytonic. In the traditional terms of historical they can be classified as South East Frankish (the dialects of c1[010gy di8.1e �lB�elglan and Dutch Limburg and also an small area in Germany around the of Dillken) and Central Frankish, the latter consisting of Ripuarian in the north and Moselle Frankish in the south. Synchronically, the dialects two contrastive tones, traditionally called a 'bumping' (falling) tone a 'dragging' tone (Schleifton and Stosston). The latter has a variable in modem dialects, but phonologically it is a high tone. We ;':as!;Urrle that this was also the case when the tones came into existence. At initial stage of their development the distribution of the two tones was ly predictable. It is this historical stage we are interested in. This stage in the 7th century and ends in the 9th • I The data in this section are taken from Maasbracht, a (tonally) very :conse:rvaltl' ve dialect in the heartland of what is now the Dutch province of This is the native dialect of the first author (see also Hermans 994). We refer the interested reader to the fol1owing important sources for l::�,C;��;;;pon1rlrnlat'l on as to the historical distribution of the tones: Frings ( 1 9 1 3 , 1 6), Welter ( 1 929), Dols ( 1 953), Schmidt ( 1 986) and De V an ( 1 999). the sake of brevity we wiI1 only discuss monosyllabic forms. This does do any harm, because at the initial stage of the tonogenesis, iMiill�!::� -I)ol){syllabic and monosyllabic forms behaved identically. Long vowels that were either mid or low in West Germanic received a ;:;'�'§gg�; ';. Jailing tone. This is illustrated in (5). The falling tone is indicated with an :: H.L-s:up,erscrr·pt over the vowel.

a

tones devel oinpedg dinophthongs later thanmustthehavecentury because monophthongi zatiknow on of theTheGennani c fal l appl i e d before they di d , and we zatOpenion stSylartlaedbletoLengtheni work theng startcent ury.work.AlsThere o, the itsonesevidmust ·h�vemonophthongi existedthatbefore e d to ence thi s rul e was acti v e al r eady before t h e begi n ni n g of the e (Kyes 1969). I

.

that

suggesting

7th

in

7th

l Oth c

ntury

2 1 2 Ben Hermans and Marc van Ooslendorp

(5 )

Long mid and low vowels received falling tone

[bre :H Lf]

e: < e:

[ho : HLt]

WGM e: ' letter' WGM o:

'hat'

WGM

' snow'

L [vra:H x]

'n

ice '

' ea

rly '

« ai)

[i�H L r]

' honor'

WGM �: « au)

[lu�HLn]

' bread'

'thread'

e:

e:f]< e o L [ le : H

'wage'

WGM a: ' heavy'

Short vowels followed by Irl or a velar nasal also received a falling tone, as shown in (6). (6)

Iql and Irl received falling tone L [v alJH ] HL [balJ ] HL [rolJ ] [ba�L] HL [h ar ] [h�rHL]

'to catch '

' afraid' 'rank' '

severe '

proper name 'wire gauze'

L [stOlJ H ] HL [brIIJ ] HL [h01J ]

[vreIlL] HL [strer ] HL [ krer ]

' standpast '

' bring'

' hangpast ' ' tar'

. 'star' 'cart'

If a short vowel is followed by an obstruent, then the level high tone appeared. A few examples illustrating this are given in (7).

Against the sonority scale 2 1 3

Short vowels followed by an obstruent have a level high tone 'head '

[k�Hp] [buHk] [wtt]

' hilly goat' 'white'

[ laHf] [vol\] [moHs]

'cowardly' 'fluid ' 'must'

The level high tone consists of a simple tonal element, phonologically; we . . represent this with H. It is inserted to satisfy a (hard) constraint requiring the head mora of the word carry the tonal segment H. The high tone of first mora is not allowed to spread to the second mora, due to another Y;Ciom,tralmt according to which a non-head mora should not have a high tone De Lacy 2002b for the family of constraints regulating the relations hetwe(�n tonal structure and prosodic structure). Accordingly, the schematic �?,j;eDl·esfmtl:ttl·(m of a bimoraic word with a level high tone is as in ( 8). The level high tone H

I

II

II

V

falling tone consists of two tone segments, H and L; we assume that latter segment is inserted to ensure that a mora is not toneless. The '{sc!bet1rlatOlc representation of a falling tone is therefore as follows: The falling tone H

L

I

I

J.l

Jl

V

the question is, of cQurse, why not all bimoraic words received a low :" tone. It is here that sonority becomes relevant: apparently, the second mora :' ::::�ould only carry a tone if it was sufficiently sonorous.

2 14

Ben Hermans and Marc van Oostendorp

Here we reach the critical point in our argumentation. Notice now that the low tone is allowed in words ending in a velar nasal, as the examples in (6) show. This being the case, the scalar approach to sonority makes the following prediction: if a (velar) nasal is sufficiently sonorous to carry a tone, then surely any segment of greater sonority also licenses a tone. The prediction is correct for the rhotic liquid, as we have seen in (6). However, it is blatantly wrong in other cases. Let us have a closer look at them. High vowels and falling diphthongs did not receive a falling tone. Examples illustrating this are given in ( l 0).

( 1 0)

Long high vowels and falling diphthongs received a level high tone [wi:llt] [vu : H I] [kIEiH t]

WGM i:/u:/ai/au [ti:Ht] 'far' ' dirty ' [ IE iHt] 'dress' [b:>uHm]

'time' 'grief' 'tree'

Short vowels followed by a non-velar nasal, or by a lateral liquid also got a level high tone. This is exemplified in (1 1 ).

(1 1 )

short vowels followed by a tautosyllabic sonorant consonant received a level high tone [moHn] [k loHm]

[m:>1I 1] [v :>H: l]

'man'

[mIHn]

'nasty'

' hardly'

[sloHm]

'coal mud'

' mole'

[vreH:l] [bnH :I]

' skin'

'full'

'glasses'

Although the velar nasal is sufficiently sonorous, the other, nasals are not, and neither are the lateral liquid or the high vowels. Perhaps it is possible to redistribute the consonants over the scale, such that the two liquids become separated by the velar nasal. This would create a scale of the following type: ( 1 2)

An attempt at a solution : obstruents < n,m < I < IJ < r < high vowels < mid vowels < low vowels

Against the sonority scale 2 1 5

. In this scale we can locate the cut off point between lJ and the lateral liquid To the left of this point falling tones are no longer allowed. While this correctly accounts for the consonants it still does not solve the problem of the gap constituted by the high vowels. To explain this in a scalar approach, we would need the following absurd scale: .

.

An absurd scale: obstruents < n,m < I < high vowels < lJ vowels < low vowels

<

r < mid

This scale makes many strange predictions. Let us briefly look at just one. it has been well documented that syllable peaks are sympathetic to 'relatively h ighly sonorous segments (cf. Prince and Smolensky 1 993). Now the sonority scale in ( 1 3), in tandem with Contiguity of Reference, predicts .that i f high vowels are sonorous enough to occupy a syllable's peak, then ;,�l1'·P.IV rhotic l iquids and velar nasals are able to do the same. This certainly an absurd prediction from a typological point of view. Further, ( 1 3) is unprincipled, in the sense that there is absolutely no reason why it is istructure:d the way it is. We can conclude, then, that the distribution of the . " ling tones at the initial stage cannot be accounted for within a scalar �pproach of sonority. The main problem is that it either violates Contiguity . Reference, or an absurd scale must be adopted. In strong contrast to this, the Representational Approach we have Skt�tclled before can explain the distribution of the tones quite easily. We have to say that a low tone must be linked to an IAI-bearing element. representation in ( 14a) is rejected by this constraint, whereas the one in ) is accepted. a.

H

I

�

L

I

�

b.

H

L

Jl

11

I

I

I

A

.• . The licit structure in ( 1 4b) represents mid and low vowels, but also the . · . yelar and the rhotic liquid. These segments constitute the class of IAI-

9arrying segments, as we have seen in the preceding section. ': One might object that this kind of approach makes the same kind of ��azy predictions as the scalar approach. We know that syllable peaks Javour high sonority, so one might consider a constraint stating that a

216

Ben Hermans and Marc van Ooslendorp

syllable peak must contain an IAI-element. This would predict the existence of a language that allows m id and low vowels in the peak, and in addition to that the velar nasal and the rhotic liquid, but not high vowels. Clearly, we are in need of a principled theory of visibility. We suggest that prosodic heads, like the peaks of syllables and feet, can only see those place elements that are heads. This would imply that they only access major category features and those place features that are heads. According to Van der Torre, the IAI-element is not a head in the domain of a consonant; in Irf, the element III is the head, and the velar nasal has no heading place element. If this is true, then peaks can only see the IAI-elements of low vowels and of mid vowels, but the latter is only true if the mid vowels are IAI-headed. Other subsyllabic positions, like the onset and the dependent position of the rhyme, should be able to access further details of segmental structure as we have seen in the second section. This allows a constraint that excludes a velar nasal in onset position, or a constraint that favors a velar nasal in the dependent position of the rhyme. Tone, as we have seen in the preceding section, must also be allowed to have access to further details of segmental structure. The fact that we have to rely on a visibility theory that still must be developed can admittedly be considered as a weak point, but we want to point out that something similar is true for the scalar approach. It has been p ointed out by Clements ( 1 990) that the major class features provide sufficient information to explain what is allowed in peaks. In fact, this was his main reason to get rid of the traditional sonority scale and to derive it from the major class features. While it has been established by now that Clements' claim is too restrictive in it� details, it still is true in spirit; for certain phenomena, like the phonology of prosodic peaks, the sonority scale is far too rich. Peaks can only see a modest subpart of the richly structured sonority scale. Clearly, then, proponents of scalar approaches to sonority must equally rely on some kind of visibility criterion, which still remains to be developed.

4. Motivating IAI on loJ and Irl We now tum to a further interesting consequence of our proposal; if it is true that Igl and the rhotic liquid carry the IAI-element, it is predicted that they exhibit lowering effects on adjacent vowels. In this section we will show that this is indeed the case. Veatch (199 1 ) observes that in certain dialects of Alabama vowels are lowered by a velar nasal: in words like spring, jinger, thing, etc. the vowel

Against the sonority scale 2 1 7

i s real ized as [re]. Veatch refers to this process as 'Alabama Lowering'. Interestingly, in his view this is ' an unnatural, anticoarticulatory effect' . (Veatch 1 99 1 : chapter 9). It might be true that Alabama Lowering cannot be understood phonetically, but it can be understood phonologically given point of view just developed; it is a case where the IAI-element seeks . the peak position of the nucleus, thereby lowering the vowel. Something similar happens in Shiaoxing Chinese. According to Zhang (2005), a constraint is active in this language, stating that high vowels, or glides, are not tolerated after a velar nasal: * [lJ] [+high] : [lJ] cannot occur before any [+high] (semi-)vowel e believe that this constraint would be more profitably formulated in terms of Element Theory, as a case where the IAI-element spreads to the 2r::,,;eaIIC providing independent motivation for the hypothesis that a velar nasal carries such an element? The lowering effects of Irl are well known. Lindau ( 1 985) observes that Irl frequently lowers a preceding vowel. In Canadian English, for •. in!starlce, [E] is lowered to [re] before Irl, resulting in neutralization. There is distinction, for instance, between merry and marry (Woods 1 993). ' ]:'JnerlOnlel1la like this provide independent motivation for the hypothesis that phonological representation of the rhotic liquid contains the IAIAnother piece of evidence in favour of the same assumption comes so-called non-rhotic varieties of English. The literature on this topic is we cite Kahn ( 1 9 76), Broadbent (199 1 ), Ortmann ( 1 998), Bakovic • . (1 999) and Kramer (2005) as examples of work from many different · tn(�OrtmC;al backgrounds that are all compatible to a view which takes Irl to ,bave an IAI element. . .. Many varieties of English avoid vowel hiatus with an intrusive Irl, but only if the preceding vowel is not high. This is an entirely productive '£�,.J)rOlcess. which happens for instance also in English accented pronunciation . .. .. . . foreign languages and in loanwords. Some examples il lustrating this are in ( 1 6) .

s (2005) Alewthofoughsyllheabledoesrhymesnot saythat thithes velexplaricnasal itly, itinalasocodaseemscantruonleyinbeZhang' vinon-hi preceded by a gh vowel.

.2

2 ] 8 Ben Hermans and Marc van Oostendorp

( 1 6)

j ' etais d6ja[r] ici UEFA[r] officials

(Wells 1 982:226) (Sebregts 200 1 :25)

The scholars mentioned all note that these phenomena find a parallel in the fact that after high vowels hiatus is resolved by the insertion of a homorganic glide: ( 1 7)

the key[j] is the zoo[w] is

Although implementations differ, Ir/ is a glide for the non-high vowels, according to the scholars just cited. This allows them to formulate both types of hiatus resolution in a uniform way; both instrusive Irl and homorganic glide insertion are cases where the vowel's features spread to the right to fill the empty onset. In terms of Elements this is independent motivation that the rhotic liquid carries an IAI-element. 5. Conclusion

Obviously, several questions remain open. For one thing, it would be interesting to know whether similar asymmetries between velars and other places of articulation can be observed within other manner classes. At present, we have no evidence that velar obstruents would display a lowering effect similar to the velar nasal, for instance, although a logical extension of the claims made here would lead to such a prediction. Another, more theory-internal, issue is why it is that IAI, rather than III or lUI, makes the segment more sonorous. in principle, Element Theory would treat these elements as completely symmetrical, .but that does not seem to conform our analysis of the facts. Leaving these questions open for future research, we think we have still reached a conclusion of some general theoretical importance. We have contrasted two views of sonority effects: one which is based on sonority as a scale (which may be motivated either in terms of phonetics or in terms of phonological complexity), and one in which specific phonological elements contribute to the sonority of a segment. In particular, we have seen that IAI is part of the representation of the liquid Irl and the nasal fg/, and this makes these segments behave as more sonorous than other l iquids and nasals in their tone licensing capacity. This predicts that these two segments will interact with low vowels in another way, and we have provided some evidence for that as well.

Against the sonority scale 2 1 9

References Anderson, Gregory. 2005. Areal and phonotactic distribution of N. In: Marc van Oostendorp and Jeroen van de Weijer (eds.), The internal structure of the phonological segment. BerlinlNew York: Mouton de Gruyter, 2 1 7- 1 34. Bakovic, Eric. 1 999. Deletion, insertion, and symmetrical identity. Manuscript, Harvard University. Bolognesi, Roberto. 1 998. The phonology of Campidanian Sardinian: A unitary account of a self-organizing structure. Doctoral dissertation, University of Amsterdam. Broadbent, Judith. 1 99 1 . Linking and intrusive r in English. University College London Working Papers in Linguistics 3: 28 1 -302. Brockhaus, Wiebke. 1 992. Final devoicing: principles and parameters. Doctoral dissertation, University College, London. Clements, George N. 1990. The role of the sonority cycle in core syllabification. In: J. Kingston and M. Beckman (eds.), Papers in Laboratory Phonology I: Between the grammar and physics of speech. Cambridge: Cambridge University Press, 283-333. .. ' .Crosswhite. Katherine. 1 999. Vowel reduction in Optimality Theory. Doctoral dissertation, University of California, Los Angeles. Willy. 1 953. Sittardse diftongering. Sittard. , Michael, and Ton Goeman. Haags. Den Haag: Sdu. Theodor. 1 91 3 . Studien zur Dialektgeographie des Niederrheins zwischen Dusseldorfund Aachen. Marburg. Theodor. 1 9 16. Die Rheinische Accentuierung: Vorstudie zu einer Grammatik der Rheinische Mundarten. Marburg. ·�;:HSl rrjc: John. 1 994. English sound structure. London: Blackwell. John, and Geoff Lindsey ( 1 995). The elements of phonological representation. In: 1. Durand and F. Katamba (eds.), Frontiers of Phonology, 34-79. Harlow, Essex: Longman. Hermans, Ben. 1994. The composite nature of accent. With case studies of the . Limburgian and Serbo-Croatian pitch accent. Doctoral dissertation, Free University Amsterdam. Daniel. 1 976. Syllable-based Generalizations in English Phonology. Doctoral dissertation, MIT. · J{Jiim er, Martin. 2005. English schwa insertion before liquids and phonological opacity. Manuscript, University ofTroms0 . . Kyes, Robert L. 1 969. The Old Low Franconian Psalms and Glosses. Ann Arbor: University of Michigan Press. . Lacy . • Paul de. 2002a. The formal expression of markesness. Doctoral dissertation. University of Massachusetts. Amherst. Lacy, Paul de. 2002b. The interaction of tone and stress in Optimality Theory. Phonology 1 9, 1-32.

220

Ben Hermans and Marc van Oostendor.p

Lindau, Mona. 1 985 . The story of Ir/. In: Victoria Fromkin (ed.), Phonetic Linguistics: Essays in honour of Peter Lade/oged. Orlando: Academic Press,

1 57- 1 68. Oostendorp, Marc van. 200 I . Nasal consonants in Dutch and some related systems. Neerlandistiek. nl 200 1 .08. http://www.neerlandistiek.nVpublish/artic1es/

000049/artic1e.html Ortmann, Albert. 1 998. Consonant epenthesis: Its distribution and phonological

speCification. In: Wolfgang Kehrein and Richard Wiese (eds.), Phonology and Morphology ofthe Germanic Languages. Tilbingen: Niemeyer, 5 1 -76.

Prince, Alan, and Paul Smolensky. 1 993. Optimality Theory : Constraint interaction in Generative Grammar. Report no. RuCCS-TR-2. New Brunswick, NJ:

Rutgers University Center for Cognitive Science. Rice, Keren. 1 992. On deriving sonority: A structural account of sonority relationships. Phonology 9, 6 1 -99. Sebregts, Koen. 200 1 . English [r]-Liaison: Rule based theories, Government Phonology and Optimality Theory. M.A. thesis, University of Utrecht. Scheer, Tobias. 1 999. A theory of consonantal interaction. 20 1-237.

Folia Linguistica 32,

Scheer, Tobias. 2004. A Lateral Theory 0/ Phonology. BerlinlNcw York: Mouton de Gruyter. Schmidt, Jiirgen E. 1 986. Die mittelfriinkische Tonakzente (Rheinische Akzentulierung). Stuttgart. Smith, NONal. 2000. Dependency Theory meets OT. A proposal for a new approach to segmental structure. In: Joost Dekkers, Frank van der Leeuw and

Jeroen van de Weijer (eds.), Optimality Theory. Phonology, syntax and acquisition. Oxford: Oxford University Press, 234-278. Taeldeman, Johan. 1 980. Inflectional aspects of adjectives in the dialects of Dutch­ speaking Belgium. In: Wim Zonneveld et a1. (eds.), Phonology. Den Haag: Martinus Nijhoff,0265-292.

Studies in Dutch

Torre, Erik Jan van der. 2003 . Dutch sonorants: The role of place of articulation in . phonotactics. Doctoral dissertation, University of Leiden. Vaan, Michiel

de. 1 999. Towards an explanation of the Franconian tone accents. O Amsterdamer Beitrage zur A lteren Germanistik 5 1 , 2 3-44. -

Veatch, Thomas, C. 1 99 1 . English vowels: Their surface phonology and phonetic implementation in vemacular dialects. Doctoral disseriition, Univers ity of Pennsylvania. Wells, John. 1 982. Accents ofEnglish. Cambridge: Cambridge University Press. Welter, W. 1 929. Studien zur Dialektgeographie des Kreisen Eupen. Bonn. Wiese, Richard. 2000. The Phonology of German. Oxford: Oxford University Press. Woods. Howard. 1 993 . A synchronic study of English spoken in Ottawa. In: Sandra Clark (ed.), Focus on Canada. Amsterdam: Benjamins, 1 5 1 - 1 78

WrigRobert ht, Richard. reviandew ofDonca perceptual cues(eds),and cue robustness. In: Bruce Hayes, Ki r chner, Steri a de Cambri d ge: Cambri d ge Uni v ersi t y Press, Di s sertat i o n, The Phonol o gy of Shaoxi n g Chi n ese. Doct o ral Zh. ang,LeidJiensheng. University. Again�t the sonority scale 22 1

A

Phonetically Based Phonology.

2005.

34-57.

"

Classifiers, agreement and honorifics in Japanese Masaru Honda

1.

Introduction

Japanese is considered to he a language which lacks syntactic agreement as observed in many Indo-European languages. It does have Case-marking on argument NPs (DPs), but hardly any agreement morphology appears in the verbal domain. I It could be the case that features such as gender, person and number, namely q>-features, play no grammatical role in Japanese in the way that they do in the Indo-European languages. However, it turns out that semantic categories are significant in Japanese in making syntactic distinctions. ( 1 ) demonstrates that alternation of two verb forms involves the semantic feature [± animate] on the subject NP. (1)

a.

Gakusei-ga iru. Student-NOM be 'There is a student. ' b. Hon-ga aru. Book-NOM be 'There is a book.' c. *Hon-ga iru.

The alternation in ( 1 ) cannot be considered a lexical selection, since both verb forms only bear the same existential meaning. This is similar to variations of the be verb in English, if we disregard the difference in type of feature involved . A large number of languages in the world employ semantic . · categorization for syntactic processes, and classifiers often make semantic categories visible in such processes.2 In this squib I will argue that Japanese is one of the classifier languages, displaying at least some of the characteristic properties of noun class systems (Aikhenvald 2000, 20-22),

(2000, for discussions on this matter in relation to and case assiEmonds gAiknment. languages.henvald (2000) for detailed, comprehensive descriptions of classifier I

Z

See

See

366)

PF

LF

Classifiers. agreement and honorifics in Japanese 223

and that Japanese honorifics can be reanalyzed as a type of agreement :, UC1i3�U on the relation between a classifier and its associated verbal domain.

Noun categorization and classifiers in Japanese lassifiers are characterized by their reference to semantic categories of the they accompany: animacy, size, specificity, individuality, etc. They induce agreement on constituents inside or outside the domains in \Wblich they occur. 3 In Japanese, numeral classifiers such as ( 1 ) are well known. San-nin-no gakusei-ga ki-ta. three-CL-GEN student-NOM come-PAST 'Three students came.' QP san-nin consists of a numeral and a classifier, whose selection jepenCls exclusively on the semantic class of the head noun modified by the . If the noun is han 'book', for instance, the classifier satu should fol low numeral. Numeral classifiers are semi-open class items including wp:roxll' mately thirty eight typical members (Aikhenvald 2000, 1 06). Other types of classifiers in Japanese are less known or hardly ij),resltigiilte,d, but certain post-nominal elements, whose categorical status is identified, can be considered candidates for classifiers. More :pe<;itic;ally, titles such as -sensei or -san as in Yamada-sensei, attached to )er.;onal names and other human nouns, are likely to be classifiers. I will iiscuss them shortly in connection with Japanese honorifics.

Honorifics in Japanese his pioneering work on honorifics in generative grammar, Harada ( 1 976) :..""........... '" that the Japanese honorific system is not merely a collection of expressions but a grammatical system. Three types of honorifics are traditionally distinguished in Japanese, as ';)illtlstnlted in (3).4

See Ogacla(2002) forwithidetnatihleedMidinsicussi otnsfraofmework. QuantifiSeeer FlalsooatMiandyagawa struct(1989). ures with . 1jhllumeral ssi fi ers m al i s :Abbreviatiofinsc verbusedmorphol here ogy,are:LV=lHT=honori ficandtitle, refers HP=honori fic prefi x, :;�in:,VM=honori i g ht verb, t o Verbal Noun the sense of Kageyama 99 1). (I

VN

224 Masaru Honda

(3) a. Yamada-sensei-ga hon-o o-yom-i-ni nat-tao Y amada-HT-NOM book-ACC HP-reading-HVM 'Mr. Yamada read a book.' b. Watasi-wa Yamadasensei-ni o-tegam-o I-TOP Yamada-HT-DAT HP-Ietter-ACC HP-sending-Lv o-okuri-sita. (PAST). 'I sent a letter to Mr. Yamada.' hon-o C. Watasi-wa kinoo yom i-mas-i-ta. 1-TOP yesterday book-ACC reading-HvM 'I read a book yesterday. ' In Harada's terminology, (3a) is called 'subject honorifics' (henceforth SH) and (3b) 'object honorifics' (henceforth OH), both of which fall under a single category, 'propositional honorifics.' (3c) is an instance of 'performative honorifics' (henceforth PH), distinct from SH and OH in that it does not require any honorified NP in argument positions.

(4) a. Ame-ga fut-ta. Rain-NOM fall-PAST 'It rained.' b. Ame-ga fur-i-mas-i-ta. C. * Ame-ga o-fur-i-ni-nat-ta.

(No honorifics)

(PH) (SH)

Syntactic configurations and morphological patterns of Japanese honorifics are summarized in (5); NPs in bold fac� represent honorified NPs.s

(5)

SH: NP-NOM . . olgo-VN-ni-nar-u OH: NP-NOM NP-DAT/ACC olgo-VN-sur-u PH: V/A-mas-Ides-u .

...

In the propositional honorifics, an NP in an argh�ent position, subject in SH and object in OH, needs to refer to a person who should be respected. 5

The

VN-sur

construction is extensively found in Japanese. For detailed

discuss ion of this type of construction, see Kageyama ( 1 99 1 ). When it is used in OH, VNs take the honorific prefixes 0- or go- depending on whether they are of Japanese origin or of Sino-Japanese origin respectively. The honorific construction has alternative verb fonns; for instance, the passive verb form -(rJare- frequently occurs as forms. See Harada ( 1 976, 504).

an

honorific verb form. For suppletive

Classifiers, agreement and honorifics in Japanese 22 5

Such an NP is normally followed by a title or the general honorific suffix ­ san.6 It is not unreasonable to assume that those honorific titles and L ')."'.....,.I1.r l1·'f' suffixes are classifiers since they denote semantic classes of the , IIU 'Ll .... they are attached to. Thus, Japanese honorific NPs have the structure (6). NP + CL where CL = Classifier ','a. JI....."",nu.:>..

, honorifics establish a sort of agreement relation between an ',arj�I1nerlt NP and a verb; more precisely the honorific form on the verb with an honorific title or a classifier in its argument position. COlnSl<1er the contrast in (7). Yamada-sensei-ga hon-o o-yom-i-ni nat-tao h. Yamda-ga tegami-o yom-da. c. *Yamada-ga tegami-o o-yom-i-ni-natta. a.

(= 4a)

(7c) shows, the honorific sentence is ungrammatical without a proper :las;sl'ller on the argument NP. However, (8) is grammatical. Yamada-sensei-ga hon-o yom-da. might argue that the grammaticality of (8) poses a problem for the that an agreement relation holds between the NP with an honorific �las:Sltlc�r and a verb. In (8) the verb does not take any honorific form, even the subject NP is followed by an honorific classifier. This, however, not problematic. ' Consider the well-known paradigm of the be verb in ·sh. I am sick.

She is sick. He is sick. d. We are sick. e. They are sick. C.

, , ' Note that not only personal names but also occupational titles can have this H�eneral personal suffix -san. Thus butyoo-san 'Mr.lMs. Director' is a natural ,��:pre:ssi(Jln in Japanese. There are many such combinations including gakusei-san .!Ms. student,' and nikuya-san 'Mr.lMs. butcher.'

226 Masaru Honda

The form of be varies in agreement with the subject. Note, however, that when a pronoun is morphologically distinctive with respect to any given feature, the form of be is not always distinctive. The pronoun they, for instance, is distinct from we with respect to person. But the form of be is neutral with respect to the feature. Table 1 shows how each feature is morphologically realized on the subject and the predicate in the paradigm in (9). Y means that the feature is morphologically realized, and N means that it is not. Table 1 PREDICATE

SUBJECT Person

Y

Number

Y

Gender

N

Person

Y

Number

Y

Gender

N

Person

y

Number

Y

Gender

y

Person

Y

Number

Y

Gender

N

Person

Y

Number

Y

Gender

Y

Person

Y

Number

y

Gender

N

Person

Y

Number

Y

Gender

N

Person

N

Number

Y

Gender

N

Person

Y

Number

y

Gender

N

Person

N

Number

y

Gender

N

Note that more agreement features are visible on the subject pronoun than on the verb be. However, there is no case in which more features are morphologically manifested on the verb than on the subject pronoun. In the case of Japanese honorifics, where semantic features enter into agreement, a similar pattern obtains. Let the features involved here be [± honorific] . Table 2 ARGUMENT H onorific

Y

PREDICATE Honorific

N

. HonQrific . ; N ··Hon.odfic , Y Honorific

Y

Honorific

y

Honorific

N

Honorific

N

Among four possible combinations of the honorific features on an argument and a predicate in Table 2, only the one in the shaded cells, in

Classifiers. agreement and honorifics in Japanese 227 ":;:'']2:",:,, '' ,...,h . .... h

the predicate manifests more features than the argument, is not . permitted, This is exactly the case in (7c). �{i . . Every honorific NP should be followed by a classifier in SH and OH, which agrees with the honorific form of a verb, As I have discussed so far, relation between an honorific classifier and a verb is considered af!l'eelment in the same sense as the subject-verb agreement in the Indo­ languages. More precisely, honorific agreement holds not heltw«:,en the head noun of an honorified NP and a verb, but between a lassifier and a verb, just like I or Agr and 0 agree in gender, person and ber. Thus, the classifier -sensei in (3a) agrees with the honorific verb o-yomi-ni-naru. Here I disregard the exact structure of the verb �'i��L90mlp l(;:x in Japanese. Most of the examples of OH which Harada cites in his paper are those :,=:;:;;�:.I" which the subject is the first person. In my own judgment, the case of with the third person NP in subject position sounds somewhat ??Hanako-ga Yamada-sensei-ni o-tegami-o o-okuri-sita. )en1teD(�e ( 1 0) would be perfectly OK if it were embedded in a statement iiTej�tlv addressed to Mr. Yamada as in ( 1 1 ). Yamada-sensei, [Hanako-ga Yamada-sensei-ni o-tegami-o o-okuri­ sita] soo-desu-yo. Yamada-CL (vocative) [ . . . J guess 'Mr. Yamada, T guess Hanako sent you a letter. ' my judgment is correct, agreement in person controls OH, which might ugJl�est that honorifics are primarily subject-verb agreement. In fact, the "......,... __... feature relates to certain other grammatical contrasts as well. Watasi-wa kore-ga tadasii to omo-i-masu. I-TOP this-NOM correct C think '1 think that this is correct.' b. *Yamada-wa kore-ga tadasii to omo-i-masu. c. Yamada-wa kore-ga tadasii to omo-tte-i-masu. The first person and the third person contrast in ( 1 2); each requires a different verb form. This shows that agreement in person is of considerable : importance in Japanese,

228 Masaru Honda

In Japanese honorifics, two types of features are involved; one is person and the other honorific. Presumably the person feature is a property on D, while the honorific feature a property on CL. Both types of features interact to yield three different honorific patterns, namely SH, OH and PH. Given the assumptions above, the following structure can be proposed for Japanese nominal phrases, in which CL as well as D heads its own 7 proJectIOn . •

•

( 1 3)

DP

I

D'

�

CLP

�

NP

D

CL

There are two types of agreement in Japanese; one involves 0, and the other CL. I will call the former D-agreement, and the latter CL-agreement. In Japanese 0 contains the person feature. D-agreement is similar to subject-verb agreement such as the one in English. CL-agreement, on the other hand, is more semantically based, referring to semantic classes to which the head ofNP belongs. 4. Consequences and further speculations

I have argued that CL as well as 0 enters into agreement processes in a language like Japanese. A number of questions remain. One such question might concern changes in honorific styles over the past few decades. Recently younger people are said to use little or no honorific speech. I have no space to pursue this issue here, but one thing that I can say is that even though they no longer maintain the old style of honorifics, Japanese people seem to use different forms of classifiers and verb morphology to preserve syntactic relations between arguments and predicates. 7

s

Detail are omitted in ( 13). Cheng and Sybesma ( 1 999) postulate a similar structure for Chinese nominal phrases in which CL is the head of its own projection.

Classifiers, agreement and honorifics in Japanese

229

It is not clear at this point whether the structure in ( 1 3) generalizes into '. nominal projections in other languages. Presumably gender, unlike person, a property on CL, not D. Languages may be parameterized as to relative ' strength of CL and D. In languages like English, D seems much more !�;nrlomm ' lent than CL.

cAikblen'vallQ. Alexandra Y. 2000. Classifiers: devices. Oxford: Oxford University Press .

A typology of noun categorization

. '£<'1""'",1T

Lisa Lai-Shen, and Rint Sybesma. 1999. Bare and not-so-bare nouns and the structure ofNP. Linguistic Inquiry 30: 509-542. Joseph.

2000.

Lexicon and grammar: The English syntacticon.

BerlinlNew York: Mouton de Gruyter.

g;;!:" 'lHru'ad:a, S. I. 1 976. Honorifics. In Japanese generative grammar, ed. Masayoshi Shibatani, 499-56 1 . New York: Academic Press. Taro. l 99 1 . Light verb constructions and the syntax-morphology interface. In Current English linguistics in Japan, ed. Heizo Nakajima, 1 69203. BerlinlNew York: Mouton de Gruyter. iliv:ae:awa.. Shigeru. 1 989. Structure and case marking in Japanese: Syntax and semantics 22. New York: Academic Press. Kyoko. 2002. The syntax of Japanese nominal projections and some cross­ linguistic implications. Doctoral dissertation, University of Durham.

What stranded adjectives reveal about Split-NP Topicalization Hanneke van Hoof

1.

Introduction

Split-NP Topicalization (SNPT) is an A-bar-displacement construction which splits argument and predicate NPs. SNPT in Gennan, whose syntactic properties have been resourcefully described in Van Riemsdijk ( 1 989), is illustrated in ( 1 ): (1)

[Autos] kann ich mir [keine neuen amerkanischen -] cars can I me no new American (ones) leisten. afford , As for cars, I cannot afford new American ones.'

(G)

In ( 1 ), the direct object has been split. The topicalized split NP is referred to as TOP. The split NP which remains in canonical position is referred to as REM. TOP represents a N' core of the unsplit NP, which is lacking in REM; the gap in REM is indicated by a dash '-' . In this squib, I the empirical domain is restricted to German (G) and Brabant Dutch (BrD) ? As shown in ( 1 ), more than one adjective (A) can appear in the overt part of REM. But sirtce only As that are left-adjacent to the gap in REM (cf., e.g. amerikanischen 'american' in ( 1 » are relevant for the discussion here, the notion 'stranded adjective' applies only to these adjectives. Section 2 hypothesizes that stranded As should be of a particular

SFB

nn

l

1 This squib builds upon resear�h I did at the university ·�f TUbi ge in the �ummer and fall of 2002. Thanks are due to the 44 1 Linguistische Datenstrukturen in Tubingen for providing facilities and a stimulating environment to work in. I a so thank you, Henk, for being a constant source of linguistic inspiration. 2 See Van Hoof (1997) for outline of the syntactic properties of SNPT and Split­ NP Contrastive Dislocation in Brabant variety of Dutch (D). This variety of Dutch, which allows for these split constructions, is spoken in the province of Northern Brabant. Although this squib does not present any examples of Split Contrastive Left Dislocation, the generalization that will be a g ed for here holds for this construction in Brabant Dutch, too.

ru

Left an

Dutch (BrD), a regional

.,

Split-NP topicalization 23 1

type and presents a handful examples supporting this hypothesis. Section 3 show s that this hypothesis applies to stranded As in NP-ellipsis ::construction s, too. Section 4 describes the semantic condition that has to be fulfi lled by stranded As in SNPT and NP ellipsis. Section 5 notes some p ossible theoretical consequences of this outcome for the syntactic analysis of SNPT. 2.

Only classifying adjectives can be stranded

:�There are quite some lexical-semantic classifications for As around (cf. Vendler 1 967; Bolinger 1 967; Kamp 1 975 ; Broekhuis 1 999. a.o.). The meaning of an A appears to be highly dep endent on the semantics of the noun it modifies, as well as on the pragmatic context and the creativity of the language user. However, in spite of all these factors and different ;0,���':;:;ctpmantic classifications, I would like to pose the following generalization: classifying adjectives can be stranded in REM. The distinctive feature the majority of c las s ify in g As is that besides their attributive use, they allow for a predicative use with the same meaning. But besides nredlcatn ' Ie adjectives, classifying adjectives also include As that can be ch,aralctelnz��d as 'restrictors ' in the sense of Broekhuis ( 1 999). Some As, such as awake, only have a predicative use, while some such as former and alleged, can only be used attributively. The icative use of an attributive A can be tested by using it as a §�;��onlplc;:mlent of a copula, as a secondary predicate, or as a postnominal moc11tlc�r. Although most attributive adjectives allow for a predicative somehow, the avai lability of the predicative reading always ::
232 Hanneke van Hoof

[een paar syntactische -] c. *[Atlassen] werden er atlases were there a couple of syntactic (ones) uitgegeven. published , As for atlases, some syntactic ones were published.'

(BrD)

However, if the relational adjective in (2) is focused, the grammaticality of (2b,c) improves considerably, as shown in (3): 3

(3) a. JDEZE atlassen zijn \SYNTACTISCH. (BrD) 'THESE atlases are SYNTACTIC.' niet aIleen b. [Atlassen] maken ze op het Meertens atlases make they at the Meertens institute not only [/MORFOLOGISCHE -], maar ook [\SYNTACTISCHE -] morphological (ones) but also syntactic (ones) 'As for atlases, they do not make only MORPHOLOGICAL ones, but also SYNTACTIC ones at the Meertens institute.' •

Sometimes, a relational A can be reinterpreted as a predicative A quite easily. For example, the examples in (4) show that the relational A zondags 'of Sunday' can denote a property if it is combined with the noun kleren 'clothes' (cf. (4a» , but cannot denote a property if it is combined with the noun krant 'newspaper' (cf. (4b» . The original interpretation of the NP zondagse kleren is 'a kind of clothes that is worn only on Sundays', but this interpretation has changed into 'clothes that used to be worn on Sundays'. In the latter interpretation, the A also allows for a predicative use (cf. (4a» . However, a similar shift in meaning ' does not occur in ' zondagse krant ' newspaper of Sunday'. This is probably due to a lack of extra-linguistic properties in today's context by means of which Sunday newspapers can be distinguished from newspapers appearing on other days of. the week, . like for instance 'tabloid' versus 'broad sheet' format. The contrast between the SNPTs in (5), which cann�t be cancelled with the aid : bf of fo�us, is confirmed by the predication test in (4).

Focused words are written in and '\', a falling pitch accent.

3

SMALL

CAPITALS; 'r indicates a

rising pitch accent

Split-NP topicalization 233

a. De kleren waren zondags. the clothes were Sunday-like b. * IDlE krant was \ZONDAGS. that newspaper was Sunday-like

(D/BrD) (D/BrD)

(BrD) [Kleren] draagt hij (aileen maar) [zondagse -] . clothes wears he only but Sunday-like (ones) ' As for clothes, he wears (only) ones that are sunday-like. ' (BrD) b. *[/KRANTEN] leest hij aileen maar [\zONDAGSE -] . newspapers reads he only but Sunday (ones) ' As for NEWSPAPERS, he reads only ones that are SUNDAY-like'

a.

1(, J�d�/erlb-rlelated adjectives attribute a property to an event or situation the noun indirectly refers to. Therefore, they often occur with verb-derived . houns. For example schnelle 'fast' in (6a) does not denote an inherent �.-nlnp.rlhl of the noun, but a property of the verbal action that is associated it, i.e. verlaufen 'to proceed'. Although these As initially resist a rirCjc11c:au',re use, they often can get a 'derived' predicative reading if they focused, as illustrated in (6). 4 A parallel result holds for the SNPTs in a. Eine schnelle Hochrechnung. 'A fast expansion.' b. *Die Hochrechnung war schnell. 'The expansion was fast.' c. IDiESE Hochrechnung war \SCHNELL. 'THIS expansion was FAST. '

(G) (G) (G )

a. *[Hochrechnungen] hatte das Programm rein paar schnelle -] a couple of fast ( ones) expansions had the program (G) durchgeftihrt. carried out ' As for expansions, the program had carried out a couple of fast ones.'

ot 'That the predicative interpretation must somehow be derived becomes clear from the fact that adverb-related adjectives cannot occur in postnominal position.

234 Hanneke van Hoof

h.

[/HOCHRECHNUNGEN]

hatte das Programm [ein paar \SCHNELLE had the program a couple of fast (ones) hei def Darstellung des IENDERGEBNISSES at the presentation of the final result (G)

expansions durchgefiihrt (aber carried out but stUrzte es \AB). crashed it down 'As for EXPANSIONS, the program had carried out FAST ones ... (but when it had to present the FINAL RESULT, it crashed DOWN).'

Finally, it appears that also restrictor adjectives (cf. Broekhuis 1 999:52t) can be stranded in SNPT. Adjectives in this class restrict the reference of the noun in a direct, almost deictic way. The class of restrictors is special because these As can be stranded in REM but cannot be used predicatively, as shown by the example in (8):

(D/BrD) Bepaalde mannen. 'Particular men. ' ( D/BrD ) b. */OEZE mannen zijn \BEPAALD. 'THESE men are PARTICULAR. ' c. [Mannen] kunnen haar [bepaalde -] wei bekoren. (B rD) men can her particular (ones) PRT please 'As for men, particular ones can please her though. '

(8) a.

As shown by the examples in (9) and ( 1 0), also intensional adjectives can 5 only be stranded if they allow for a pfedicative interpretation: (9) a.

Een nagemaakte euro. 'An imitated euro.' b. De euro was nagemaakt. 'The euro was imitated. ' c. [Finse euro' s] verzamelt-ie [nagemaakte -]. Finnish euros collects-he imitated (ones) 'As for Finnish euros, he collects imitated ones.'

5

(D/BrD) (DlBrD) (BrD) .

Partee (200 1 ) observes independently that only predicative As can be stranded in Polish split NPs. In this paper, Partee argues in favour of a revision of the traditional ideas about the semantics of intensional privative adj ectives (cf. Kamp 1 975) and claims that privative adjectives (cf., e.g. nagemaakt ' im itated ' in (9)) have to be subsective .

Split-NP topicalization 235

(l O) a. Ein angeblicher Baron. 'An alleged baron. ' b. */OIESER Baron ist \ANGEBLlCH. 'THIS baron is ALLEGED .' c. * [/BARONE] mochte sie nur [\ANGEBLICHE -] heiraten. barons wants she only alleged (ones) marry 'As for BARONS, she wants to marry only ALLEGED ones.' 3.

(G)

(G) (G)

Stranded adjectives in NP-ellipsis constructions

In , the preceding section, it has been argued that only classifying As can in REM and that this class of As consists of predicative As and '::." ,flestr'lct()r As. It has been shown that some As wh ich do not denote an ;'�Jiherent property of the head noun can still gain a derived predicative ,jnterpretation if they are focused. It appears that exactly the same :generalization can be made for stranded adjectives in NP-ellipsis :�;c�[)ns�tructJC')ns in German and Brabant Dutch, too. So, for example, the N' in (l l a) is ungrammatical, and so is REM in ( l Oc); the elided N' ' ,(l lb) can for instance be interpreted as broek 'trousers' but not as krant ,,'neWSJ:lape:r' (cf. the REMs in (5a,b» . , ) a. *Sie mochte nur [NP ANGEBLICHE ] heiraten. 'She wants to marry only alleged (ones).' b. Hij had aIleen [NP een ZONDAGSE ] meegenomen. 'He had taken along only a Sunday-like (one).' c. Er kunnen haar [NP bepaalde -] weI bekoren. 'Particular (ones) can please her though.' -

(G)

-

(D/BrD) (D/SrD)

beck ( 1 995) and Kester ( 1 996) argue that the elided site in German, and English NP-ellipsis is the empty pronominal predicate pro, must be syntactically licensed by overt agreement morphology on a 'ferrmalllt element left-adjacent to pro. I also want to draw attention to the study of Broekhuis and Strang about the semantic nature of the adjectival complement in the '\P.8rtiltive-{!reniti\,e construction in Dutch. They conclude that the adjectival ;:qomp.lemlent of this construction (cf. (12» has to be compatible with a predlc:ati\'e use. '�:��·nn.r'

[vreselijk-s]] [NP [iets] something terrible-GEN

(D)

236 Hanneke van Hoof

Broekhuis and Strang ( 1 996) do not consider the class of restrictor adjectives. But in the context of the discussion here, it is interesting to see that restrictors can appear in the complement position of this construction, too, as shown in ( 1 3): ( 1 3 ) a.

[NP [iets] something b. LiP (iets] something

[ dergelijk-s]] such-like-GEN [bepaald-s]] particular-GEN

(D)

(0)

The conclusion drawn by Broekhuis and Strang in combination with my observation in ( 1 3) corroborate the generalization that has been posed in the previous section, which assumes restrictor and predicative As to build a common class.6 This result is consistent with Kester's analysis for the Partitive-genitive construction (cf. Kester 1 996:208), in which the adjective with the genitive suffix c-commands an elided noun, which she analyzes as pro, as indicated in ( 1 4): ( 1 4)

[N' [leuk-s] [NP [iets] something nice-GEN

(pro]] ]

(D)

Since the second member of the Partitive genitive is assumed to have the same structure as the simple NP-ellipsis construction, the observation that the second part of this construction has to be a predicative or restrictor A follows as a matter of fact. The next section attempts to answer the question why only classifying . As can strand in SNPT and NP ellipsis. . 4. A semantic licensing condition on the empty N'

At first glance, it may seem somewhat surprising -that the observation that · adjectives cannot always be stranded in SNPT and NP ellipsis has gone unnoticed for such a long time. Lobeck ( 1 995) and Kester ( 1 996) only concentrated on the syntactic licensing conditions of the empty N' in NP ellipsis. Also most studies on SNPT seem to have assumed that apart from the ordering restrictions among the As in TOP and REM, the distribution of As in split NPs is unconstrained. On the other hand, it is not so much 6

The generalization that only predicative As and restrictor As can appear as a second member of the Partitive-genitive construction can also be made for German and Brabant Dutch.

Split-NP topicalization 237

surprising because there are hardly any As that can be stranded in no A-N combination at all . In addition, many As that cannot strand at first go can still become well-formed stranders with the aid of focus. This does not imply that the generalization that only classifying As can strand is an empty one, but this situation requires a precise understanding of what makes an A . ,.,"' .... . '", n 'classifying' . Since it has been shown that stranded As have to be morphologically jpf1ected in order to license the empty N' in NP-ellipsis (cf. Lobeck 1 995; l.<.ester 1 996) as well as the empty N' in the REM of SNPT (cf. Fanselow the ' classifying' relation between the stranded A and the empty N ' serves the purpose of licensing the empty N' in both too. S ince predicative As and restrictor As do not constitute a tlcms, j���·.;&onstrU(� ±'bflLomlogeneous syntactic class,7 it is highly improbable that classifying As a role in the syntactic licensing of empty N 's. Therefore, the hYIPotJl1esl' s that classifying As function as semantic licensers of empty N's ; sef�ms to be a viable option. So, why can only classifying As semantically license an empty N'? My lVpc)thc�SlS is that only classifying As are able to select a subset of the kind :ete:rrf�d to by the empty N ' . This semantic property of stranded As is i.rot)ablly helpful in finding the antecedent for the empty N' , Since the ....n."aT1h., denoted by a predicative A can always be intersected with the ft"i'\"'P�' denoted by N', predicative As select a subset of N', Restrictor As a subset of N' in a deictic way. Therefore, I propose the following, lrellmllnalry semantic-licensing condition on empty N, :8

restrinictcommon or As restriwictththdete ereference ofquanti the NPfiersin athdianrectwitway, theycatiseem to more rmi n ers or h predi v e As. it;urltlennore, resto beriphrctorsasalcan. Finharaldlyly, restri be modi fihave ed, iton contrast wid atththeprediuppercativparte As,of can al s c t o rs be l o cat e adjecti valgllyorderi nggeneral hierarchy, inn contrast wizthedprediherecatiaccords ve As. with Sleeman's Jnt:e r e:s t in ' , t h e i z ati o hypot h esi .bsc::�rvaltsemanti iOJ1S aboutc natureeloflipthesis adjecti in Frenchve is(cf.of Slcrucieemanal importanceSleforemantheconcl udesng l i c ensi �:'elid(�N'd N's ins iFrench, n French.is Howeve�, the clcatedss ofthanAs iwhin German ch can semanti cally liDutch, cense far more restri and Brabant .J:Predi cati�"eand Shihonly be cwhisicfyh iimg plifietheys thataretheycognihavetiveltoy relbeevantveryinlothew isense of-Sproat n the adjectival order hierarchy. •

1 996).

NP

As can

( 1 988),

las

n

238 Hanneke van Hoof

( 1 5)

Semantic-licensing condition on stranded adjectives: A stranded A must be able to select a subset of the property denoted by the empty N' in order to facilitate the identification of the antecedent for the empty N'.

What is the role of focus in making an A classifying? Apparently, focus helps to construe a predicative reading for As that do not have one in the default reading of the A-N combination. For example, although syntactisch 'syntactic' in (2) does not denote a default property of atlases, it can still become interpreted as a property of atlases if this A is contrasted with a comparable, but different A (cf. (3» . However, focus cannot have the same effect in (5b) because of pragmatic reasons. Adverb-related As (cf. (6») modify the event or situation in which N fulfills a semantic role. How can these modifiers become classifiers ofN if they are focused? The answer to this question seems to require a theory that allows the meaning of common nouns to be flexible to a certain extent. For example, the meaning of deverbal nouns, as in (6) and (7), could be assumed to be associated both with a type structure and with an event structure at the same time (cf. Larson 1 995 for similar ideas). If a common noun does not project its type structure, then the event structure can become relevant for classification and As can restrict the common noun's kind indirectly over its event structure.9 Note that if this i dea about the meaning representation of common nouns is on the right track, then focus on a stranded A can probably also force the empty N' to adapt its kind meaning slightly, so that A can denote a subset of it. We can also understand now why so many ungrammatical examples of SNPT can be saved by the insertion of a focus particle . with scope over REM . :

Drubigwi(1997) sketblechesmeania vinegwrepresent on the semanti c properti esnouns: of kinds"Many that seems ttos accord t h flexi a ti o n for common l i n gui s have pOiives:ntedwhereas out thatadjecti therevesis denote basicsidenotati onal edis,fference betweennatural nounskinandds. adject n gl e properti nouns denote associatedshare,withupa ltoargea certandainotn amount clearlyofdelvariimiatteidon,number ofmeani propertngieofs whitheKindsckihncantheid-denoti rbemembers but the ng tenn cannot be reduced to those properties (cf. Drubig 1997: , ',

9

48)".

a

a

Split-NP topicalization 239 5.

Possible consequences for the analysis of SNPT

The observation that stranded As in SNPT have the same morphological properties as in NP ellipsis goes back to Fanselow ( 1 988). This observation reappears in the relevant literature every now and then and is claimed to be valid for quite some languages. Of course, this observation gains even more importance by the observation that stranded As in NP ellipsis and in the REM of SNPT have common semantic properties, too. Meanwhile, the ,generalization that stranded As have to be predicative has also been noticed for other languages, namely for split NPs in Polish (cf. Partee 200 1 ), split NPs in Romance (cf. Mathieu 2004) and split NPs in Modem Greek (cf. Ntelitheos 2004). So, also the semantic part of the similarities between ' REM in SNPT and NP ellipsis enjoys cross-linguistic support. Hence, the question arises: How should this generalization be expressed by the analysis of SNPT? Under the assumption of the pro analysis for NP ellipsis . Lobeck 1995; Kester 1996), one can adopt a syntactic analysis of SNPT , which is base generated as an independent elliptic NP ( cf. low ( 1 988) and Van Hoof (2005) for an overview of different ' )fiS of this idea in the literature). A less committing option, im lplememtl:ttl( assumes the similarities between and NP ellipsis to be more l!plnCJldentall, could be a subphrasal-movement analysis which leaves the category pro. A new, refreshing analysis, wh ich takes the common n..r'n""rtll:·� of NP ellipsis and split NPs in Modern Greek as a starting point, be found in Ntel itheos (2004). He rejects the pro analysis for NP is and argues that NP ellipsis should be analyzed in terms of 'copy delete' of partial NPs. Therefore, for the time being, the problem of g;tli:ndilng an appropriate theory expressing the generalization above seems to reduced to the chicken or egg problem.

REM

,

REM

�ollmgler, Adjectives in English: Attribution and predication. Br�()eDutch khuils. occasi Hans.onal papers Tilburg University. Modem Grammar of Broekhuis. Hans, and Anke De partitieve genitiefconstructie. 'Drubig, Hans Bernhard. of TObingen.Arbeitspapiere des Sonderforschungsbereichs University .

.

Dwight. 1967. Lingua 1 8: 1 -34. 1 999. Adjectives and adjective phrases. 2, Strang. 1 996. Nederlandse Taalkunde 3 : 22 1 -238. 1 997. Some cross-categorial generalizations of focus structure. 340 1 09,

240 Hanneke van Hoof

Fanselow, Gisbert. 1 988. Aufspaltung von NPn und das Problem der 'freien' Wortstellung. Linguistische Berichte 1 1 4: 9 1 - 1 1 3 . Hoof, Hanneke van. 1997. Left Dislocation and split topics. In Materials on Left Dislocation, ed. by Elena Anagnostopoulou, Henk van Riemsdijk, and Frans Zwarts, 275-305. Amsterdam: John Benjamins. Hoof, Hanneke van. 2005 . Split Top icali zation. In The Blackwell Companion to Syntax, ed. by Martin Everaert and Henk van Riemsdijk, chap. 62. Oxford: Blackwell. Kamp, Johannes A. W. 1975. Two theories about adjectives. In Formal Semantics for Natural Language, ed. by Edward L. Keenan, 123- 1 55 . Cambridge: Cambridge University Press. Kester, Ellen Petra. 1 996. The nature of adjectival inflection. Doctoral dissertation, University of Utrecht. Larson, Richard. 1 995. Olga is a beautiful dancer. Ms., LSA, New Orleans. Lobeck, Anne. 1 995 . Ellipsis: Functional heads. licensing, and identification. Oxford: Oxford University Press. Mathieu, Eric. 2004. Split scrambling in Romance. In Current studies in comparative Romance linguistics, ed. by Martine Coene, Gretel de Cuyper, and Yves D'Hulst, Antwerp: University of Antwerp. Ntelitheos, Dimitrios. 2004. Syntax of elliptical and discontinuous nomina/so Master's thesis, University of California, Los Angeles. Partee, Barabara H. 200 1 . Privative adjectives: Subsective plus coercion. Ms., University of Massachusetts, Amherst. Riemsdijk, Henk van. 1 989. Movement and regeneration. In Dialectal variation and the theory ofgrammar, ed. by Paola Beninca, 1 05- 1 36. Dordrecht: Foris. Sleeman, Petra. 1 996. Licen..ving empty nouns in French. Doctoral dissertation, University of Amsterdam. Sproat, Richard, and Chitin Shih. 1 988. Prenominal adjectival ordering in English and Mandarin. In NELS 1 8, ed. by James Blevins, 465-488. Amherst: University of Massachusetts, GLSA. Vendler, Zeno. 1 967. The grammar of goodness. In Linguistics in Philosophy, by Zeno Vendler, ch. 7, Ithaca: Cornell University Press.

'Past tense interpretations in Dutch

is a well-established fact in the aspect literature (Boogaart 1999; lebrandse 2000; Verkuyl 2005) that the S imple Past in Dutch is neutral w th < i respect to perfective and imperfective aspect: when combined with . predicates it allows both readings. The Dutch Simple Past thus does not have the same semantics as the English Simple Past, which only allows readings. Instead, it functions more like certain Romance ..... ..� r1·.,. � t tenses (French Imparj ait, Italian Impel:fetto). This squib presents mal-semantic analysis of the Dutch S imple Past that explains its dual ii'for . nature as the interaction of its essentially imperfective semantics additional pragmatic effects that may yield perfectivity (section 2). The ::' SlflISlI" Sl' S predicts the pattern of results of an aspectual comprehension task , =�,'::;'. three past tenses with Dutch adults (section 3) and is superior to an .. ;,�;;:�i1'::altc�rn:ati·'Ie analysis that would claim that this tense is ambiguous between possible readings.

Analyzing three Dutch past tenses D UtCh Present Perfect (tegenwoordig voltooide tijd), Simple Past �:' ;�� d�� �� � verleden tijd) and Periphrastic Progressive (aan-het :; V;.�j� :d ?l!:ClnnstnJ'�tiflln ) may all refer to past-time situations. Whereas the Perfect and !. ���il�rol!fe!)si,'e function aspectually very similarly as in English (but see De '':is'l��rt 2000 on some differences between the Dutch and English Perfect), Dutch Simple Past does not behave like its Engl ish counterpart. �>;:;i��i; , ,lj.oo;gaaLJt ( 1 999) argues that for telic eventuality descriptions of type event, �: :S';:"i:0:'�:::tI,p, : Dutch S imple Past allows for a perfective as weII as an imperfective reading, unlike the English Simple Past which always yields a perfective reading for events (see also De Swart 2000; Verkuyl 2005). In narrative a Simple Past on event predicates may yield a perfective or . :'imperfective interpretation; in non-narrative contexts it basically has an . . • imperfective reading. The perfective reading is illustrated with the following example from Boogaart ( 1 999:26). The Dutch Simple Past of :

.

'

242 Angeliek van Hout

draaide 'turned' in the short narrative sequence in ( 1 ) must be translated with a Simple Past of English turned. Voordat ik het huis verliet, deed ik aIle gordijnen dicht before I the house left did I all curtains closed en sloot aile deuren, maar ik draaide het gas niet uit. but I turned the stove not out and closed all doors ' Before I left the house, J closed all the curtains and locked all the doors, but I didn 't tum off the stove. '

(1)

The continuation test in (2) illustrates an imperfective interpretation of a Simple Past combined with an event predicate. The English Simple Past wrote lacks this interpretation; instead, the appropriate translation of the Dutch Simple Past form schreef (literally: ' wrote') is a past progressive in English, was writing (examples from Boogaart 1 999:73).

Marie schreef een brief toen ik haar net zag en Mary wrote a letter when I her just saw and is er misschien nog steeds mee bezig. ze she is there possibly even still at busy #'Mary wrote a letter when I just saw her and she may still be working at it. ' 'Mary was writing a letter when I just saw her and she may still be working at it . '

(2)

The other two Dutch past tenses each have only one meaning: The Present Perfect on an event description yields a perfective entailment and does not pass the continuation test, (3), indicated by the #-sign, whereas the Periphrastic Progressive is imperfective: it strips the event of its culmination mOment and therefore passes the contin1,lation test, (4).

(3)

#I

Marie heeft een brief geschreven en ze is er . misschien and she is there possibly Mary has a letter written nog steeds mee bezig. at busy even still # 'Mary has written a letter and she may still be working at it.'

Part tense interpretations 243

(4)

Marie was een brief aan het schrijven toen ik haar net Mary was a letter at the write when I her just zag en ze is er misschien nog steeds mee bezig. busy even still at saw and she is there possibly , Mary was writing a letter when I just saw her and she may still be working at it.'

Another test for perfective and imperfective readings is found in simultaneity effects in discourse contexts, or lack thereof. The Dutch Simple Past combined with an event predicate in (Sa) is ambiguous, allowing both a simultaneous reading (letter-writing was going on at the moment of John's entering) and a sequential reading (letter-writing started after John 's entering), witness the two possible translations in English (Past Progressive and Simple Past, respectively). The Present Perfect has only a perfective reading, (5b), and the Periphrastic Progressive only an imperfective reading, (Sc). 1 (5) a. John kwam binnen. Marie schreef een brief John came inside. Mary wrote a letter ' John entered. Mary wrote/was writing a letter.' h. John kwam binnen. Marie heeft een brief geschreven. lohn came inside. Mary has a letter written 'lohn entered. Mary has written a letter.' binnen. Marie was een brief aan het schrijven. c. John kwam John came inside. Mary was a letter at the write 'John entered. Mary was writing a letter. ' So, a telic predicate combined with Simple Past in Dutch is optionally . perfective or imperfective, whereas the English Simple Past on such a predicate is necessarily perfective. Boogaart ( 1 999) attributes this difference to the fact that English has. a dedicated form for imperfective past, i.e., the Past Progressive. The presence of alternative forms within a language affects the exact set of interpretations for each. The English Simple Past is the unmarked form; it does not allow for imperfective readings because English has another, marked form to express imperfective past meanings. The Dutch S imple Past, on the other hand, is ambiguous,

The sameDutdichffversus erencesEnglbetween Dutch Simplaree Pastobserved versusin Present Perfect , and i s h Si m pl e Past between Sequence of Tense contexts (Boogaart 1999; Hollebrandse 2000). I

244 Angeliek van Hout

according to Boogaart. In addition to perfective readings, imperfective interpretations are an option for lack of an alternative form that claims that reading (apparently, on this explanation, the Dutch Periphrastic Progressive has not grammaticized to the same extent as the English Progressive). In De Swart's (1 998) model of the interaction between a predicate's lexical aspect and the grammatical aspect contributed by tense or aspect, both kinds of aspect make reference to the same kinds of ontological entities: state, process and event. States are non-dynamic and homogeneous; processes are dynamic and homogeneous; a�d events are dynamic and quantized (Bach 1 986; Krifka 1 992). De Swart assumes that the verb and its arguments describe eventuality descriptions at the level of VP (Verkuyl 1993; Krifka 1992). Aspectual operators such as progressive or perfect aspect sit in Asp or T heads and function as eventuality modifiers which " ... map sets of eventualities (of a certain type) onto sets of eventualities (of some possibly other type)" (De Swart 1 998:349). Aspectual operators may thus trigger an aspectual shift from one kind of eventuality type to another. Alternatively, certain tenses may be aspectually sensitive, which means that they require a particular type of eventuality as their input. If the input eventuality does not fit this requirement, an aspectual shift into the appropriate type is coerced before tense can apply. In the subsequent discussion I focus on the effects of the three Dutch past tenses on telic eventuality descriptions, i.e., those that denote events. De Swart analyzes the English Progressive as an aspectual operator which maps an event or a process onto the state of that event or process going on. For a telic predicate combined with a progressive, the operator thus strips the event of its culmination point, so that the sentence no longer entails completion. I believe that the Dutch Periphrastic Progressive functions in exactly the same way. As for the Dutch and English Present Perfect, De Swart (2000) proposes that both involve an aspectual operator which requires a quantized eventuality of type event as its input and gives as its output a state which is the consequent state of the input event, i.e., the state that holds right after the event is over (Moens 1 987; Moens and Steedman 1 988). The Present Perfect thus entails completion of an event. The English Simple Past is a pure Tense operator according to De Swart ( 1 998); it anchors the eventuality input in the past, conserving its original eventuality type. The English Simple Past is thus ��aspectually transparent in the sense of letting the lexical aspect 'shine through'" (De Swart 2000). The combination of Dutch Simple Past with an event is more flexible, however, as it allows for perfective and imperfective readings . Therefore, it cannot be a pure tense operator (unlike the English Simple

Past tense interpretations

245

Past). On the other hand, since the Dutch Simple Past does not trigger one unified aspectual effect across all eventuality types, it is not an aspectual operator (unlike the Progressive or Perfect). Instead I believe that the Dutch Simple Past functions like the French Imparfait, which De Swart ( 1 998) analyzes as an aspectually sensitive tense: it requires a homogeneous type as its input. States and processes are homogeneous, and so the Simple Past simply preserves their homogeneous nature when locating them in the past. For events, on the other hand, aspectual coercion must first create a homogeneous type. A Simple Past on an event predicate thus coerces the state of that event in progress: an imperfective reading.2 This proposal does yet not suffice, since an event predicate with Simple Past may also have a perfective interpretation. I argue · that the option of a perfective interpretation is a pragmatic effect, not a semantic one. Semantically the Dutch Simple Past requires homogeneous eventualities and anchors them in the past. Its semantics thus does not entail completion. But a conversational implicature may give rise to a completion interpretation. When an event in progress was happening in the past, it may Just as well have progressed to its culmination moment and be completed; . the semantics does not rule this out. This implicature only arises with event descriptions in the Simple Past, not with states or processes, whose . ;eventuality descriptions do not include a culmination moment. In a way, the coercion effect is undone and the underlying telic eventuality type comes back. So the semantics of the Dutch Simple Past coerces an imperfective reading when this tense combines with an event description, and a pragmatic implicature explains the perfective reading as an alternative interpretation. This analysis makes a strong prediction: al l else being equal ·Dutch native speakers will prefer an imperfective reading over a perfective one; they will allow a perfective reading only if there is no imperfective option. The alternative analysis on which the Simple Past is ambiguous predicts that both interpretations are allowed in all contexts, with no clear preferences. The experiment described in section 3 tests these predictions.

imAlperftemati ectivve.ely the Dutch Simple Past carries habitual aspect, which also

2

is

246 Ange/iek van Houl 3. Interpreting Dutch past tenses

I tested 15 native adult speakers of Dutch, undergraduates at the University of Groningen. They were tested in a group and wrote down their answers to the questions on answer sheets. The experiment was based on short stories each accompanied by a series of two pictures plus a final empty slot for which subjects had to select the right picture. The picture book page of the castle-building story is shown in Figure 1 . One story goes as follows: One day Mickey Mouse was on the beach. He was playing in the sand. What would he build? The curtains closed, so we couldn't see any further what happened. Let' s ask the giraffe to look behind the doors. Giraffe, what did you see there?

Figure 1 : Picture book for castle-building story

-

The giraffe, a hand puppet who was introduced earlier, could see what happened behind the curtains because he had such a long neck.3 The giraffe spoke the test sentence which contained one of the three past tenses: Present Perfect, Simple Past or Periphrastic Progressive, (6). .

The adults whose data are presented here, served as a control group in a more comprehensive study that also tested Dutch 2 and 3-year-olds' interpretation the three past tenses, and compared them to how Polish children and adults interpreted perfective and imperfective aspect. I refer the reader to Van Hout (2005) for a complete presentation of the Polish data, Van Hout (in prep.) for the Dutch data and Van Hout (submitted) for a comparison of Dutch, Polish and Italian.

3

Past tense interpretations 247

Mickey Mickey b. Mickey M ickey c. Mickey Mickey

( 6) a.

Present Perfect heeft een kasteel gebouwd bu ildparticiple has a castle bouwde Simple Past een kasteel bu iltsimplePast a castle was een kasteel aan het bouwen Periphr. Progr. the buildlnfinifive was a castle at

Then the experimenter showed two pictures and asked if the right picture was there. Subjects could choose one picture, or they might say that the right picture was not there or that both qualified. 1 presented different kinds of situations (completed, incomplete and ongoing) in three conditions. The conditions varied the picture pair combinations: (i) completed/ongoing; ii) completed/incomplete and iii) ongoing/incomplete. The completed picture showed M ickey standing next to a finished castle, not working at it anymore. The ongoing picture showed Mickey in the process of working at a sand building, the castle was clearly not yet finished. The incomplete picture showed a half-finished castle and Mickey walking away from it, having stopped his building activities. The three picture choices for the castle-building story are shown in Figure 2.

Figure 2: Conditions of picture combinations for castle-building story Completed situation

Incomplete situation

Ongoing situation

A --

�., . , --;.

Picture pair conditions: (i) completed/ongoing, (ii) completed/incomplete, (iii) ongoing/incomplete All VP predicates were telic and transitive, consisting of a verb and a quantized (singular) count noun, for example, build a castle, draw ajlower, eat a fish. Telicity was crucial in this experiment, because it compares the aspectual contributions of d.ifferent tenses on events--eventualities with an inherent culmination moment. The aspectual hiteractions of the three tenses under investigation with telic predicates show up as different entailments: the perfect will introduce a consequent state to the event, thus entailing

248 Angeliek van Houl

completion, the other two tenses lack this entailment, as argued in section 2. The design thus included two factors: tense (3 levels) and picture pair combination (3 levels), yielding 9 conditions. For each condition there were 3 items, yielding a total of 27 items. The test was taken in three sessions, one for each tense. In addition to the test items there were 3 control items for which the right picture was not among the two. These were used to ascertain that subjects were paying attention and were able to say that neither picture qualified. Before the experimental items were presented, there was a training session of 4 items during which subjects were made fam iliar with the procedure and the task. The three bar graphs in Figure 3 present the results of each tense for each of the three conditions. The bars specify for each condition how often subjects chose which situation, or said that neither or both pictures qualified.

Figure 3: Results Dutch past tense interpretation for each picture pair combination Present Perfect 1 00% 90% 80% 70% 60% 50% 40% 30% 20% 1 0% 0%

r-���----�

El both

• neither

m incomplete ' lB ongoing

. ��--�---

com/ong

com/ir:tc .

onglinc ·. ·

C completed

�

Past tense interpretations 249 Simple Past 1 00% 80%

• both

• neither

60%

e incomplete

40%

B ongoing

o completed

20% 0%

com/ong

com/inc

�ng/inc

Periphrastic Progressive 100%

I :

80%

6 both

60%

• neither

40%

m ongoing

iii incomplete

o completed

20% 0% ��--��----�--� com/ong

com/inc

�ng/inc

The results show a clear pattern. With the Present Perfect subjects only allow the completed situation and with the Periphrastic Progressive only : the ongoing situation; if the required situation was not an option, they said ,'i��§�iij? . that neither picture qualified. With the Simple Past, they accepted both 'completed and ongoing situations. These choices depended on the picture pair. When the ongoing situation was one of the two pictures in the pair (in completed/ongoing and ongoing/incomplete conditions), they chose it. But when it was not one of the two presented pictures (as in the completed/incomplete condition), they opted for the completed situation.

250 Ange/iek van Hout

This skewed pattern in the results supports an analysis on which the Simple Past is essentially imperfective, and may alternatively get a perfective reading if imperfective is not an option. It does not support an analysis that claims that the Simple Past is essentially ambiguous. The latter would predict general acceptance of both ongoing and completed situations independent of which picture pair options was shown. 4. Conclusion

The goal of this paper was to present a semantic analysis of the three Dutch past tenses (Present Perfect, Simple Past and Periphrastic Progressive) that explains the unambiguously perfective interpretation of the Perfect and imperfective interpretation of the Progressive versus the seemingly ambiguous nature of the Simple Past. This perfective/imperfective flexibility is described in the aspectual literature (Boogaart 1 999; Hollebrandse 2000; De Swart 2000; Verkuyl 2005), but has not received a formal analysis so far. I propose that the Dutch S imple Past is an aspectually sensitive tense that requires a homogeneous input, like the French Imparfait (De Swart 1998). It is not a pure tense operator like the English Simple Past. And it is not ambiguous. As it requires a homogeneous input, the S imple Past will coerce an imperfective interpretation when it combines with a tel ic predicate. A conversational implicature may sometimes lead to an interpretation on which the event progressed up to its culmination moment, which yields completion, i .e., a perfective interpretation, overruling the imperfective semantics. The latter effect will only show up in contexts in which an imperfective reading is not possible. This analysis of the Simple Past leads to the prediction that an imperfective interpretation is the preferred option; the perfective reading will only become manifest when an imperfective interpretation is not an option. This prediction is supported by the results from an aspectual comprehension study with Dutch adults. References Bach, Emmon. 1 986. The algebra of events. Linguistics and Philosophy 9, 5- 1 6. Boogaart, Ronny. 1 999. Aspect and temporal ordering: A contrastive analysis of Dutch and English. Doctoral dissertation, Free University at Amsterdam . Hollebrandse, Bart. 2000. The acquisition of sequence of tense. Doctoral dissertation, University of Massachusetts at Amherst.

Past tense interpretations 25 1 Hout, Angeliek van. 2005 . Imperfect imperfectives: On the acquisition of aspect in Polish. In Aspectual inquiries, ed. by Paula Kempchinsky and Roumyana Siabakova, 3 1 7-344. Springer: Dordrecht. Hout, Angeliek van. Submitted. Perfect understandings: Acquisition of perfect tense in Dutch and Italian and perfective aspect in Polish. Lingua. Special issue on Tclicity and Perfectivity, ed. by Heidi Harley and Rafaella FoUL Hout, AngeJiek van. In preparation. Perfect and imperfect understanding: Acquisition of three past tenses in Dutch. Talk presented at Language Acquisition between Sentence and Discourse, Radboud University Nijmegen, 2005. Poster presentation at BUCLD 30, Boston University, 2005. Krifka, Manfred. 1 992. Thematic relations as links between nominal reference and temporal constitution. In Lexical Matters, ed. by Ivan Sag and Anna Szabolcsi, 29-53. Stanford: CSLI Publications. Moens, Mark. 1 987 Tense, aspect and temporal reference. Doctoral dissertation, University of Edinburgh. Moens, Mark and Mark Steedman. 1 988. Temporal ontology and temporal reference. Computational linguistics 1 4, 1 5-28. Swart, Henriette de. 1 998. Aspect shift and coercion. Natural Language and .

.

Linguistic Theory 1 6, 347-385. Swart, Henriette de. 2000. Tense, aspect and coercion in a cross-linguistic perspective. In Proceedings of the Berkeley Formal Grammar conference, ed. by Miriam Butt and Tracy King. Stanford: CSLI Publications. Nerkuy l, Henk. 1 993 . A Theory ofAspectuality. Cambridge: Cambridge University Press. Verkuyl , Henk. 2005. How (in-)sensitive is tense to aspectual information? In Crosslinguistic views on tense, aspect and modality, ed. by Bart Hollebrandse, Angeliek van Hout and Co Vett 1 45- 170. Cahiers Chronos 1 3 . AmsterdamlNew York: Rodopi.

Why phonology is the same Harry van der Hulst

1.

Introduction

In this short contribution I would like to discuss and defend the idea that phonology and (morpho-)syntax are organized in parallel ways. Thus, I take issue with v iews that explicitly regard phonology as different, or implicitly adopt different theoretical models in both domains. The idea that phonology and syntax are parallel is, of course, hardly new, and the fol lowing discussion will identify some relevant sources. I felt, however, that to address this subject in a volume that is dedicated to a linguist who has always shown a vivid interest in both areas, regarding them as equally important and theoretically sophisticated, is especially fitting. Henk van Riemsdijk is one of the few linguists who, although being mainly active in the area of syntax, would systematically attend all phonology talks at, for example, GLOW conferences. Even though trying to do the opposite is more demanding for someone who is primarily a phonologist, I have always found his presence in linguistics inspiring and admirable. 2. The general organization of the grammar

I

adopt the view that the two fundamental units of language are words and sentences. Even though the distinction · as such is riddled with theoretical issues, problems and borderline cases, I believe that generally accepted distinctions such as the following testify to the fundamental natUre of the separation: Table 1 Syntax Phonolo1!Jl Semantics

Word Morphology Lexical phonology Lexical semantics

Sentence Syntax Post-lexical phonology Sentence semantics

Of course, providing labels for this distinction proves nothing in itself, in particular since some of them are just restatements of the distinction at

Why phonology is the same 253

some specific level. Traditional terminology does not necessarily hold 'the truth '. Despite such objections (and many others that could be made), I take the distinction as a fruitful directive in thinking about the organization of mental grammars, not only in research, but also in explaining issues to students (see Van der Hulst 2005). The lexicon plays a role in this distinction but hardly a defining one. It all depends on what one takes the . lexicon to be. Certainly, if the lexicon contains idiomatic expressions and the like, it can hardly be said that it contains all and only words. Side stepping this issue as well (referring for further discussion to Van der Hulst, . . to appear a), let us look at the internal organization of grammatical modules, of which we now postulate six (three at each level). Each component appears to have at least the following two ingredients: a. Basic units, primitives b. Combination rules Thus, each module accounts for a compositional structure of some sort in either words or sentences. Putting the semantic modules aside for the moment, I will adopt the following tenninology for the basic units and combination rules in each of the remaining modules:

Table 2: The organization of grammatical components

,Let us now move on and take into consideration some facts of language that ,we have so far ignored. We have assumed that the basic units combine into :�Clfger chunks while remaining unchanged. However (some might say: "', ' Iy) more is going on. In particular, it would appear that basic sometimes change in the combinations that they occur in. I refer to phenomenon as the variability of linguistic units (cf, also Van der 1st, to appear c). Linguistic units are like chameleons. They adapt to the ; )environment that they appear in. Let us consider the variability of . phonemes and words. We find two types of variability: a. Variabi lity in the inherent properties h. Variability in the distributional properties

254 Harry van der Hulst

At the level of words, variability of inherent properties refers to inflection. We note, for example, that in English verbs takes on various endings and the crucial point to take home is that these endings are 'forced upon' the verb by the properties of the sentence context that the words appear in. Hence we have a set of rules that add endings to verbs. These rules look a lot like word formation rules but they are not word formation rules. The inflectional rules specify affixation of words that is necessary for these words to occur in sentences, in combination with other words; they do not make new words. Since we will encounter rules with a similar function in the other components of the grammar, I will use the general term adjustment rules.

The second type of adjustment rules that we appear to need in syntax is the transformational rule. The distributional properties of words differ in different sentential contexts. In a statement, for example, we find the direct object next to the verb, while in sentences we have to look for it at the beginning of the sentence. Transformations are adjustments, like inflectional rules, in the sense that they modify the basic structures that are delivered by the combinations rules. Let us now look at the need for adjustment rules in phonology. Let us assume that English has a phoneme Ipl which occurs in words such as pin, spin, nap, jloppy etc. Close examination of the way in which this phoneme is pronounced in these various words reveals that there are differences. In particular, the Ipl that occurs at the beginning of a syllable with stress is pronounced with a little puff of air (called aspiration). To account for this variability we need rules that adjust the properties of the phoneme Ipl to its environment. We call the different realizations of a phoneme allophones, and the rules that account for them allophony rules. Allophony rules are adjustment rules that play a role which is analogous to inflectional rules. What about the second type of adjustment rules, those that would be analogous to transformations in syntax? It seem to me that all phonological rules that affect segments as a whole, substituting thein � by others (as in electric electricity), deleting them, inserting them or, in rare cases, reversing their order, qualify for this category; I will call them P-rules for short. It has of course long been noted that P-rules are transformational rules in terms of their formal properties, just I ike it has long been noted that phonotactic rules are analogous to phrase structure rules. Let us, then, add the adjustment rules to our model: -

Why phonology is the same 255

Table 3: The organization of grammatical components (take 2) Word phonolof,07 Sentence pbonolo� Morphology Syntax

Basic units phonemes

combination rules Phonotactic rules

word forms

prosodic rules

morphemes

Word formation rules

words

Phrase structure rules

aqjustment rules allophony rules P-rules

inflectional rules T-rules

Table 3 raises the obvious question as to whether the two open cells also contain adjustment rules. Numerous studies in the area of 'prosodic' or 'post-lexical' phonology have shown that, indeed, there is a lot of phonological action at the level of the sentence. However, I will not here try to c lassify such processes in inherent and distributional. As for morphology, for inherent adjustment rules, we could think of rules adding so-called linking units in compounds (hond-e-hok, schaap-s-wol). However, I ' ll leave it to the morphologists among us to develop the (underdeveloped) notion of adjustment rules in this module. (Crucially, one should not place rules that change the phonological shape of morphemes in this category; those are phonological P rules. Morphological adjustment rules must deal with morphological units.) What about semantics? On the one hand, one might expect to see that both semantic modules will contain a set of basic semantic units as well as a set of rules according to which these units can be combined; and then, in addition, we expect to need adjustment niles. The search for basic semantic units at the world level is ongoing and regarded by many, turning to prototype theories, as fruitless. Certainly, both morphemes and words, when combined into larger units shows variability of meaning, which suggest rules that manipUlate inherent properties. Sentence level scope . relations also suggest rules that are analogous to syntactic transformations. If we regard semantics are the third pillar of grammar (as suggested in Jackendoff 2002), we arrive at the following diagram:

256 Harry van der Hulst

(3 ) The organization of the grammar and o f the grammatical components Phonology Phonemes Phonotactic rules Allophony rules

Phonology Word forms Prosodic rules Adjustment rules

Lexicon words

Semantics word meaning

�sentences 1 Morphology Morphemes WFRs Adjustment ru les

Semantics Sentence meaning

Syntax Words PSRs Inflection rules

In the two semantic boxes I have left undecided exactly what kinds of basic units and rules are needed. In all other boxes we find three things: basic units, combination rules and adjustment rules (WFRs = word formation rules; PSRs is phrase structure rules). The bottom line appears to be this: We have words and we have sentences and at both levels we have three systems of rules that assure that each unit (word or sentence) is wellformed phonologically, morpho­ syntactically, and semantically. All six systems (or at least those that regard phonology and syntax) appear to be organized in a very similar way. 3. Bringing in phonetics

As an alternative to v iewing semantics as the third, independent pillar of the grammar, many linguists have sought to establish a close and direct relationship between syntax and semantics, for example in the tradition of Montague-grammar and many more recent approaches. Syntactic structures are constructed and hand-in-hand a semantic representation (or interpretation) is constructed. This view creates an asymmetry between phonology and syntax. A distinct property of phonology is that it deals with

Why phonology is the same 257

units and structures that, as such, have no meaning . (Alas, we phonologists spend out lives on dealing with meaningless matters; may be that's what's different) But wait, just like syntactic units and constructions have cognitive content (i.e. meaning), totally and utterly independent from their phonological form, phonological units and constructions have phonetic content. Thus, one would expect an analogous situation on the side of fonn: phonology specifies units and complex structure and as such structures are formed, phonetic rules .compute the sum total phonetics. Indeed, explicit parallels between the hand-in-hand efforts of syntax and semantics and of phonology and phonetics have been explored in Wheeler' s Categorial Phonology (Wheeler 1 98 1 , expl icitly based on a conception inspired by Montague style grammar) and we find the same parallelism in Declarative Phonology (cf. Scobbie 1 997, Scobbie et aI. 1 996). (This whole class of approaches both to syntax-semantics and phonology-phonetics share the characteristic of being skeptical of adjustment rules, but that could just be a ·· coincidence. ) One might argue that both cognitive content and phonetic content are external to grammar in a narrow sense, and that precisely in this sense, grammar is a mechanism to link cognitive content and phonetic content (or, as we say, meaning and form). Hence, I would like to revise the diagram in (7), leaving out details: words

phonetics -

PhonOIOgy

()

syntax

----+

semantics

sentences 'The thin arrows simply means 'characterize' or 'check', i.e. the role of phonology and syntax is to check the wellfonnedness of linguistic , , expressions (words and sentences). The fat arrows imply a dependency " relation. The phonetic and semantic-cognitive side of language is (if the arrows point the right way) dependent on the phonology and syntax, 'respectively. In this diagram there is no dependency between phonology : ,)ind syntax, no central engine that drives the system (as in standard i'��c:homskyan models). One might argue that this is wrong, that syntax drives ' L::!fue whole thing. However, one might then, in addition, argue that what /really drives the whole thing is semantics, without any doubt the central ·

258 Harry van der Hulst

aspect of human language. Thus, we revise the structure again. This diagram expresses an implicational hierarchy in that semantics (in the general sense of cognitive structure) does not presuppose anything . Looking at the other end of the implicational chain one might say that each step presupposes all prior steps: (5)

words

phonetics +--

PhonOIOgy

�)

syntax

�

semantics

sentences Also, given the close link between phonetics and phonology on the one hand and between syntax and semantics on the other, we expect that it is not always clear how certain phenomena on one side (fonn) or the other (meaning) are best dealt with . Syntacticians quarrel with semantic colleagues or their own semantic alter ego, and phonologists do the same with reference to phonetics (cf. Van der Hulst, to appear c). Needless to say that all the above issues are heavily dependent on the specific theories that one is committed to, although no such commitment should keep one from seeing the importance of the issues at hand and exploring alternatives. Whatever one likes to see as the central engine, and however one resolves the boundary conflicts and divisions of labor, the above has made clear, I hope, that there is not much difference between phonology and syntax. Indeed, from the very start, Chomksy himself argued for a parallel design, especially with regard to the recognition of the class of adjustment rules, claiming that his ideas for the design of syntax were inspired by his ideas for the design of phonology. Subsequent developments have shown a divergence, at least in what is sometimes called 'mainstream generative grammar' (Van der Hulst 2004). Whereas, for example, the distinction between lexical and post-lexical phonology flourished and lead to the recognition of certain diagnostic properties of both classes of rules, a similar distinction in syntax worked out rather differently (mainly to recognizing word fonnation as lexical, but not so-cal led bounded transfonnation relations) (cf. Kiparsky 1 978, Hoekstra, Van der Hulst and Moortgat 1 98 1 ). More recently, we have seen a spectacular acceptance of Optimality Theoretic approaches of phonology language, while approaches to syntax have not been affected in the same way. Thi,s, of course, makes no

Why phonology is the same 259

rational sense, unless it is the case that people doing phonology are genetically different from people doing syntax. This seems unlikely (as I'm sure Henk van Riemsdijk would agree). More likely is it that such different theoretical choices have everything to do with fashion, good public , relations and Chomsky'S dominance in syntax. All things being equal, given that phonology and syntax are parallel in the ways suggested here, it . would be highly unlikely that OT is good for phonology and bad for syntax. It is either good for both, or for neither (cf. V an der Hulst and Ritter 2000, Van der Hulst 2004). 4.

Structural analogy

The idea that phonology and syntax are analogous is, as mentioned at the beginning of this article, neither mine nor ne�. I simply adopt it as a working hypothesis and I believe that it leads to fruitful results. The idea has received a name in the work of John Anderson, the developer of so­ ' called dependency approaches to both morphology, syntax and phonology: .structural analogy (cf. Anderson 1992, Van der Hulst, 2000, 2003, to appear b). Anderson is meticulous in tracing the roots of this idea in ' linguistics. Without going into specific formulations, let us say that the assumption of structural analogy holds that different modules of the ,grammar are organized in the same way, making use of the same structural relations. Modules differ due to the fact that they start out with different , sets of primitives. In addition, Anderson reckons with differences that are to the interaction between components. I take this to mean that differences in dependency between modules might affect the organization of one or the other module. Due to Anderson's work I personally got �convinced by the idea that phonology and syntax are not different at all, but in fact, highly parallel. The parallelism in his dependency approach go far 1>eyond the idea that both modules deal with hierarchical structure and .include specific claims concerning the kind of relationship (one of d�pendency) that holds between the units in such structures. Very similar Jqeas have been pursued in a related model, Government Phonology. Most :of mainstream generative phonology has ignored this important line of research and thus a chance to explore the parallelism between syntax and : phonology. In extreme cases, this parallelism has been explicitly denied (cf. Bromberger and Halle 1 989 cf.. Van der Hulst, 2000, to appear a). . Why would structural analogy exist? It was once expressed to me by ' Morris Halle (p.c.) that this idea is at odds with the idea of modularity. If modularity is taken to mean that different components have their own

260 Harry van der Hulst

specific vocabularies and tasks, why would one expect similarities in their internal organization, he said, adding: one would expect the opposite. True, it is all a matter of expectation (based on general considerations). My expectation of structural analogy is based on the idea that it would be unreasonable to assume that the different modules of the mind use dramatically different mechanisms. A lot has been learned in the recent decades about the development of biological organisms. One thing that I am thinking of here is the idea (pushed especially in the evo-devo approach --evolutionary developmental biology- that differences between organisms do not arise because they are genetically totally different, but because the same genes are used in different ways, at d ifferent points in the development, more or less often in different places of the structure and so on. This idea squares nicely with the discovery that organisms tum out to have much less genes than was previously believed. One can do a lot with a relatively small number of genes if they are recycled, used over and over again within the same organism. If, as is now generally believed, the mind (including the mental grammar) is based on a genetic substrate (and develops environmentally), it makes sense to expect that different modules use and recycle the same types of neuronal connections. If the idea that phonology and syntax (and other mental modules) are similar because they are based on a neural substrate that emerges on the basis of reactivating the same genetic instructions we might want to say that the similarities are not just analogies, but, stretching the term, homologies. The resemblance is due to common genetic decent. There is another reason for expecting similarities, however, one which goes beyond considerations regarding, grammatical modules or, indeed, mental modules in general. As Abler ( 1 989) points out there are far­ reaching similarities between the organization of language grammars and the organization of, for example, chemistry and physics. In all these domains we find discrete units that combine into 'ever , larger structures. Indeed, this almost seems a point of logic. How .else can 'complex structures be understood? Others such as Simon ( 1 996) and Yolk (2002) have pointed out that certain structural patterns are pervasively recurrent not just in the human mind, but in the (human) universe. It is tempting to speculate on the consequences of such broad-ranging similarities but it would seem that my time is up.

Why phonology is the same 26 1

References Abler, William

1 989. On the pariculate principle of self-divers ifying systems. Journal ofSocial and Biological Structures 12: 1 -2. Anderson, John M. 1 992. Linguistic representation: Structural analogy and stratification. Berlin and New York: Mouton de Gruyter. Bromberger, Sylvain and Morris Halle 1 989. Why phonology is different. Linguistic Inquiry 20/1 : 5 1-70. Hulst, Harry van der. 2000. Modularity and modality in phonology . In Phonological knowledge: its nature, ed by N. Burton-Roberts, P. Carr and G. Docherty, 207-244. Oxford: Oxford University Presss. Hulst, Harry van der. 2003 . Cognitive phonology. In Germania at alia. A l inguistic th Webschrift for Hans den Besten on the occasion of his 55 birthday . Hulst, Harry van der 2004. Phonological dialectics; A short history of generative phonology. In Linguistics today - Facing a greater challenge, ed. by P.O .J. van Sterkenburg, 2 1 7-243. Amsterdam: John Benjamins Publishing Company. Hulst, Harry van der. 2005. A mind for language. KendalllHunt Publish ing Company.

Hulst, Harry van der. to appear a. On the parallel Organization of linguistic components. In

Phonology and syntax- the same or different.

Special issue of

the journal Lingua, ed by R. Bermudez-Otero and P. Honeybone. Hulst, Harry van der. to appear b. Dependency Phonology. In: language and linguistics. 2nd edition. Elsevier.

·

Hulst, Harry van der. to appear c. De vorm van taal. Hulst, Harry van der and Nancy Ritter.

2000.

The encyclopedia of

Nederlandse Taalkunde.

The SPE-heritage of OT.

The

Linguistic Review 1 7/2-4: 259-290.

Hoekstra, Teun, Michael Moortgat and Harry van der Hulst. 1980. Introduction. In Lexical Grammar, ed by T. Hoekstra, H. van der Hulst and M. Moortgat, 1-48. Dordrecht:

Foris 2002. Foundations of language. Brain, meaning, grammar,

Jackendoff, Ray. •

Evolution. Oxford: Oxford University Press. 1 978. Recent development in generative phonology. In The Nordic Languages and Modern Linguistics. Vol. 3, ed by 1. Weinstock, 42-50.

Kiparsky, Paul.

University of Texas, Austin Texas.

Scobbie, James.

1 997. Autosegmental Representation in a Declarative Constraint­ based Framework. New YorkILondon: Garland press. [Original dissertation: Attribute Value Phonology. Edinburgh, 1 99 1 ] ' ;:Slcobbie, James, M., Jolm S. Coleman and Steven Bird. 1996. Key aspects of Declarative Phonology. In Current trends in phonology: Models and methods, Vol. II, ed by Jacques Durand and Bernard Laks, 685-7 10. European Studies · .

.

� of Salford .

Research Institute (ESRI) and Manchester: Universi

• Simon, Herbert. 1 996. The sciences of the artificial. 3 The MIT Press.

edition. Cambridge etc.:

262 Harry van der Hulst Yolk, Tyler. 1 995. Metapatterns across space, time and mind. New York: Columbia University Press. Wheeler, Deirdre. 1 98 1 . Aspects of a categorial theory of phonology. PhD dissertation, Umass.

Recursively linked Case-Agreement: from accidents to principles and beyond runy Huybregts

1.

Introduction

All languages incorporate a fair amount of 'noise'. Chaotic relics from the past without obvious systematic links to their core systems. An example may be the inflected complementizers of numerous Germanic dialects. Sometimes, however, these historical accidents and dialect mixtures can be shown to systematically interact with more principled, internally directed, systems of language. Here we will argue that the morphosyntax of complementizers is crucially implemented in core systems of basic - recursion and interface conditions imposed on 'narrow syntax.' Properties of Case valuation of 'parallel' argument chains are recursively linked to Agreement processes, which, we will argue, define the (im)possibility of that-trace effects. Using Chomsky's (2005b) intriguing terminology, we might say that ' 1 5' factor' chance properties interact with '3fd factor' general language-independent principles to explain '2nd factor' conditions of 1languages, including UG. 1 -

2 An analysis of complementizer-trace effects .

As is well-known subject extraction" across an adjacent complementizer ·�"�" lbehlav��s differently across the world's languages. For example, that-trace effects occur in English ( 1 ) and French (2) but not in Spanish (3 ) or West­ ....&0" •• . (4). These different behaviors reflect to some extent effects of the :path-dependent evolutionary history of the language. Romance complementizers (que, che) differ idiosyncratically from Germanic complementizers (that, da) and are homophonous with interrogative rather ,..

ready Thein titlecentury Cartesiastudy n ratiofonalthisosem tdays, hese issues were insightfully ,disAlcussed. of a famous mentalconcern istic inteforpretr thaetio'tnhreeof unifactors' versalingrammar, clhearlofy lasuggest s thAnat (tathtereemptwaswas made deep t h e growt nguage. to explof agener in language-dedi cyated,directuniedversalmechani propertismsesofofthelanguage ('('graieneral e ' ) i n t e nns a l , i n t e rnal l sonnee'). 1

: raisonnee, _

1 7th

a

Grammaire generale et

a

mind

264 Riny Huybregts

than demonstrative pronominals. Similarly the parametric variation of 'agreeing' compJementizers in West-Flemish (and several Dutch dialects) vs. 'non-agreeing' compiementizers of English and other languages reflect historical accidents. But in the latter cases these 'accidents' are associated with conditions on extraction that may be given a more principled account in tenns of 'third factor' considerations, i.e. principles that may not be specific to language but are organism-independent in that they reflect general properties of complex organic systems (Chomsky 2004, 200Sa,b). ( 1) (2) (3) (4)

* [cp who do you think [cp ec that [TP tWH read this book ]]] *[cp qui crois-tu [cp ec que hp tWH a lu ce livre ]]] [cp quien crees [cp ec que [TP ha tWH leido este libro ]]] [cp wien peinz-je gie [cp ec da [TP tWH dienen boek gelezen eet ]]]

In French and English there are alternative, language-specific, ways to get around this constraint: deletion (5) or modification (6) of the complementizer that causes the problem . See Perlmutter ( 1 97 1 ), Chomsky and Lasnik ( 1 977), Taraldsen ( 1 978), Pesetsky ( 1 98 1 ), Kayne (1 980), and Chomsky ( 1 98 1 ) for some successive and increasingly insightful analyses of this effect. (5) (6)

[cp who do you think [cp ec 0 [TP tWH read this book ]]] [cp qui crois-tu [cp ec qui [TP tWH a lu ce livre ]]]

In contrast, Spanish and West-Flemish have their parameters 'selectively stabilized' to license these effects in different ways. Properties of 'null subjects' (Chomsky 1 98 1 ) and 'agreeiilg complementizers' (Rizzi 1 990) allow for 'proper government' of the wh-traces in these languages.2 The various treatments · of the that-trace effect add up .to increasingly explanatory acco!lnts which document the intellectual history of the field in an interesting way.3 Initially, the effect was yiewed . as . more or 'less 'accidental' and was accounted for in language-particular and 'construction­ specific tenns. These ' 1 st generation principles' included e.g. Perlmutter 's

En aa n licensed position. ou Anagnostopoul (1998), CP ( c p e EPP-conditioton Spec-TP,wh-pbrase ic neSpec-vP es inSpec­u l minating di(2001). scussion these considerations a broadera contextgeneral see Freianddin andil uVergnaud

2

xtr ctio in 'null subject' languages is from post-verbal A mi im list refonnulation, using ideas due to Alexiadou and would allow for direct movement of the from to wh h is not c sary 'n l w ithout the ' ara ll l ' movement subje t languages' for the to be satisfied). 3 For more of in

Recursively linked Case-agreement 265

Generalization, Chomsky and Lasnik's that-trace filter, the Nominative Island Condition (Taraldsen, Pesetsky, Kayne), and were descriptive . generalizations rather than truly explanatory principles. This state of affairs changed completely with the Principles-and-Parameters framework, which was a major leap forward with abstractly formulated principles of universal grammar that were no longer construction-specific or language-particular 'but reflected the properties of internally directed mechanisms of I-language more directly. The '2nd generation principles' of this model were language­ dedicated in a general sense and included e.g. ECP to account for the that­ . ' trace effect (Chomsky 1 98 1 , 1 986; Rizzi 1990). The complementizer that, mtervening between the Spec-CP and Spec-TP positions, acted as a 'min imality barrier' blocking proper government of the offending variable by the intermediate element in Spec-CP. With the Minimalist Program the notion of government became no longer formulable and the explanation of the that-trace effect was lost. Assuming that the Strong Minimalist Thesis is basically correct this effect must now be shown to follow from deeper design features or general properties of computational efficiency. If this goal could be achieved we would have explained the condition in language-independent terms and we would have moved the is to a higher level of explanation (Chomsky's 'third factor,).4 We try to show that in cases of that-trace violations, the '3rd generation 'principle' of Full Interpretation, an interface condition (IC) which requires all un interpretable elements be eliminated at the CI-interface, is such a .

As shown in Chomsky (2005a,b) basic recursive properties of internal give rise to the concept of 'parane) movement'. Since we will rely heavily on 'parallel movement' to , explain complementizer-trace phenomena this notion will be briefly discussed below. To this end, :consider some important data exemplifying long distance agreement in ;�:'Icl�la[lld i'c (adapted from Holmberg and Hr6arsd6ttir (2003)).

"E;;lUl"'1�n".

a. * [cp hvaaa stUdent veist �u [cp ec a5 [TP tWH vir5ast [vp tDP t v [rp hestarnir vera seinir ]]]]] 'To which student do you know that the horses seem to be slow'

'.'; 4 See Pesetsky and Torrego (200 1 ) and Szczeglelmak ( 1 999) for alternative minimalist accounts (in tenns of derivational economy and conditions on phasal movement of left edge elements respectively). •

•

•

266 Riny Huybregts

b. veist l>u [cp ao hp )Jessum student viroast [vp tDP tv [TP hestarnir vera seinir ]]]] ' Do you know that the horses seem to this student to be slow ' c. *Jlao viroast [vp einhverjum stUdent tv [TP hestamir vera seinir ]] 'It seems to some student that the horses are slow' The problem here is, Why does the trace of DP in Spec-VP b lock long distance agreement between the tensed verb in T and the embedded nominative DP in (7a) but not in (7b)? What is it that (7a) and (7c) have in common and which makes (7b) different? Chomsky's answer is that Spec-VP in (7a), just as in (7c), but not in (7b), is a completed, well­ formed, A-chain, which blocks agreement. The reason is that 'parallel movement' simultaneously merges the wh-phrase in Spec-VP to TP and CP, giving rise to two independent chains, a consequence of phase-based computation with heads of phases controlling AgreelMerge. The features driving syntactic computation reside in the heads of phases, viz. C or 'strong' v (annotated as v *), but they may be ' spread' to the elements these phasal heads select, viz. T or V . In particular, the Agreement-features of C,v* probe categories that are checked in the specifier positions of T,V (acting as probes by proxy with features inherited from C,v *). Focusing on (7a), the operator wh-phrase in Spec-CP and the argument phrase in Spec-TP have been independently selected by the edge feature of C and Agreement-features of T, inherited from C, from the single source position Spec-VP. The edge feature of C is checked in Spec-CP but its Agreement-features are checked in Spec-TP by proxy. We therefore have two A-chains, (Spec-TP, Spec-VP) an� (Spec-VP), in the derivation of (7a). However, in (7b) the search domain of Probe T excludes the head of the A-chain (Spec-TP, Spec-VP), and furthermore, the element in Spec-VP cannot be a blocking factor for agreement in (7b) since only the maximal chain (or its head) can be an intervention factor. However, the singleton chain ( Spec-VP) can be such a factor causing an intervention effect in (7a) and (7c). Spec-v*P, the first Merge position of subjects, rather than Spec-TP, their second Merge position, is the relevant Goal for Probe C . In fact, the completed argument chain (Spec-TP, Spec-v*P) would have all its features valued and therefore be ' inactivated' and ' invisible' for Probe C. This analysis leads us to expect that that-trace phenomena should not be restricted to Spec-TP positions. If true, there is additional motivation for

Recursively linked Case-agreem ent 267

rejecting an account in tenns of the notion of government (ECP), unformulable in minimalist versions of linguistic theory.5 There is independent evidence, pointed out in Chomsky's On Phases (Chomsky 200Sb), for saying that the element in Spec-TP (T tensed) must be ' inactive ' because the argument-chain it heads is fully valued. The Subject-Island asymmetry between subjects of transitive clauses vs. 6 subjects of intransitive clauses is one piece of evidence for this position. Th is analysis can be extended straightforwardly to some classical problems .. of linguistic theory. Consider for instance some syntactic properties of French clitic en, which may precede (en-arriere), as in (8a), or follow (en­ avant), as in (9), the DP into which it was base-generated. But since the analysis of unaccusatives (9) could be assimilated to the analysis of passives (8b), en-avant could be actually reanalyzed in all these cases as en-arriere (ordering cliticization of en before raising DP).7 (8) a. it n' en a pas publie la preface b. la preface n ' en a pas ete publiee

: (9)

la preface n' en est pas arrivee

The invited and welcome conclusion at the time was that en-avant is i��generally excluded (since it gives rise to improperly bound traces) and the · . ill-formedness of ( 1 0) falls out as a special case of this condition of proper :binding. But the invited inference is invalid if all surface subjects including . transitive subjects start out inside the verbal projection (v*P or VP). Cliticization from Spec-v*P is no longer straightforwardly blocked . .5

We wou ld expect there to be languages which exclude both subj ect and object extraction across overt com pl ementi zers. Selayarese (Finer 1 997) and Bahasa Indonesia (Saddy 1 99 1) seem to be such lan guages . Discussion of these languages . must be deferred for reasons of limitations of space . . 6 Consider (i) *the truck ofwhich the driver caused the accident vs. (ii) the truck of :which the police found the driver. See also (iii) the truck of which the driver was killed in the accident. The contrast is not between subjects vs. objects but rather between 'transitive subjects' vs. ' intransitive subjects' (which go along w ith 'objects) and must be determined prior to the raising of OP to Spec-TP . . Furthermore, the contrast between (iv) who was (*therej hurt in the accident vs . . (v) there was no one hurt in the accident falls out from an analysis incorporating '�parallel movement.' See Chomsky (2005b) for detailed discussion and insightful analysis. 7 See Ruwet ( 1 972) for early discussion and analysis.

268 Riny Huybregts

( 1 0)

*Ia preface n'en a pas provoque personne

The clitic has been extracted from DP but extraction is blocked from transitive subjects. As shown by the contrast between ( 1 0) and (9)/(8b), the effect is determined by the base structure of the element from which the clitic has been extracted, not by its derived structure. The question is, Why does this asymmetry exist? Transitive subjects also originate in the verbal projection selected by T. So why do they constitute islands for clitics? The concept of 'parallel' movement forced on us by the assumptions underlying phase-based computation gives us an immediate answer. ( l 1 ) a. [cp C [TP T [v p DP v* [vP V [DP D [NP . . en ]]]]] ] b . [cp C [TP T [v p [DP 0 [NP .. en ]] [ v* VP ] ]]] ..

..

..

..

At the phasal level v*P of ( 1 1 a), illustrating in part the derivation of (8a), the clitic can be internally merged to v*P from the first merged 'object' position before it is subsequently merged to TP on the next phase CPo Such a derivation is blocked for transitive subjects in the Spec-v*P position of ( 1 1 b), in part illustrating the derivation of ( 1 0), since this position is not within the domain of the phasal head v* controlling internal Merge. Furthermore, v*P is a strong phase and transitive subject DPs in Spec-v*P cannot be internally searched by a higher phasal C probe (locality effect of minimal search). The ill-fonnedness of ( l 0) thus receives a natural explanation, which in all relevant respects is identical to the one given in Chomsky (200Sb) for linguistic data showing a Subject Island asymmetry . (cf. footnote 6). The analysis of (8b) and (9) is straightforward and will not ' . be discussed here. 8 Let's consider now in some detail the derivation leading up to a that­ trace effect in the grammar of English. .

( 1 2)

.

. [cp wh-XP that [TP .

a

..

T [v.p P v* [vp read this book ] ]] ]

Assuming the concept of 'phasal derivation' with the heads of phases control ling the computation, the q>-features and 'edge' feature EF of C spread to the head of the complement it selects, here T. C, and T by proxy, act as Probes looking for appropriate Goals in a minimal search. The 8

Further empirical support for phase-based 'parallel movement' may come from a reasonable analysis (omitted here) of the locative inversion contrast illustrated by (i) the town near which (she believed) was situated an old ruin vs. (ii) *the old ruin which (she believed) near the town was situated (see footnote 6).

Recursively linked Case-agreement 269

linguistic object P in Spec-v*P carrying a wh-feature as well as q>-features is such a GoaL Agree applies and movement is 'parallel' in the sense that � merges internally to TP and CP simultaneously: it is selected by T (q>­ features) and it is selected by C (edge feature). Feature spreading leads to independent movements to Spec-TP and Spec-CPo 'Parallel movement' thus creates two independent chains (i) (Spec-CP, Spec-v*P) and (ii) (Spec'. TP, Spec-v*P). The relation between Spec-CP and Spec-TP is indirect only. As a consequence two independent argument chains are derived. A-Chain CHI : (Spec-TP, Spec-v*P) A-Chain CH2: (Spec-v*P) '. The cp-features of the subject in Spec-v*P agree with the features of its Probe T and its own Case-feature will be valued to reflect this agreement. The chain CH I is therefore well-fonned (all its features have been valued) and can be no longer computationally active. That is why extraction from Spec-TP, merging the head of the argument chain ( l 3a) to a phasal Hcategory containing it, is impossible. Direct movement from Spec-v*P to 'Spec-CP is computationally straightforward but the result is an argument chain CH2 which is imperfect. Since C is not involved in any cp-feature (this task it has delegated to the tense category selected by it), , ·the Case-feature of this chain cannot be valued. The result is an ill-formed chain in violation of the principle of Full Interpretation. The contrast between ( l ) and (5) follows from properties of 'spreading' if we assume that 'zero' declarative compiementizers spread .. . features to the Tense elements they select while holding on to them. :This is what 'spreading' means. For overtly expressed complementizers we ",,,U.4"" Y ..... two different situations: either (i) features are evenly spread over C . and T as in the case of 'zero' complementizers, or (ii) C parts with (some ot) its features, transferring these to T without retaining them. The former ('spreading'), with morphologically expressed complementizer 'agreement, is illustrated by West-Flemish� the latter option ('transfer'), with non-agreeing overt complementizers, by Standard English. .. . . The 'offending' complementizer of English ( 1 ) is involved in the ·" :' . analysis given above for (12), which results in a violation of an interface condition (Full Interpretation}. In contrast, the ' licensing' complementizer ,of English (5) allows for a. 'double check' for the edge features and cp­ . " footures of C and T. The wh-phrase in Spec-v*P is independently selected ipy C and T, more precisely by the features of C and of T, inherited from C, . " )vhich happen to be the same (a property of 'spreading'). Edge features and ·

'

270 Riny Huybregts


( 1 4) a.

West-Flemish complementizers have their features evenly spread over C and T. The same features act as Probes twice: edge and agreement features of C select the 'subject' in Spec-v*P and the edge and agreement features that T has inherited from C do the same. The 'subject' wh-phrase in Spec­ v*P (its first Merge position) targets simultaneously TP and, CPo Merging to TP and CP results in 'parall�l' chains but each of these ch�ins has its Case­ feature independently valued as a reflex of successful i'agreement The analysis is analogous to the one given above for the 'zero' complementizer 10 of English. 9

For detailed discussion of complementizer agreement in West-Flemish see Haegeman ( 1 992). JO Complementizer agreement correlates also with (obligatory) ' doubly filled comp' (DFe) as shown by (i) k-weten nie [wa *(da) dienen vent - geeten eet] and (ii) k-weten nie [wa *(da-n) die venten - geeten een] ' I don't know what that/those guy/guys haslhave eaten' . These three correlated properties

Recurs ively linked Case-agreement 27 1

�ii;:o��:(W'

We assume that standard Dutch complementizer dat 'that' has abstract agreement features reflecting the evolutionary history of the language. Complementizer agreement is abundantly manifested in numerous regional dialects of Dutch. 1 1 Abstract agreement is nothing more than a residual factor of local dialects preserved in standard varieties of present day Dutch. Assum ing this 'resistant trait' view of abstract complementizer agreement to be correct, standard Dutch will be analyzed in exactly the same way as West-Flemish. Still, it has been suggested that intuitions of that-trace effects in standard Dutch are not always unequivocally sharp, or that these effects are licensed in one variety of standard Dutch, Dutch A, but not in another variety, Dutch B (Maling and Zaenen 1 978). In fact, Dutch A and Dutch B do not exist as regionally or socially defined dialects (as observed by Maling and Zaenen) but may flourish at the organismal level (1languages). An expected outcome of the biolinguistic perspective adopted by the Principles-and-Parameters framework. Random, chaotic distribution of these different 'dialect' speakers across the Dutch population can be reasonably explained as a 'threshold effect' of 'cue-based' parameter setting (Lightfoot, 1 999). We have argued that 'doubly-filled . COMP' (DFC), agreeing complementizers and licensing of that-trace effects . are correlated properties of language (e.g. positively in West-Flemish and negatively in standard English). The profligacy of dialectal cues stands in sharp contrast with the parsimony of cues of standard Dutch: complementizers do not overtly agree, DFC is optionally and marginally used, and that-trace effects are evidenced in complex data structures only. ,All in an these 'stimuli' may be too 'poor' for many organisms to act upon. 'Selective stabilization of parameters only works if the relevant cues are robustly present in the environment. This may explain the random and unstable distribution of standard Dutch intuitions of these phenomena. Stability at the organismal level, where I-language (a psycho linguistic condition) receives a natural interpretation, does not contradict randomness ' (complementizer agreement, DFC, and licit that-trace effects) can be given a unified analysis. The reason 'doubly filled compo is excluded in Standard English but obligatory in West-Flemish must be that features relevant for satisfying interface conditions cannot be deleted. This inhibition could be a consequence of the concept of 'spreading' under the assumptions of phase-based computation. Agreement features spreading from C to T act together (as a unit). Deletion of the complementizer in DFC contexts entails deletion of agreement in C and therefore deletion of agreement in T. As a consequence, Case-evaluation of the Nominative : argument is obviated and the result is a 'crash' that cannot be avoided. II See Zwart ( 1997) for explicit discussion and for further references.

272 Riny Huybregts

at the population level, where E-language (a sociolinguistic condition) may find its more natural habitat. 12 To conclude, we have argued that phasal derivation with its Agree­ based Probe-Goal mechanisms of Internal Merge gives rise to 'parallel movement' operations resulting in parallel argument chains. Each of these chains m ust satisfy interface conditions of Case-valuation on its own. In particular, each A-chain must have its Case-feature valued as a reflex of Agree. In environments that show that-trace violations, Case-Agreement properties of 'parallel ' chains are unvalued and lead to crashing derivations at the semantic interface. 3. Conclusion

Seemingly 'accidental' features of language such as the ' agreeing' complementizers of some Germanic dialects are shown to be related to processes that belong to the core computational mechanisms of language. Idiosyncratic properties of subject extraction across complementizers can be given principled analyses in terms of Full Interpretation, an interface condition that is imposed on 'narrow syntax' by language-external cognitive systems. Specifically, complementizer-trace effects show that Case-Agreement is an integral part of the basic recursive operation of (internal) merge. Accidental properties of language are explained as the outcome of general, language-independent, conditions imposed on language-dedicated systems of VO. Chomsky's story of 'three factors', essentially. Experiential 'first factor' elements in the growth of language thus receive a principled explanation from ' second factor' mechanism s of genetic endowment that are constrained by 'third factor' architectural conditions on complex systems. One can hardly fail to notice that significant advances have been made in the explanation of these phenomena ever since they were added to the research agenda of generative grammar, advances which have become possible only recently, with the biolinguistic perspective on language as adopted by minimalist versions of the Principles-and-Parameters framework.

12

For an intriguing discussion of the biolinguistic perspective see Pi attelli­ Palmarini (1 989), where 'selectivist' solutions are favorably compared to ' instructivist' accounts in biology as well as in the cognitive sciences.

Recursively linked Case-agreement 273

References Alexiadou, Artemis, and Elena Anagnostopoulou. 1 998. Parametrizing AGR: Word order, V-movement, and EPP-checking. Natural Language and

Linguistic Theory 1 6: 49 1-539. Chomsky, Noam . 198 1 . Lectures on Government and Binding. Dordrecht: Foris. Chomsky, Noam. 1 986. Barriers. Cambridge, Mass.: MIT Press. Chomsky, Noam. 2004. Beyond explanatory adequacy. In The Carthography of Syntactic Structures. Vol. 3, Structures and Beyond, ed. by Adriana Belletti. Oxford: Oxford University Press. omsky, Noam. 2005a. Three factors in language design. Linguistic Inquiry 36. 1 : Ch 1 -22. Chomsky, Noam. 200Sb. On Phases. (MIT unpublished ms.) Chomsky, Noam, and Howard Lasnik. 1 977. Filters and Control. Linguistic Inquiry 8.3 : 425-504. Finer, Daniel. 1 997. Contrasting A-dependencies in Selayarese. Natural Language and Linguistic Theory IS: 677-728. Freidin, Robert, and Jean-Roger Vergnaud. 200 1 . Exquisite connections: Some remarks on the evolution of linguistic theory. Lingua I l l : 639-666. Haegeman, Liliane. 1 992. Theory and Description in Generative Syntax: A Case Study of West Flemish. Cambridge: Cambridge University Press. Holmberg, Anders, and Thorbj6rg Hr6arsdottir. 2003. Agreement and movement in Icelandic raising constructions. Lingua 1 1 3 : 997- 1 0 1 9. Kayne, Richard. 1 980. Extensions of binding and case-marking. Linguistic Inquiry

1 1 . 1 : 7S-96. Lightfoot, David. 1 999. The Development of Language: A cquisition, Change. and Evolution. Oxford: Blackwell. Maling, Joan, and Annie Zaenen. 1 978. The nonuniversality of a surface filter. Linguistic Inquiry 9.3 : 47S-497. Perlmutter, David. 1 97 1 . Deep and Surface Structure Constraints. New York: Holt, Rinehart & Winston. Pesetsky, David. 198 t . Complementizer-trace phenomena and the Nominative Island Condition. The Linguistic Review 1 : 297-343 . Pesetsky, David, and Esther Torrego, 200 1 . T-to�C movement: Causes and consequences. In Ken Hale: A Life in Language, ed. by Michael Kenstowicz, 3 55-426. Cambridge, Mass.: MIT Press. Piattelli-Palmarini, Massimo. 1 989. Evolution, selection, and cognition: From 'learning' to parameter setting in biology and the study of language. Cognition

3 1 : 1 -44. Rizzi, Luigi. 1 990. Relativized Minimality. Cambridge, Mass.: MIT Press. Ruwet, Nicolas. 1 972. Theorie Syntaxique et Syntaxe du Franr;ais. Paris: Editions du Seuil.

274 Riny Huybregts Saddy, Douglas. 1 99 1 . Wh-scope mechanisms in Bahasa Indonesia. In More Papers on Wh-Movement, ed. by Lisa Cheng and Hamida Demirdache, MIT WPL 1 5, 1 83-2 1 8. Szczegielniak, Adam. 1 999. 'That-t Effects' cross-linguistically and successive cyclic movement. In Papers on Morphology and Syntax. Cycle One, ed. by Karlos Arregi et al. , MIT WPL 33, 369-393. Taraldsen, Knut Tarald. 1 978. On the NIC, vacuous application and the that-trace filter. Bloomington, Ind.: Indiana University Linguistics Club. Zwart, Jan-Wouter. 1 997. Morphosyntax of Verb Movement: A Minimalist Approach to the Syntax ofDutch. Dordrecht: Kluwer.

Enfoldment as Economy Takashi Imai

1.

Introduction

Recent ideas on economical and optimal considerations in biolinguistics noW perceived as the Minimalist Program (MP) are not so new, but rather could be uncontentious in other fields of the natural sciences such as physics and chemistry for instance. One can think of Fennat's Principle of Least Time that states that light propagates between two points in such a way as to minimize its travel time or Jacobi's Principle of Least Action defining that a true trajectory makes the action stationary or the Principle of Least Potential Energy saying that a system in a state of rest in an inertial reference frame occupies a stationary value of its potential energy function. (Cf. Lemons ( 1 997).) Such 'Least effort' notions are natural selections for physical as well as linguistic (grammatical) principles explicated in a series . of Chomsky'S recent work, J and would be eventually common for everything. In the MP, the derivation and representation in the Internal (intentional) language (I-language) should be optimal. In other words, conditions andlor principles on economy and optimality impose best derivations and representations. A particular language (grammar), L may select a set of optimal operations in accordance with the principles on economical and optimal conditions. Here, let us summarize the two fundamental and important principles proposed by Chomsky 'in the MP as follows:2

(1)

·

The Least Effort Principle Eliminate anything unnecessary in the following cases: (i) superfluous elements in representations ( ii) superfluous steps in derivations

I For the Minimalist Program that we follow here, see Chomsky ( 1 9 95), (2000a) , (2000b), (2004), (2005). Ana for further topics on linguistics and the natural sciences, see Imai (2002). 2 The defmition parts of the principles (1 )-(2) are paraphrased here based on Chomsky (2000a).

276 Takashi Imai

(2)

The Last Resort Principle Avoid crash in derivations as possible.

Those principles are applicable for any operations and processing in biolinguistics, and could eventually be compatib le in natural science. 2. Matching Projections in the Minimalist Program

In the principles-and-parameters approach Haider ( 1 988) investigates the Matching Effect maintaining two premises: no empty categories and no empty derivations. An immediate consequence of Haider's work is now quite compatible with the Minimalist conception in the sense of economy and optimal derivations. The classical matching projection is defined as in (3) : (3)

A matching projection is a projection superimposed on an existing projection such that the nodes of the primary projection serve as secondary nodes of the superimposed proj ection. (Haider 1 988: (35)

The process that Haider proposes is a kind of universal operation like 'Superimpose a projection, P on an existing projection, P' .' See Haider ( 1 988) and Groos and Riemsdijk ( 1 983) for more on the matching effect. Let us propose here that superimposition is now technically called ENFOLD. 3 ENFOLD enfolds two categories in case that the first categorial head is empty. Note that those categories may be restricted to Functional Categories (FC). We will also propose the opposite operation as UNFOLD. NaturalJy, UNFOLD unfolds compressed categories (stacking of categories) to visually show the 'virtual' configurational (vertical) structures in the sense of Zubizarreta and Vergnaud ( 1 982). Some immediate consequences of enfoldQlent w�uld b.e to ·account for multiple SPECs, adjunctions, raising, head m ovement, that-trace effect, structural parallelism between sentences (clauses) and noun phrases. To mention a few, we will consider multiple SPECs. It is not so impossible to regard mUltiple SPECs as stacking of XPs, i.e. maximal projections. Hence, SPECs in multiple SPEC projections must be equidistant in the sense of Equidistance of Chomsky ( 1995). Bohm ( 1 980) inspired me to motivate the use of his terms, which are 'enfold' and 'unfold'.

J

En/aidmen! as Economy 277

Next, let us consider Heads. A complement merges with a head H, whi ch leads to creating a head ascendant structure. (HAS). A question arises. Does maximal projection, XP merge with HAS, creating another maximal projection? There may be two options as follows: 1.

2.

maximal projection merges with a HAS. This case is not an adjunction, but no distance among them (HASs). A maximal projection merges with another maximal projection. This case is an adjunction. A

Let us consider the fol lowing tree: (4)

A

B

� A �

C

D

E

;In (4), A denotes either a maximal projection or HAS. Suppose that E is a ' • head, then, A is a HAS assuming a category E. , Suppose vP and TP. Then, in certain environment, vP and TP can be enfolded. It is tempting to speculate that the difference between English ,,a ', nd French can be uniformly explained by enfoldment. Consider the 4 . ,following: a.

h.

[TP [vp DP-subj VP ] ]

U

hp/vp DP-subj [ vp V OP ]]

In support of the VP internal subject hypothesis, no DP subject raising to SPEC TP is assumed. Instead, the operation, ENFOLD applies to enfold TP -

4

AlB indicates a superimposed matching projection, now called an enfolded structure .

278 Takashi Imai

In support of the VP internal subject hypothesis, no DP subject raising to SPEC-TP is assumed. Instead, the operation, ENFOLD applies to enfold TP and vP here. The outcome is straightforward. Once the enfolded structure is created, then, the verb raises to v, but does not necessarily raise to T in French. On the other hand, the verb similarly raises to v in English, but no Tense affix hopping to the verb is necessary. 3. Subject wll-in-situ in English

Let us now consider the following: Who saw John?

(6 )

As argued in the earlier MP, WH in the subject position appears in-situ, hence, no displacement takes place. We will return to this in section 4 .

[cp C[+Q) [TP who[+Q] saw John ]]

(7 )

Only WH features move to CP-SPEC for feature checking. Hence we get ( 8): [cp Op{+Q] C[+Q] [TP who saw J ohn ] ]

(8)

I

I

feature checking

However, it causes problematic in that it lacks generalization. Other WH words can move to SPEC-CP from non-subject positions while WH in the . subject position appears in situ. Notice however, if one assumes matching projection of CP-TP, then that it is not impossible to assume that virtually the subject WH does not copy/move at all, but does stay in situ. Practically the subject WH proves to be in SPEC-CP as follows: (9)

a. b.

V, see is raised to v-head from V and as noted in section 2, if it is in fact the case that vP, TP and CP are enfolded in a triple way, then, see in an appropriate form is in CIT/v, manifesting a V2 phenomenon weakly in

En/oldment as Economy 279

English. Note that English is a defective V2 language. Furthermore, the analysis may extend to wh-in-situ languages. Consider the following Japanese examples: Taroo-ga nani-o katta ka (=no)? Taroo-NOM what-ACC bought Q ' What did Taro buy?' b. Dare-ga sono hon-o katta ka (=no) who-NOM hat book-ACC bought Q ' Who bought that book?'

( 1 0) a.

Note that Japanese is an SOY language and wh-in-situ. ka is traditionally regarded as a Q marker, which is a scope marker of interrogation of WHo Note that no is its colloquial form. It is not so impossible to say that ka is a visible counterpart of the invisible feature of Q in C-head . In other words, it is pronounced at PHON. Morphologically, ka is an enclitic, which is not independently used. Now, let us assume the status of Clitics as follows: (11)

Clitics may only be visible at PHON in the paral lel structure of morphology .

Only feature(s) of Clitics may exist in narrow syntax. A node with a bundle of features can project its projection. In Japanese, CP, TP and vP are stacked by ENFOLD. Consequently we will postulate here that WH replacement is not involved in wh-in-situ languages such as Japanese, Korean, Chinese among others in any sense: no feature movement at LF nor syntactic movement.s With the revival of the classical matching effect as a new operation: ENFOLD, the projections of empty head may be enfolded into the actual projection derived by the operation of Merge. 4. V2

.

and Matching Projections

In Germanic languages, V2 is respected as a default setting even in English. In the previous section, it is argued whether WH in the subject position was in situ or not. Consider again:

S

See Watanabe ( 1992) on Japanese syntactic movement of WH among others.

280

(12)

Takru'hi Imai Who saw John?

As noted earlier, this sentence seems superficially to be a wh-in-situ case. Chomsky (2005) posits that WH is extracted to the edge of vP, and then further on to the edge of CP, regarding CP and vP as phases. For the sake of simplicity of expository purposes, we will use vP instead ofv*P, which is Chomsky's term. The edge-feature as well as the Agree-feature in C­ head can trigger WH movement. T-head, to which the Agree-feature is inherited from C-head, raises WH to SPEC-TP from the edge of v P. Then, the edge-feature of C-head attracts WH to SPEC-CP from the edge of TP. We will however propose that sentence (12) is a matching effect structure. TP and vP are enfolded by ENFOLD, then, CP and TP are further enfolded. Consequently, CP-TP-vP can be enfolded. In other words, who is in the edge of vP, ofTP, and ofCP. The two features ofC, i.e. the Agree- and the edge-features, directly seek the goal of WH, who in SPEC-vP in the enfolded structure as ( 13).

(13)

[cPITP/vP who [c'rr'/v' C/T/v [vp wRe see John]]]

Verb see in an appropriate superficial form is a head of v, T and C. Enfolding of categories is a more economical option in the sense of effective derivation.Consider: (14)a. *Whom did who see? b. Who saw whom? In multiple WH questions, ENFOLD applies to fold CP, TP, and vP. Thus, the structure (14b) is roughly represented as follows: (15)

[cprrP/vp who [c'rr'/v' ClTlv [vp wOO see whom]]],

�

'

.

�r

' . ' j

' " In a classical analysis in the pa � ; ungrammati �al . (l 4 iesorts to the violation of the Superiority among others. But, it can be ruled out by virtue of the Matching projection violation. It is impossible for Acc. WH to move/copy, since CP and TP are already enfolded, thus, if Acc WH moves to SPEC-CP/TP, this move is inadequate, because Ace. WH is incorrectly

Enfoldment as Economy 281

raised to· SPEC-TP as well as SPEC-CP. Hence, whom cannot find the landing site of its own.6 (16)

[CPffP who [C'ff' did [who see whom]]] t

X

Economy of derivation is a priority criterion: If any operation whatever is more economical than other ones, it applies first, which is a default setting of universal operations. It is suggested that ENFOLD should apply at

SPELL-OUT.

5. Concluding Remarks What we have seen thus far is how the matching effect tells us about the optimal and economical consideration along the line of minimalist program. An immediate consequence of ENFOLD in matching projections would derive straightforwardly from accounting for multiple SPECs, adjunctions, raIsmg, head movement, that-trace effect, structural parallelism between sentences and noun phrases among others which we will leave here for further investigation. References

Bohm, David. 1980. Wholeness and the implicate order. London: Routledge and Kegan Paul. Chomsky, Noam. 1 995. The minimalist program. Cambridge. Mass.: MIT Press. Chomsky, Noam. 2000a. Minimalist inquiries: The framework. In Step by step: Essays on minimalist syntax in honor ofHoward Lasnik, ed. by Roger Martin, David Michaels, and Juan Uriagereka, 89-155. Cambridge, Mass.: MIT Press. Chomsky, Noam. 2000b. Derivation by phase. In Researching and verifying an advanced theory of human language: Explanation of the human faculty for constructing and computing sentences on the basis of lexical conceptual features. ed. by Kazuko Inoue, 1 -45. Chiba: Kanda University of International Studies. Also In (200 1) Ken Hale: A life in language, ed. by Michael

Kenstowicz, 1-52. Cambridge, Mass.: MIT Press. 6

Note that since CP and TP can be enfolded in a matching situation, no head movement of did takes place from T to C in (16). Notice however that TP and vP cannot be enfolded because of the lexical item, did appeared in T-head. Thus, a structure as (14a) is never generated at the outset in the matching structure. I am ignoring some technical possibilities here.

282 Takashi Imai

Chomsky, Noam. 2004. Beyond explanatory adequacy. In The cartography of syntactic structures. Vol. 3, Structures and beyond ed. by Adriana Belletti, 104- 1 3 1 . Oxford: Oxford University Press. Chomksy, Noam. 2005. On phases, Ms., MIT, Cambridge, Mass. Groos, Anneke and Henk van Riemsdijk. 1 979. Matching effects in free relatives. In Theory of markedness in generative grammar. Proceedings of the 1979 GLOW conference, ed. by Adriana Belletti, Luciana Brandi, and Luigi Rizzi, 1 7 1 -2 1 6. Pisa: Scuola Normale Superiore. Haider, Hubert. 1 988. Matching projections. In Constituent structure. Papers from the 1987 GLOW conference, ed. by Anna Cardinaletti, Guglielmo Cinque, and Giuliana Giusti, 1 0 1 - 1 2 1 . Annali di Ca' Foscari, XXVII, 4. Imai, Takashi. 2002. Toward an optimal theory in generative grammar. In Working papers on minimalism III --A basic scientific research into modularity of language faculty, ed. by Yukio Takahashi, 1 -9. Tsuru, Yamanashi: Graduate School, Tsuru University. Lemons, Don S. 1 997. Perfect form---Variational principles, methods, and applications in elementary physics. Princeton, NJ: Princeton University Press. Watanabe, Akira. 1992. Subjacency and s-structure movement of WH-in-situ. Journal ofEast Asian Linguistics 1, 255-291 . Zubizarreta, Maria-Luisa and Jean-Roger Vergnaud. 1 982. On virtual categories. In Papers in syntax, ed. by Alec Marantz and Tim Stowell, 293-303 . MIT Working Papers in Linguistics 4. Cambridge, Mass.: MIT, Department of Linguistics and Philosophy, MITWPL.

"GP, I'll have to put your flat feet on the " ground Jon athan Kaye

Government Phonology (OP) in its various flavours has been characterised by the absence of any level of phonological structure that could be called 'phonetic'. Since the theory of phonological representations which forms part of GP consists of a small set of privative elements, it follows that at any point in a phonological derivation, the resulting phonological string is immediately interpretable. Put another way, there is no more detail relating to ultimate realisation present in the output of a phonological derivation than there is in its input. The input string is as interpretable as the output string; they are just interpreted as different things. There is no sense in which the output strings are somehow closer to a physical manifestation than are the input strings. In fact they are equally close or far away. Phonological derivations simply change one well-formed and complete phonological string to another. A second attribute that characterises GP is that it views phonology as entirely autonomous from 'phonetics', in the usually understood sense of this term. Concretely, GP (or at least the version that I practise) explicitly rejects the commonly held but unsupported and highly implausible view that "[the] features that figure in discussions of phonetics and phonology are from this point of view instructions for specific actions of an articulator." (Halle, Vaux, and Wolfe 1 998:2). As we shall see, there are compelling reasons to reject this assumption. Finally, it should be noted that GP assumes the following, called the Phonological epistemological principle, to wit: the only source of phonological knowledge is phonological behaviour. Thus, phonetics (still 'in its normal sense) plays no role in the postulation of phonological objects nor the interaction of such objects. With all this in mind, it is reasonable to ask how GP deals with the grounding problem. The grounding problem may be expressed in this way. :It is indeed a fact that humans communicate with each other and much of 'this communication involves the production and transmission of an audible . :$jgnal on the part of the speaker and the reception and interpretation of that signal on the part of the hearer. The aim of this article is to present a view of how this might be done, and still be consistent with the views of GP

284 Jonathan Kaye

expressed in this section. I must begin this quest with a brief look at what exactly is transmitted from mouth to ear - the nature of what is (misleadingly) called 'the speech wave' . The very name 'speech wave' is a major misnomer. It implies that its entire content is devoted to the transmission of linguistically relevant information. If such were the case then characterising phonological objects as sets of instructions to the articulatory apparatus would make some sense. If we move our tongue in such-and-such a way, then this will obviously affect the sound emitted and consequently change the linguistically relevant infonnation contained in the sound wave. But nothing could be further from the truth. There is, to be sure, linguistic content in the sound waves we produce when we speak, but this only only a part - perhaps a small part of what they contain. I would suggest that a large part of the so-called speech wave is devoted to dispersing human group-recognition signals: human oral pheromones, if you will, commonly known as ' accent' . Consider this case. I lived and worked in London from 1988 to 1 999. English of a North American variety is my native language. I consciously strove to learn and use London lexical forms to the extent they differed from my native ones. Never in the 1 1 years that I lived there did anyone from Britain ever mistake me for a native Londoner (although I was sometimes mistaken for Irish, in short conversions). Of course these observations were entirely accurate and entirely characteristic of a universal human ability: to recognise con-specifics of the same group. Being social primates it is hardly surprising that humans have this ability. And how are these group membership signals communicated? We can eliminate the senses of taste, touch, sight and smell since accurate identification can be done over the telephone. Clearly these signals are transmitted by sound as part of the so-called speech wave. We see then that our oral emissions contain not only linguistic content but also signals as to our group identity. To these we need to add cues that allow for individual identification (my friends recognise my voice over the phone) and transmission of emotional states. All these, and perhaps more, share the same speech wave and are produced by the same vocal apparatus. With these facts in mind, it becomes apparent that the claim that phonological objects are sets of instructions to the articulators is untenable. One brief example should suffice to bring home this fact. In Quebec French there are two ways of pronouncing an "r" type sound as in words like rouge [red] or jreiner [to brake]. This "r" can be manifested as an apical trill similar to the Spanish one or as a uvular similar to the standard European French variety. The articulatory gestures involved in the production of each

·"A'

']

"GP, I'll have to put your flat feet on the ground" 285

of these "r" types are quite different. As Michel Prairie has shown in . numerous articles on the subject, the selection of one or the other production method is tied to a battery of variables. The trilled variety tends to be used by older, rural and less wel1-off speakers whilst the uvular version is more often found amongst younger, urban and better-off speakers. From a phonological point of view however, the two are identical. Their distributions are the same. They trigger or undergo the �ame phonological events. Yet, according to the standard view, phonological representations are to be seen as a set of instructions to the . articulators. The articulations of the two Quebec "r"s are quite distinct. It . follows that their phonological representations ought to be equally distinct and that this distinction should somehow show up in their phonological behaviour. All this is, of course, quite false. It is also difficult to reconcile the claim that phonological representations are sets of instructions to the articulators with the observed . behaviour of the congenitally deaf. If articulatory instructions were 'hard­ . ' into the linguistic component of the brain we would not expect the extreme difficulty that the non-hearing have in oral speech production. The muscular development of the deaf is unaffected and it would seem that this "'::�ll'f" l'r.n latory approach would not predict that the deaf would have difficulty implementing such a set of instructions to the articulators. Finally, it is worth noting that linguistically identical utterances can be using significantly different articulatory gestures. This can balJpe�n as a result of some trauma to the articulatory anatomy (a broken simultaneous use for other purposes of the oral cavity (eating) or uther human activities purposely distorting normal articulation (ventriloquism). .. .. From the above it should be obvious that phonological grounding is �coustical1y and not articulatorily based. Phonological objects such as elements or constituents (nuclei, onsets) are associated with acoustic . ignatures which are to be found somewhere in the signal. Experience '�::Jf�aCJl1es us that these signatures are quite robust, capable of withstanding :consldel�able degrees of ambient noise, distortion, and a variety of other :;i]Jiotentilally disruptive factors. They are of such a nature as to not interfere each other or with the other systems with which they share the oral ':�l1lan:nel (group-recognition, personal identity signatures, emotional states, . Clearly articulatory strategies cannot be realistically discussed while one of more of these components of the oral channel. Whatever .. :i;Pte component of the human brain that translates linguistic information into "articulatory gestures is, it is clearly outside the linguistic component since it

286 Jonathan Kaye

depends crucially on non-linguistic infonnation originating in other components. Let' s call this the 'production centre' . This component receives infonnation from a variety of dynamic sources - sources whose content may vary from utterance to utterance. The linguistic component and the emotional component would be examples of this. There are also static sources for the production centre, notably group signs and personal 10. Note that these can be adjusted as in cases where one changes languages or imitates a different accent or person. What I have in mind looks something like this.

Fig. l

The cauldron represents the oral production centre. Amongst other ingredients it receives the ,linguistic content of the, utterance under construction defined in terms of a number of acoustic images. Whatever the ultimate outcome of this productio�, the product must includ � .the�e images and include them in the appropriate temporal sequence·to ' aSs ure that the emission 's linguistic content can be decoded by the receiver of the emission. Of course the workings of this production centre, along with the nature of the acoustic signatures of the various components of the emission, remain unknown to us. This is hardly surprising with respect to the linguistic aspects of the signal given the assumption (tacit or explicit) that the entire signal consists of pure linguistic infonnation. Using phonetically­ based classificatory procedures as this basis for delving into these problems

"GP, I'll have to put yourflat feet on the ground" 287

h as h ad no success . The hulk of res earch in what sh ould be 'applied linguist ics ' in areas like automat ic speech recognit ion relies on stat istical rath er th an l inguist ic approaches . Phonol ogy sees very few serious attempts at s eparat ing the wheat from the chaff in tenns of its own areas of compet ence. If clicks are encountered in a variety of languages in southern Africa t hey a re incorporated into phonological theory [+/- glottal s uct ion] wit h discuss ions about the ingress ive velaric air st ream mechanis m but with litt le or no underst anding of their phonolog ical behaviour. It seems reas onable t o s uppos e that clicks will turn out t o be familiar hous ehold ty pe phonological objects pronounced in a funny way, their 'clickiness ' being non-l inguis tic in nature. In cannot be repeat ed too often that the 'speech wave' carries much informat ion t hat is non-linguist ic. We have s een that above in th e case of the two Quebec " r"s . Phonological cues in the form of acoust ic s ignat ures are pres ent in the s ignal to be sure. They are, h owever, padded out with much that is non-linguist ic. Ident ical linguistic object s may appear to be quite different but their phonolog ical behaviour bet rays their underlying s imilarity. Another interest ing case of th e confusion of linguist ic with non­ linguist ic distinct ions involves ' implos ives '. In my phonet ics classes I learn ed all there was t o know about their manner of product ion. I learned . they were produced us ing the ingress ive glottalic air stream mechanis m. A closure was made (t ypically at t he lips) in t he supraglottal cavity. The glott is was closed and lowered caus ing a reduct ion in air pressure in t his cavity below t hat of th e at mosph eric press ure. Cons equent ly , when the c losure in the s upraglottal cavity was released, air rushed into the mouth giving rise to an implos ive. Years after learn ing all this I h ad my first real experience in fieldwork on a language that actually had these st range o bjects . I quickly realised that all th is· phonet ic knowledge gave me abs olut ely no clue as to what kind of animal they were - what kind of phonological beh aviour t hey would dis play . Working on Vata, a Kru language of t he Ivory Coast, I had a chance to have an up close and · · personal look at these beasts . Vata is a t one lang uage and displays t he us ual ga mut of tonal s preading behaviour. I learned t hat certain consonants blocked t onal s preading (opaque consonants) wh ilst others didn't. I can us e the lett ers b, d, j, g , v and z to des cribe th e s et of opaque consonants . Other cons ona nts l ike w, y, I, m, n, ny, ng did not have this effect . Wh at about my implos ives . Which g roup did t hey belong to? I realis ed t hat all my lm owledge of phonetics did not help me one iota to ans wer this quest ion. It turns out th at 6, the labial implos ive and t he only one that Vata poss ess es, behaved like th e non-opaque group. Tone happily spread t hrough it . Even

288 Jonathan Kaye

more revealing was the fact that in related western Kru languages such as Wobe and Ouere, an implosive a appeared where "1 " was found in related languages. This funny a had the same distribution as "I" did elsewhere and possessed all the same phonological properties. The difference between this type of implosive cf (there are other types but I won't go into that here) and "I" seemed to be purely cosmetic and without any linguistic significance. Again implosive a and "1" contain the same linguistic content but their identity is masked by non-linguistic (group signs) factors. The claim is that when our knowledge of the precise nature of these acoustic signatures deepens, the same set will be found in these seemingly different objects. I illustrate what I mean in Fig. 2 below:

o

I

x

Fig. 2 The two Quebec "r "s Just as two asplnn tablets from different manufacturers may differ superficially in shape, colour, texture and so on yet have the same pharmacological effects, so phonological objects may appear to be different , . 'phonetically and yet contain exactly the same 'active ingredients' and be identical from a linguistic point of v iew. As my own experience with implosives shows, knowledge of the methods of production of linguistic objects offers no hint as to their status. Failure to understand that the 80called speech wave contains' much more than linguistic infonnation has led to the sorry state of phonology that we see today � A. few minutes. of reflection on the nature of this speech wave would, suffice to' show that phonological objects could not possibly be defined as sets of instructions to articulators and perhaps help phonology escape from the abyss into which it has fallen. References

Halle, Morris, Bert Vaux, and Andrew Wolfe. 1998. On feature spreading and the representation of place of articulation. MS. MIT.

On parameters and on principles of pronunciation Richard S. Kayne

1. On parameters

The point thatChomsky ( 1 995, 1 60) makes concerning Case chains can be put as follows: I

(1)

Languages themselves are never the immediate locus of parametric variation.

Rather, parametric variation across languages/grammars is to be thought of in terms of varying features/properties of corresponding items of the lexicons of the languages in question. A familiar question is whether the locus of parametric variation in syntax is always restricted to functional as opposed to lexical items. Such a . restriction seems plausible, although there are some potential counterexamples worth considering. One has to do with nominal gender. For example, French and Italian have strongly similar gender systems, yet the word for sea is feminine in French (la mer) and masculine in Italian (i/ mare) . This looks like a parametric difference associated with mer/mare, which is a lexical rather than functional item. Alternatively, though, it is feminine gender itself (an item of the . lexicon - cf. Ferrari (2005» that has slightly different properties in French VS. Italian. One common property of feminine gender in the !\yo languages 'is that, to the (partial) extent that it is subject to arbitrariness, it can be 'associated with a list indicating which lexical items (including suffixes :Williams ( 1 98 1 ) - which reduces the size of the list considerably) have feminine gender in unpredictable fashion. In French this list (arbitrarily) contains mer, whereas in Italian it does not contain the corresponding mare. In this way we can think of the mer/mare difference as being associated . most directly with a functional item (feminine gender).

:2 . This paper corresponds by and large to a talk presented at the Structure of " Parametric Variation conference in Newcastle in September, 2005 and to material · . presented in class lectures at the 2005 LSA Summer Institute at M.I.T.

290 Richard S. Kayne

This parametric property of feminine gender may look complex, in that it consists of a long (but finite) list of nouns. But this is counterbalanced by the fact that looking at things this way avoids having to have a large number of feminine nouns associated with a gender parameter directly. From this perspective, feminine gender can be merged with a noun only if (in the unpredictable cases) that (suffixal) noun is contained in its list, in essence a selectional property of feminine gender. (Feminine gender via agreement, on adjectives, determiners, etc. will involve something more.) A second potential counterexample to the claim that parametric variation is limited to functional items comes from prepositional complementizers of the sort found in French and Italian. In both languages, infinitives are often preceded by no overt preppsitional complementizer (the contexts are not identical in the two languages). In both, prepositional complementizers frequently do appear (again, the contexts are not identical). The prepositional complementizer is usually de in French and di in Italian. Sometimes it is a in French and a in Italian. A m inimal difference is found with their counterparts of try: essayer de... in French vs. provare a in Ital ian. If French essayer differed from Italian provare in being parameterized for taking de rather than for a (and the reverse for provare) we would be associating a parameter with a non-functional item (assuming try to be lexical). Alternatively, it might be that at least the prepositional complementizer a/a is associated with a list of (subject and object control) verbs that it selects for and that differs somewhat in French vs. Italian. In Italian provare is on the list associated with a; in French essayer is not on the list associated with il. An apparent problem is that a seems to be in the wrong place relative to the verb, if a is to be the selector (parallel to feminine gender). The problem would dissolve if prepositional complementizers are merged outside VP, as in Kayne (2005, chapters 5, 7 and 9). (Strictly speaking, the relation between prepositional complementizer and verb (or VP headed by it) will be mediated by a K(ase) morpheme.) ...

2. Intra-language parametric variation

A possible objection to the preceding is that within Italian there are actually two verbs for try (cf. try and attempt in English) and they differ from each other in choice of prepositional complementizer:

On parameters and on principles ofpronunciation 29 1

(2)

(3)

Gianni ha provato a cantare. 'G has tried a singinf ' Granni ha tentato di can tare. ' ... attempted di...'

The earlier difference between French essayer de... and Italian provare a.. . might then appear misleading, since Italian tentare di... seems to show that what's at issue is not really a difference between French and Italian at all. The correct reply to that objection, I think, is that that's exactly what is expected, given ( 1 ). Languages are not the locus of parametric variation. Italian a contains in its verbal selection list provare but not tentare, an example of intra-Italian parametric variation. Another example of the same general sort comes from French vs. English: (4) (5 )

Jean est assez grand pour.. . 'J is enough big for/to . .. ' John is big enough to ...

Alone among degree words in Engl ish, enough follows the associated adjective, as shown by (5 ) as opposed to (7) and (8 ). (6)

(7) (8)

*John is enough big to ... John is too/so/as/how big. *John is big too/so/as/how.

In French assez patterns regularly with the other degree words: (9)

Jean est trop/si/aussi grand.

The word order difference between (4) and (5) can be thought of parametrically in movement terms. In English the adjective must move higher past enough (cf. in part Bresnan ( 1 973, 285 ) and lackendoff (1 977, 15 1» . No comparable movement takes place in (standard) French. But again (as with prepositional complementizers), this is not just a FrenchlEnglish difference, since we see something similar entirely within English, insofar as the other English degree words don't share the property of enough that induces movement of the adjective. And again there is nothing unexpected here. Given ( 1 ), iCs not languages that differ parametrically, at bottom, but rather particular elements of the lexicon (arguably only functional elements) that differ parametrically. So there' s no reason at all not to expect that in many languages there will be cases in which similar elements within that

292 Richard S. Kayne

language have differing properties (such as enough vs. the other degree elements in English). Another way to put this is ( 1 0). A stronger formulation (that seems plausible) is ( 1 1 ). (10)

(1 1)

Parametric variation occurs within languages as well as across languages. The parametric variation that occurs within languages is of exactly the same sort as the parametric variation that occurs across languages. The elements subject to it are the same in both kinds of cases, and the features/properties in question are, too.

3. Back to lexical vs. functional

In addition to gender and prepositional complementizers, the count/mass distinction appears to involve parametric variation associated with lexical rather than functional items. For example, English and French differ with respect to grape: (12)

( 1 3)

Give us some grapes/*grape. Donne-nous du raisin/*des raisins.

In English grape normally has count noun properties and can be used as a mass noun only in special contexts. In French raisin is on the contrary normally a mass noun. (It can be used as a count noun in special cases.) This looks like a parametric difference associated with lexical items. Alternatively, we might try to relate this kind of variation to ( 1 4) and (15 ) by taking lexical nouns to always be accompanied by either number or amount or (cf. Kayne (2005, chapter 8» unpronounced NUMBER or AMOUNT, depending on cc:mnt vs. mass. (14) (15 )

John has a large number(*amount of friends. John has a large amountl*number of moriey : : : :

"

" ',

""

If we take nouns like number and amount and their unpronounced counterparts to be functional, not lexical, elements (cf. the notion 'semi­ lexical' in Corver and Van Riemsdijk (200 1», then we can say that these functional nouns select for lexical nouns (again, in the unpredictable cases in terms of a list - the selection configuration needs to be made precise). English grape and French raisin differ with respect to which list they are on. If so, we are looking at a property of functional nouns (and only indirectly at a property of lexical nouns). English NUMBER/number

On parameters and on principles a/pronunc iation 293

selects for grape, French AMOUNT/quantile selects for raisin (the relevance of markedness needs to be worked out). Consider now the case of missing lexical items, of the sort illustrated English shallow having no direct counterpart in French. This might be by th ought of as a parametric diff erence centered on a lexical (as opposed to functional) item (realizable in one language but not in another). But that would appear to leave out the fact that French can readi ly express what English expresses using shallow, namely with peu profond ' lit tle deep'. Th e alternative th at suggests its elf is that shallow in English is necessarily accompanied by LIITLE BIT: t

!,

( 16 )

! I ! �,l . !

!�

��, i

t.. I i =

i

I .1,

i, �'

I�

�..

r';;:

l,:'ii:l;'

l�:, >

I"

i

�'. .'

Th is lake is LITTLE BIT shallow.

LITTLE BIT 'selects' for certain adjectives, including shallow, in English . Th e same is true in French, except that French has to pronounce BIT (as peu - cf. Kayne (2005 , section 12 .4». Thus this FrenchlE nglish difference may be rethinkable as a property of LITTLE and/or BIT, both arguably functional rather than lexical. A cas e that may lie at the intersection of shallow and grape is one pointed out (p.c.) by Peter Svenonius, who noted that various languages presumably lack a direct counterpart of reindeer. In fact French seems to lack a word corresponding to English moose. Thinking of the fact that the plural of moose is moose and th at such a zero plural is widespread in E nglish with names of animals, it may be that English has an un pronounced .c lass ifier for (large) animals that is associated with a selection list that contains moose. French would have a comparable classifier wh ose selection list contains no parallel item, so the parametric difference would be associated more d irectly with the animal c lassifier than with moose itself. There is some similarity between this case and the Hale and Keyser (2002) discussion of laugh in E nglish vs. Basque, which expresses laugh as overt light verb plus noun. The expectation would be that th is difference could be localized as a property of the light verb itself, which h as to be pronounced in Basque vs. English and which would count as functional, rather than as a property of lexical laugh. (In this case and in general, it is ' essential that the properties associated with parameters be limited in complexity .) As a final case bearing on th e functional vs. lexical question, consider id ioms, wh ich seem to crucially involve, in at least some cases, arbitrary ., ,choices of lexical items Gudgrn ents given for the idiomatic sense), e.g.: '

294 Richard S. Kayne

( 1 7) ( 1 8)

He just k icked the bucketl *pail. They' re going to have to close up shop/*store .

It remains to be seen to what extent these are really arbitrary choices that wou ld have to be directly associated with the lexical items in question. For example, shop vs. store might be related to ( 1 9) and bucket vs. pail to (20 ). (19) (20 )

They're going shoppingl *storing. You 've been spending ?buckets/*pails of money these days.

4. More on

enough

T he diff erence within English between enough and the other degree words seen in (4)-(8) is paralleled by a well-k nown difference concerning cooccurrence with much: (2 1 ) (22)

They have too/so/how/as much money ... They have *enough much/*much enough money ...

W ith enough in (23), we have no visible much, which is not possible in E nglish with the other degree words in (24). (23 ) (24)

They have enough money. *They have too/so/how/as money.

In present terms, the natural proposal is that (23) contains an unpronounced MUCH (cf . Jackendoff ( 1 977, 1 52» . But that by its elf leaves us with a curiou s coincidence. Enough is the one degree word to occur with u npronounced MUCH rather than with pronounced much, and also the one degree word that imposes the order 'adjective + degree word', as in smart enough (vs. *smart too, etc.). A step toward linking these two properties is to say (25) and (26 ). Therefore the structure of (23) is as in (27). (25) (26 ) (27)

MUCH in (23), lik e overt much, is an adjective (cf. the related little and also fewlfewerlfewest - Kayne (2005, chapter 8» . Like other adjectives, MUCH precedes enough. ... MUCH enough money.

We can now account for the impossibility of (24) if the following two statements hold:

On parameters and on principles ofpronunciation 295

(28 ) (29)

MUCH comes t o precede enough as a side effect of adject ives in general coming to precede enough, in which case we would expect MUCH not to be able to precede too/so/as/how. MUCH is legit imate only if it comes to preced e the degree word .

In ot her words, (24) is unavailable in part because (30) is unavailable (due to t he absence of adj ect ive preposing across degree words ot her t han enough) and in part because by (29) MUCH is not legit imate in (3 1). (30) (31)

* MUCH too/so/as/how money * . too/so/as/how MUCH money. ...

..

5. On Principles of Pronunciation The quest ion now is what dist inguishes (31) and (27), i.e. why does MUCH need t o prepose? This need s to be related to a broader quest ion: (3 2)

Why s hould ua ever al low element s t hat can perfect ly well be pronounced (e.g. as much) t o somet imes be unpronounced?

T he answer to this broader quest ion is, I th ink, t hat t he existence of non-pronunciation (at least of the MUCH type, but arguably more broadly, as I will suggest) is an automat ic consequence of t he arch it ect ure of derivat ions. (R izzi (2005) independent ly has a part ially similar idea for the cas e of t he h igh portion of root cont ext s.) Assume t hat t he fol lowing holds: (33 ) a. At a given phase level, only t he head and mat erial in t he c­ command domain oft he head can (and must ) be spelled out . At a given phase level, no material wit hin (or adjoined to) a lower b. phase 'can be spelled out . I f so, t hen spellout systemat ically and aut omat ically ' fails to see' phrases in the Spec of a phase (similarly for adjunct ion, if t hat is d ist inct). Any pronounced Spec will of necessity now be t he Spec of a non-phase. (For example, an overt wh -phrase must be in t he Spec of a non-phase.) Assume furt her: (34)

(33 ) is t he only source of non-pronunciat ion (at least of elements that have a pot ent ial pronounciat ion).

296 Richard S. Kayne

Then we can say that (31) is excluded because MUCH/much did not reach t he Spec of any phase. In (27), on the ot her hand, MUCH/much arguably has reached the Spe c of some phase and t herefore can (and must) be unpronounced. (In big enough, the lexical adject ive big must not end up in t he same posit ion as MUCH/much. Note that from t he present perspect ive much and MUCH are identical, apart fr om landing site.) The quest ion ar ises as to t he status of unpronounced copies in a movement chain, a question that might be closely r elated to the pr esent discussion. Assume that successive cyclic movement passes through the Spec of a phase (alternatively, via adj unction). The corr esponding copy is typically unpronounced. This follows immediately from (33). (Apparent cases of pr onunciation of an intennediate or lower copy might be hidden cases of doubling.) Somewhat diff erent is t he quest ion of the typical non-pronunciation of t he lowest copy in a chain. We cou ld bring t his case into the present framewor k by claiming: (35)

Moved phrases must or iginate in the Spe c of a phase.

(And by (33) t hey must not end up in t he Spec of a phase, if they are pronounced.) I f (35) holds, then there is no longer a need t o st ipulate t hat only t he highest copy of a chain is (typically) pronou nced. I am assuming t hat there is no covert movement of the LF type (if only because ther e is no LF component - cf. Chomsky (2005». The question also arises as to whether t here can be movement from Spec,non-phase t o a c-commanding Spe c,phase. Long movement of t hat type may be excluded for PIC reasons. Whether short movement oft hat type e xists is left an open quest ion. 6. Further types of non-pronunciation

We also have cases of non-pronunciation related to the presence of a non-c­ commanding ant ecedent , as e.g. with VP-deletion. The non-pronunciation of the VP will fol low from (33)/(34) if that VP is the Spec of some phase (or phases, if it moves). Note in passing t hat Chomsky's (1995, 203) i dea of linking VP­ delet ion direct ly t o VP-destr essing does not cover the whole range of cases: (36) (37)

JOHN didn't break the window, YOU did (*break t he window). Who br oke the window? HE did (*break the window).

On parameters and on principles ofpronunciation 297

In these, a destressed in situ V P is impossible. French allows unpronounced objects of certain prepositions in a way that English does not. Alongside (3 8) one has (39). On a tire sur lui.

(39)

'one/they have shot on him' On a tire dessus. , ., .shot of-on'

By (33)/(34)/(35), the unpronounced object must be in the Spec of a phase, the head of which might be the de whose appearance needed explaining. (Alternatively, de itself might head a phase through whose Spec the unpronounced object has passed. In some cases there may be an unpronounced counterpart of de.) What about unpronounced heads? If the unpronounced head is the head of a projection that contains no pronounced material at all, we can speak of an unpronounced phrase, and fall back into the previous discussion. What, then, of an unpronounced head whose maximal projection does contain pronounced material (in complement or in specifier position)? A familiar apparent such case is pseudo-gappingIVP-subdeletion, as in: He praises you more often than he does us. An idea that goes back to work by Jayaseelan ( 1 985) is that the stranded . phrase us in (40) must have been moved out of the VP prior to 'deletion'. But why couldn't the verb alone have been 'deleted' under identity with the matrix verb? Assume as earlier that (3 3) is the sole source of non-pronunciation (at least for elements that have a possible pronunciation). Then the verb alone could not possibly be unpronounced, since head positions themselves (whether of phases or non-phases) will never fail to be seen by the operation of spellout. Spellout will fail to see V in (40) if V is within a phasal Spec, in which case pronounced us must not be, i.e. must have . moved out (to the Spec of some non-phase), which is what we needed to show. Note that the 'must' of (33a) excludes classical head movement. Head movement that would leave a pronounced copy behind might in tum be : excluded if the doubling alluded to just above (35) in incompatible with : head positions - cf. Kayne (1994, 59) on coordination.

298 Richard S. Kayne 7. Back again to

enoug/z

The deviance of (4 1 ) can now be attributed to (33) (assuming the appropriate landing site(s) for muchIMUCH). As is well-known, (41) contrasts with (42), which in tucn correlates with the fact that (23) cannot have the interpretation of (42). Put another way, little must be pronounced, unlike much, even when preposed to enough . Rather than an infonnal notion of 'recoverability' , it might be that (42) contains an unpronounced negation (that must be licensed by overt little - cf. the earlier discussion of shallow) and that the presence of that negation prevents littlelLITTLE from reaching the Spec of a phase. (4 1 ) (42)

*10hn has much enough money. (1)10hn has little enough money.

8. Conclusion

I. The restriction of parametric variation to functional elements may be tenable despite a number of apparent counterexamples. II. The pronunciation or not of material that has a potential pronunciation may be entirely and automatically regulated by principles (33)-(34)-(35) that make reference to phases and to spellout; spellout will have a 'blind spot' for material in the Spec of a phase. (It remains to be determined how closely the notion of phase needed here matches the notion of phase needed for movement and the PIC (and what exactly determines what categories count as phases).) If (33)-(34)-(35) are correct, then ua has no need for an additional operation of deletion or ellipsis. References

Con struction in Engli sh , Bresnan, Joan. 1973. Sy ntax · of the Comparative Clause . linguistic Inquiry, 4, 275-343. . ... Chomsky, Noam 1995. The Minimalist Program, The MIT Press, Cambridge, Mass. Chomsky, Noam. 2005. Three Factors in Language Design, Linguistic Inquiry, 36, 1-22.

Corver, Norbert, and Henk van Riemsdijk, eds. 2001. Semi-lexical Categories, Mouton de Gruyter, Berlin. Ferrari. Franca 2005 . A Syntactic Analysis of the Nominal Systems of Italian and Luganda. How Nouns Can Be Formed in the Syntax, Ph.D dissertation, New York University.

On parameters and on principles ofpronunciation 299 Hale, Kenneth and Samual Jay Keyser 2002 Prolegomenon to a Theory of Argument Structure, MIT Press, Cambridge, Mass. Jacken do ff, Ray 1 977. X' Syntax: A Study of Phrase Structure. MIT Pres s, Cambridge, Mass. Jayas ee lan , K.A. 1 985. Incomplete Deletion of VP's, Compared Constituents and Clauses, ms., Central Institute of English and Fore ign Languages, Hyderabad

(India).

Kayne, Richard. 1 994. The Antijymmetry ofSyntax, MIT Press, Cambridge, Mass. Kayne, Richard. 2005. Movement and Silence, Oxford University Press, New York. Rizzi, Luigi. 2005. Grammatically-Based Target-Inconsistencies in Child Language, ms., University of Siena. Williams, Edwin. 1 98 1 . On the Notions Lexically Related and Head of a Word, Linguistic Inquiry, 12, 245-274.

What to do with th ose

fools of a crew?

Evelien Keizer

1. Introduction

Topic of this paper are binominal constructions with a demonstrative determiner, as illustrated in example (1 ) : ( 1 ) a. that idiot of a boy b. this fool of a filmmaker c. those crooks of attorneys (www.geocities.com/got_mike2004/susiepart3 .html) d. these darlings of western bankers (www .iacenter.orglwarcrime/wcbecker.htm) It is generally assumed that one of the nominal elements in these constructions functions as the head, with the other functioning as a modifier. Unfortunately, there is no consensus among linguists about which noun performs which function. Various tests, both semantic and syntactic, have been proposed to determine headedness, but the result has been far from unequivocal. After a brief introduction into the issue of headedness in binominal NPs in Section 2, this paper will focus on the role of the first determiner, in particular the demons�ative determiner, in establishing which of the two nominal elements functions as the head in these constructions (Section 3). Section 4 subsequently offers a functional analysis of these constructions. Some conclusions are presented in Section 5. 2. Headedness in binominal NPs

According to Quirk et al. (1985: 1284-1285), constructions of this kind allow for a paraphrase in the form of a copular sentence, i.e. with a predicate relation between the two nouns (a fool of a policeman: the

policeman is a fool; her nitwit of a husband: her husband is a nitwit).'

1 Bennis et al. (1998), dealing with the Dutch binominal construction, also assum e a predicative relation between the two nouns. Their analysis is based on the idea

What to do with those fools of a crew? 30 1

Napoli ( 1989: 222), too, notes that N 1 "acts as a predicate to the NP introduced by of', and adds that the semantic relations between the two elements are the same as those in other predicational constructions. Moreover, there must be a matching of selection restrictions between the two nominals that mirrors the restrictions in copular sentences (*this building is a nitwit: *this nitwit of a building) and the two elements must match for semantic gender and number (*a prince of a woman, *a prince of people; Napoli 1 989: 223). The predicative nature ofN 1 is also reflected in the fact that in many cases Nt together with the element of and the second determiner (N t + of + a) corresponds to an adjective (a foolish policeman, that idiotic prime minister; Quirk et a1. 1985: 1285). A further reason to assume that N I functions as a modifier of N2, is that Nl is invariably evaluative (hence * a teacher of a husband) and often used figuratively (based on a simile or metaphor: an angel of a girl; Aarts 1 998: 1 2 1 ). This . seems to confirm the idea that it is N2 which determines overall reference (Quirk et al. 1 985 : 1 285; Napoli 1 989: 224; Aarts 1 998: 1 24). Nevertheless, there has been considerable confusion among linguists as . to which of the two nominal functions as the (syntactic and semantic) head of the construction. This confusion can already be found in early traditional accounts. Poutsma ( 1 926), for instance, describes the second noun as ;;}�indicating the form, department or respect as to which the person or thing indicated by a preceding noun is distinguished" (Poutsma 1926: 769), .which suggests an analysis in which the first noun is the head and the second noun (part of) a modifier. Poutsma subsequently observes, however, that "[t]he idiom is common only when from a semantic point of view the noun in the ofadjunct represents the notion qualified, that of the preceding noun the qualifying notion" (ibid.: 770), in which case it is clearly the first noun which modifies the second. Kruisinga and Erades ( 1 932) evade the question of headedness by characterising binominals as appositional constructions, with "neither element clearly dominating over the other". . "" '."'''''"''', �lsewhere, however, Kruisinga ( 1 925: 302) maintains that in some . �onstructions of this type "the first noun is clearly the adjunct" (e.g. a duck • i,Jja child), while in other cases the first noun seems to be the headword (as · do a magnificent figure of a man). It is obvious, then, that intuitions · .

that, like clauses, binominals involve predicate inversion. Thus, they assume that the binominal construction een beer van een vent 'a bear of a man' has the same :initial structure as the complex nominal constructions een vent als een beer 'a man ::: £'like a bear', the former being derived from this structure through a process of predicate inversion. ·:

H,

302 Evelien Keizer

concerning the semantic relation between the two nominal parts can hardly be considered reliable. In more recent discussions of headedness in binominal constructions, attempts have been made to find linguistic evidence, both semantic and syntactic, to detennine headedness in these constructions. The resu lt, however, is still equivocal, as, generally speaking, three approaches have emerged. First of all, there are those who propose an underlying structure in which N l functions as the head (e.g. Abney 1987: 297). The assumption that N 1 functions as the head is largely based on an assumption of fonnal mimicry: since the element of nonnally functions as a preposition, heading a PP complement or modifier, it is taken to do so in binominals as well. A somewhat different approach is advocated by Napoli (1989). She , too, proposes an underlying structure of bin om ina I noun phrases with N l as the ( syntactic) head and N2 as part of a PP-complement. At the same time, however, she argues that N2 functions as the semantic head of the construction, with Nl functioning as a predicate of N2. In other words, what we have is a double-headed approach, with N 1 functioning as the syntactic head and N2 as the semantic head.2 Finally, Aarts ( 1 998) rejects the idea of double-headedness in binominals, arguing on the basis of the criteria for headedness put forward by Zwicky ( 1 985) for an analysis with N2 functioning as the syntactic and semantic head of these constructions, with the string 'N 1 + of + a' functioning as a complex modifier. 3 Keizer (to appear) provides additional syntactic, semantic and pragmatic evidence to suggest that, although not unproblematic, this may be the most promising of the analyses proposed so far.

2 McCawley ( 1 988) even goes so far as to ass ign these constru cti ons two underlying structures: one in which N2 is analysed as the head a!ld N 1 as an adj ectival noun, while the element of and the second determiner are ignored, and one in whi ch Nl is represented as the head and of as a prep ositi on heading a postnominal PP. According to McC awley, the acceptability of these constructions depends "on the possibility of assigning both structures to a particular example": if both structures can be used to represent the example in question, the noun has successfully 'mimicked' the adj ective construction (McCawley 1 988: 740-743). 3 For Dutch, which has a s imil ar (though n ot identical) binominal construction, this construction has been discussed exten sively . Here, too, headedness has been a central issue. For analyses taking NI as the head, see e.g. De Groot ( 1 964) and Everaert ( 1992); Paardekooper ( 1956) analy ses N2 as the head. ·

What to do with those fools ofa crew? 303 3. The role of the first determiner

Among the tests proposed to determine which nouns functions as the head are semantic selection, syntactic number agreement with the finite verb, extraposition of the oj-phrase, topicalisation of N2 and concord with the . ' determiner. This section will be concerned with evidence provided by the function and the form of the first determiner. In most cases, this determiner is the indefinite article; it may, however, also take the form of a possessive or a demonstrative pronoun and, occasionally, the definite article, a numeral or a quantifier. Discussion will be restricted to possessive and demonstrative pronouns and the singular numeral one, as these seem to be most relevant for the issue at hand. According to Quirk et al. ( 1 985 : 1 285), a first determiner in the form of a possessive pronoun notionally determines the second noun, as shown by the paraphrase in (2b). As further evidence for this view, Aarts ( 1 998: 13 1 ) mentions the ungrammaticality of (2c), where one and the same :element is marked twice for the possessive. a. her nitwit of a husband b. her husband is a nitwit c. *her nitwit of a husband of hers This may be taken to suggest that generally speaking the first determiner specifies the second noun rather than the first. Such a view is supported by . the fact that where N2 is a proper name, the first determiner cannot be indefinite (*a creep of a James; *an angel ofa Rebecca). As possessive pronouns in English are not marked for number, concord between determiner and head cannot be used as evidence in these '�onstructions . This is differen� of course, with demonstratives. Now, in n,tost cases the two nouns in a demonstrative binominal agree in number .and show concord with the first determiner (see example ( I » . ,Occasionally, however, a plural N t to combines with a singular N2, as . ,shown in example (3): At about half 1 0 the next morning, I stood on the deck of the Adventure, fixing the rigging up proper and trying to sort out the yardarm which those fools of a crew had made a mighty mess of. (http://dead-men.com/fanfic_memories06.php) b. Those p rej udiced fools of a jury were totally unreliable.

304 Evelien Keizer

Clearly, in these constructions the first detenniner shows number agreement with N l . Notionally, however, it looks as though, just like in possessive binominal constructions, the demonstrative specifies N2. Thus the most likely paraphrase of the binominal those crooks of attorneys in ( I c) is These attorneys are crooks. This seems to suggest an internal structure of these constructions in which N 1 functions as the syntactic and N2 as the semantic head. However, although I do no contend the predicative relation expressed in these paraphrases, I do question the validity of the claim that the demonstrative semantically specifies N2. First of all, observe that in these examples the demonstratives do not seem to have their usual demonstrative function. Consider in this respect the examples (4): (4) a. 'That would have been all right,' he went on, 'but, just as hej was about to throw the dynamite, the fish swam away and what do you think [that idiot of a bOy]j did? (http://www.shoarns.comlBirdsBeastsandRelatives.html) b. Many years ago I was singing in school, and that idiot of a principal got so mad at me that he yelled 'I wish you'd get lost and spent the rest of your life singing to the walls' . (http://www.rheindrache.de/spukschloss-en.htm ) c. But that did not solve his problem, his mother's birthday was next week, he had counted on Arsenal beating Manchester and if that idiot of a referee had not awarded that idiotic penalty, Arsenal would have won and he would have a comfortable amount of money to spend and he would not be there counting and (http://fai b lefaible.jag-minns.com/id495.htm) recounting. In (4a), it looks as though the demonstrative can be taken to semantically specify either noun, i.e. both that idiot and that boy could be used to refer back to the referent of co-indexed he� Closer examination, however, shows that the latter is unlikely - if the intention of the speaker had been simply to indicate anaphoricity, use of the definite article would have been preferable (the boy). With the first noun, however, both the demonstrative and the definite article would have been equally acceptable. This may suggest that the anaphoric relation is signaled simply by the presence of a definite determiner (either the or that). Use of the demonstrative detenniner that, however, allows the speaker to do something more than just indicating anaphoricity or identifiability of the intended referent. Interestingly, this extra function is not the usual semantic function of demonstratives

What to do with those fools of a crew ? 305

(indicating [±proximityJ). Instead, the demonstrative has another semantic function: that of intensifying the evaluative judgement given by the speaker (a judgement the addressee is expected to share). Examples (4b&c) seem to confirm this interpretation. Here the identity of the entity referred to by means of the binominal is inferrable from the context. Normally (i.e. in regular NPs) such inferrability is indicated by use of the definite article, not by a demonstrative determiner. Again, it seems that it is the definiteness of the construction which signals the inferrability of the referent, while the extra function of the demonstrative is to strengthen the f orce of the speaker's judgement - a function the demonstrative that can also be have in regu lar NPs with evaluative adjectives (that idiotic boy; that idiotic principal).4,5 The same seems to hold for binominals with the demonstrative this. As can be seen from the examples in (5), this in these constructions can occu r in an anaphoric expression (as in (5 a), where the binominal refers back to a previou sly introdu ced discourse entity) as well as in a first-mention NP (as illustrated in (5 b), where the binominal introduces a new discourse entity). In both cases, this is used in relation to the referent of the overall construction, as determined by the second noun. In addition, however, this serves the same fupction as that in example (4): that of intensifying the evaluative qualification provided by the first noun. (5 ) a. Nemesis - the name tltis idiot of afilmmaker is hiding behind gets it totally wrongfrom the start, (http://www.rumounnachine.comlReviewslNoelie_new.htm ) b. Given the recent controversy surrounding shady recruiting tactics, the Army cannot afford to let this slide. It sends a message that high-ranking officers are above the law, and that 's a terrible message to send to potential new recruits: Go fight in Iraq while t/lis jerk of a general earnsfive times your salary, stays in the United States and cheats on his Wife. (http://www.jsonline.com/news/editorials/aug05/3 50403 .asp) ...

4 W•• M', ';C; ., ,,,;, '

Note that the same applies to constructions with a proper name, both regular and binominal. Thus, it is perfectly acceptable for a demonstrative determiner to co­ occur with a proper name, provided that this proper name is modified by an evaluative e lement. This evaluative element may take the fonn of an adjective (that creepy James) or a noun (that creep of a James). S The demonstrative that can have this function even in the absence of an evaluative adjective, in which case some kind of negative judgement is clearly implied (Tell that brother ofyours that he 's not welcome anymore). Cf. footnote 7.

306 Evelien Keizer

In other words, the demonstrative in these constructions seems to fulfil two separate functions, each applying to a different noun and at a different level of analysis: pragmatically, it functions as a marker of definiteness, indicating the identifiability (or inferrability) of the overall referent (as detennined by N2); semantically as an intensifier of the speaker's eval uative judgement of this referent (expressed by means of N 1) . Finally, it is interesting to note that this kind of analysis also applies to the numeral one. Like demonstratives, one in a binominal construction typically has a reinforcing function, strengthening the evaluative force of first noun. This is illustrated in examples (6a&b). As shown in example (6a'), one can have the same function in regular NPs, where it can be used to intensify the evaluative force of a premodifying adjective.6 (6) a. Then they 'll be like rats in a pit until he makes sense, and you gave him one hell of a beating.



a' . . . . andyou gave him one tremendous beating. b. This is one beast of a system. (http://www .edmunds.com/reviews/listltop t 0/1033 87/artic1e. html) 4. Analysis

As pointed out in Section 2, and as argued extensively in Keizer (to appear), most of the syntactic, semantic and pragmatic evidence presented in the literature so far seems to favour an analysis of binominals in which N2 acts as the head of the construction. Some of the evidence discussed in this paper confinns this view. Thus, there is every reason to believe that possessive determiners specify N2. As in the case of demonstratives and the numeral one, possessives pronouns in English can be seen as fulfilling two functions: the pragmatic function of indicating identifiability of the referent and the semantic function of indicating possession . However, since both functions apply to the second noun, this is entirely in keeping with an analysis in which this noun functions as the head. For constructions with a demonstrative determiner, results are more equivocal. What makes these constructions problematic is that in those cases where a singular N 1 is combined with a plural N2, the determiner agrees in number with N t ; this seems to support an double-headed analysis, As with the demonstrative that it is possib le to for one to have this function in constructions without an evaluative premodifier; in that case, it is heavily emphasized and the evaluative modifier is clearly implied ( . . . and you gave him ONE beating).

6

What to do with those fools of a crew? 307

with N 1 functioning as the syntactic and N2 as the semantic head. This syntactic headedness of N l may, in that case, be regarded as an example of McCawley 's (1988: 741) notion of syntactic mimicry: the first determiner 'mimicks' the behaviour of a determiner in a regular NP, showing number agreement with the first noun to follow. There may, however, be an alternative, functionally motivated explanation for the data. It has been argued that in the examples discussed the determiner that fulfils two functions, and that these functions apply to different nouns. Being definite, demonstratives serve to indicate the identifiability of the referent; in that case, they determine the second noun. In addition, they are a means of intensifying the evaluative judgement expressed by the speaker, in which case they determine N I . This, of course, is entirely compatible with an approach in which N l functions as a modifier, as it is typically modifiers that are intensified. What remains to be accounted for, however, is the fact that in that case the determiner agrees in number not with the head, but with a modifier. One of the underlying principles in functional approaches to language is that syntax is subservient to semantics and pragmatics, whereby " [s]emantics is regarded as instrumental with respect to pragmatics, and . syntax as instrumental to semantics." (Dik 1997: 8). This means that the (morpho-)syntactic features of linguistic elements merely serve to guide the addressee to the intended interpretation of an expression by indicating the semantic and pragmatic properties of the entity (an individual, event, . proposition, property, etc.) denoted by that expression. Thus, the syntactic coding of number merely serves to indicate a semantic property of some entity, while number agreement simply serves to indicate a relation between two linguistic elements. Now, the typical function of demonstrative determiners in regular NPs is to indicate two properties of the referent of an NP: its identifiability and its relative distance from the speaker. Since this referent is denoted by the nominal head of the NP, agreement may be expected to hold between the demonstrative and this head. In demonstrative binominal constructions, however, the two functions performed by the demonstrative - identifiability of the overall referent and intensification of the property denoted by the first noun - do not apply to the same noun. Like in regular NPs, identifiability relates to N2 (as denoting the overall referent);

308 Evelien Keizer

intensification, however, relates to N 1 .7 The concord between the demonstrative and N 1 simply serves to indicate the latter relation. This, then, also accounts for the fact that other plural determines, such as quantifiers and numerals, are not acceptable in these constructions (*manyltwo fools of a crew), since these, like possessives, only determine N2 and must, as such, agree in number with N2 (as in many/two crooks of attorneys). 8

What needs explanation, therefore, is not so much the fact that number agreement holds between the determiner and the modifYing noun, but that (a) a noun is used in a (pre-)modifYing function and (b) that the two nouns differ in number. The latter is, of course, not as surprising as it seems: these constructions are only acceptable when N2 is a (semantically plural) collective noun; it is therefore hardly surprising that the modifying nouns appears in the plural. The fact that a noun is used in a pre-modifying function is more difficult to account for: it is the main defining property of binominal constructions, and, indeed, the property that causes all the confusion about their internal structure. Given, however, that N 1 has this modifying function, the behaviour of the demonstrative determiner is entirely consistent with an analysis of binominal constructions in which N2 functions as the head. 5. Conclusion

First of all it needs to be emphasised that within the scope of this paper it has only been possible to look at some aspects of only one of the tests proposed for determining headedness in binominal constructions; consequently any conclusions to be drawn from the preceding discussion have to be interpreted in a wider context. Nevertheless, some contrib,ution may have been made toward,S a more complete and consistent treatment of binominal constructions. What will have become clear, for instance, is that:one can�ot always blindly trust the tests proposed to determine headedness 'in binominal constructions. Considering the role of the determiner is certainly useful, but 7

Note by the way also in the (rather unlikely) case of a deictically used demonstrative binominal (e.g. Look at that monster of a house!) the demonstrative can be argued to perform this intensifying function. S It also explains the un acceptabi lity of a creep of a James (example (3a)); here again, there is no relation between the determiner and the first noun, as both functions of the indefinite article (indicating unidentifiability and singUlarity) apply to the second noun.

What to do with those fools of a crew? 309 only if one t akes into account all it s synt act ic, semant ic and pragmat ic feat ures and funct ions. Just because in regular NPs t he det erminer shows number agreement wit h the head noun does not automat ically mean t hat this must be t rue for binominaI const ruct ions as well; binominals are, after all, no regular NPs. The very fact t hat in a binominal t he modifier takes t he form of a noun makes it possible for t he determiner to agree in number with t his modifier rat her t han wit h t he head. If we assume t hat synt act ic number agreement reflects a semant ic relat ion between t wo element s, this need not be a problem. This, in turn, means that t he behaviour of demonstrat ive determiners in binominal const ruct ions is, after all, compat ible wit h an analysis of binominals as right -headed construct ion with a complex modifier. References

Aarts, Bas.

1 998. English Binominal Noun Phrases

.

Transactions 0/ the

Philological Society 96: 1 1 7- 1 58.

Abney, Steven P. 1 987. The English Noun Phrase in its sentential aspect. Unpublished PhD dissertation, MIT, Cambridge, Massachusetts. Bennis, Hans, Norbert Corver, and Marcel den Dikken. 1 998. Predication in nominal phrases. The Journal o/Comparative Germanic Linguistics 1 :85- 1 1 7. Dik, Simon C. 1 997. The Theory o/Functional Grammar. Part 1: The structure 0/ the clause. BerlinlNew York: Mouton de Gruyter. Everaert, Martin. 1 992. Nogmaals: een schat van een kind. In De binnenbouw van het Nederlands: een bundle artiekelen voor Piet Paardekooper, ed. by H. Bennis and J. de Vries, 45-54. Dordrecht: lCG Publications. Groot, A.W. de. 1 965. Structurele syntaxis (2nd ed.). Den Haag: Servire. Keizer, Evelien. To appear 2006. The English Noun Phrase: a multifunctional approach. Cambridge: CUP. Kruisinga, Etsko. 1 932. A handbook 0/ present-day English, Part II12. 5 th ed. Groningen: Noordhoff. Kruisinga, Etsko and P. Erades . 1 947. An English Grammar, Vol. /1/. 7th ed. Groningen. McCawley, James D. 1 988. The syntactic phenomena of English. Chicago: Chicago University Press. Napoli, Donna 1. 1 989. Predication theory: a case study for indexing theory. Cambridge: Cambridge University Press. Paardekooper, Piet C. 1 95 6. Een schat van een kind. De Nieuwe Taalgids 49:93-99. Poutsma, Hendrik. 1 926. A grammar 0/ late modern English, Part II, Section II: The Parts o/Speech. Groningen: Noordhoff. Quirk, Randolph, Sidney Greenbaum, Geoffrey Leech and Jan Svartvik. 1 985. A Comprehensive Grammar o/the English Language. London: Longman. Zwicky, Arnold. 1 985. Heads. Journal o/Linguistics 2 1 : 1 -29.

Why indefinite pronouns are different Istvan Kenesei

1. The problem and the initial data

Indefinite pronouns differ from common nouns in English and a number of other languages in their position or some other feature of the construction they form with the adjectival phrases (APs) that modify them. This squib intends to throw some more light on their curious behavior by attempting to offer a new analysis. In a recent paper in reply to Kishimoto (2000), Larson and Marusic ( L&M) (2004) list the following properties of NPs containing indefinite pronouns and APs in arguing for the option of a postnominal AP position in deriving the relevant structures. =

a) Unlike NPs containing common nouns, NPs containing indefinite pronouns, or more exactly, pronominals expressing quantificational meanings (Qprons henceforth), always have postnominal APs: ( 1 ) a. every interesting book b. *every book interesting c. *interesting everything d. everything interesting b) Qprons allow for the recursion of postnom inal APs, unlike common nouns: (2) a. b. c. d.

the explored navigable rivers *the rivers explored navigable everyone present capable of lifting a horse any place available accessible by bike

Why indefinite pronouns are different 3 1 1

c) Qprons do not permit uninflected measure phrases: (3) a. b. c. d.

a [23-inch-Iong] rope a rope [23 inches long] * anything [23-inch-Iong] anything [23 inches long]

d) No adj ective that can be used only attributively can occur with Qprons: (4) a. b. c. d. e.

some live thing This thing is alive/*live. *a thing live a thing alive *something l ive f. something alive

e) Only stage-level interpretation is available in Qpron - AP constructions, i.e., the individual-level interpretation available in (5a) below, according to which the stars have the general property (size, luminosity, etc.) of visibility, but they needn't be visible now, is not possible in (5b), and neither is it possible in (5c). (5) a. the visible stars (include Capella, Betelgeuse, and Sirius) b. the stars visible c. (Show me) everything visible f) Unlike prenominal APs, which allow for both restrictive and nonrestrictive interpretations, postnominal APs and APs in construction with Qprons only tolerate restrictive interpretation. Every unsuitable word was deleted. ' Every word was deleted, they were unsuitable.' ' Every word that was unsuitable was deleted. ' b. Every word unsuitable was deleted. # 'Every word was deleted, they were unsuitable.' ' Every word that was unsuitable was deleted.' c. Everything unsuitable was deleted. n ' Everything was deleted, everything was unsuitable. ' ' Everything that was unsuitable was deleted. '

(6) a

3 1 2 lstwin Kenesei

g) Comparative adjectives with complements can precede common nouns and then may yield incompatible readings; no such reading is available in Qprons followed by comparative APs. (7) a. b. c. d.

a taller person than Max I#a taller person than this bookshelf

a person taller than this bookshelf someone taller than this bookshelf

Larson and Marusic (2004) conclude that the NPs containing Qprons and APs cannot be derived from an [AP N] structure by moving the noun in front of the AP, as Abney ( 1 987) or Kishimoto (2000) proposed. Rather, they derive from structures of the fonn of (8a), where the Qpron corresponds to the D-N unit, or from (8b), where it is a morphologically complex D. Prenominal APs arise either by AP-preposing or from originally prenominal APs, cf. (9). -

(8) a. [oP D -N AP] b. [op 0 [NP 0] AP] (9) a. [op 0 APi NP Ii ] b. [op D AP NP] 2. More data, more questions

We will take as our starting point L&M's (2004) conclusion that the APs occurring with Qprons must always be predicative. This is the characteristic that accounts for most of the properties listed above, and especi ally for those under c) through f) . But the fact that these adjectives are predicative does not reveal why Qprons are different from common nouns with respect to the positioning of the AP. Note that their analysis is no more than a stipulation: Qprons must be fol lowed by APs. Besides, the curious behavior of Qprons is conspicuous also in languages in which APs are as a rule prenominal, such as German, Russian, or Hungarian. Compare the following examples from Hungarian. ( 10) a.

akar-sz? [Mi lehetetlen]-t what impossible-ACC want-2SG 'What do you want that's impossible? '

Why indefinite pronouns are different 3 1 3

beszel-ek. b . [Barki francia]-val anyone French-with speak- I SG 'I'll speak with anyone French.' c. [Valaki [AP Anna-hoz hasonl6]]-t someone Anna-to similar-ACC 'I saw someone similar to Anna. ' d. *[Francia barki]-vel beszel-ek. French anyone-with speak- l so

hitt-am. saw- l so

( 1 l ) a.

[Az [AP Anna-hoz hasonl6] lany]-t hitt-am. the Anna-to similar girl-Acc saw- l SG 'I saw the girl similar to Anna.' b. *[A lany [Anmihoz hasonI6]]-t lattam.

But even though there are no postnominal APs, the generalization holds that the APs cooccurring with Qprons must be predicative. In the followin g examples lopott 'stolen' is a lexical adjective derived from the verb lop 'steal' just like the closely related past participle el-lop-ott 'PERFECTIVE PFx-steal-PPRT, stolen' . However only the former can occur predicatively and, consequently, in construction with a Qpron. ( 1 2) a. A kicillitott aut6k lopott-ak voltak. the exhibited cars stolen-PL were 'The cars exhibited were stolen.' b. * A kicillitott aut6k ellopottak voltak. c. lelentsd, ha [barmi ]opott]-at tal{tl-sz. report-IMP if anything stolen-ACC find..2sG 'Report if you find anything stolen.' d. * lelentsd ha barmi ellopottat talcilsz. . More�ver, even in a language that has postnominal APs, as in French, Qprons are not simply followed by APs but are separated from them by the linker de, as was first noticed in case of the following examples by Kayne '(1 994). (13) a. Qui *(de) serieux as-tu rencontre? DE serious have-you met who *(de) extraordinaire n'est arrive ce matin. b. Rien nothing DE extraordinary not-is happened this morning

3 1 4 Istvan Kenesei

Similar linkers surface in modifier constructions in a number of other languages, as the Romanian example below illustrates, for which Rubin (2002) proposes the structure in ( 1 4c). ( l 4) a. Problema este (*de) grea. problem-the is DE tough 'The problem is tough.' b. Problema este curios *(de) grea. problem-the is curious DE tough 'The problem is curiously tough.' C. [ModP curios [[Mod de] [AP grea]]] His ' Modifier Hypothesis' accords well with facts from other languages, including Rubin's examples from Chinese and Tagalog, or the Hungarian prenominal, and often non-predicative, attributes in ( 1 5)-( 1 6). ( 1 5) a. [op a [NP kenyer [pp a polc alattltegnap-rol]]] the bread the shelf below/yesterday-from 'the bread below the shelf/from yesterday' b. [o P a [ModP [pp pole alattltegnap PO]i [[Mod -i ] [NP kenyer tj ] ] ]] 'idem ' ( 1 6) a

[op a [NP Iud [pp a leghosszabb nyak-kal]]] the goose the longest neck-with 'the goose with the longest neck' b. [op a [ModP [pp leghosszabb nyak PO]i [[Mod -u ] [NP Iud ti ]]]] 'idem' .

While Rubin (2002) is not concerned with what underlies the options in positioning modifiers with respect �o what they modify, it provides a possible framework for L&M's (2004) AP-movement hypothesis, which we will adopt here. As a final piece of data that L&M (2004) list but do not account for, observe ( 1 7), in which although both adjectives must be 'postposed ', their order is not irrelevant. ( 1 7) a. everyone tall present b. *everyone present tall

Why indefinite pronouns are different 3 1 5

Before our proposal is put fOlWard, let us summarize the properties accrued so far. Qprons occur with APs if (i) the AP can be predicative, and (iia) on the opposite side with respect to the usual order in case of common nouns with or without a linker, cf. ( 1 ), ( 1 0), or (iib) with linkers if on the usual side, cf. ( 1 3). I 3. Proposal

If the APs participating in the construction under investigation are predicative, suppose that aU NPs containing such adjectival modifiers emerge from relative clause-like structures as small clause (SC) complements to an N head, which in effect is a functional category supplied with features only requiring an SC as its sister. The SC contains the minimal predication [sc NP - APNPIPPINP], of which only AP is illustrated below. This structure draws heavily on Kayne's ( 1 994) and Vergnaud's ( 1974) original suggestions, but differs from them in that the small clause is a complement of N, rather than of D, for obvious reasons of 'ordering restrictions and possible recursion ofNP. First then N is merged to SC: [NP N [sc NP AP]] [sc something/book interesting]

Next, the NP moves (a la Kayne and Vergnaud) recursively to Spec,NP providing for the relative clause structure and interpretation. [NP NPj N [sc Ii AP]] [NP somethinglbooki N [sc tj interesting]]

Now, following Rubin (2002) Mod is merged to NP, which is where linkers · in Tagalog, Romanian, Chinese, and Hungarian are placed and which will ;in effect 'make room' for the surface position of prenominal attributive Jllodifers, whether APs, PPs, clauses, or other. Languages with strictly 'postnominal APs do not project the Mod l ayer.

iStO)

[ModP Mod [NP NPj N [sc Ij AP]]

. . ) :The case of Basque, in which a linker occurs between the modifier and the Qpron (Jon Ortiz de Urbina, p.c.), is not illustrated for lack of space, and neither are Thai, :'which falls under (iib), c[ Den Dikken and Singhapreecha (2004), and Japanese, which has alternative Q AN and AN Q orders (Senga Tom, p.c.). -

-

3 1 6 Istwin Kenesei

Developing Rubin's suggestion further, the heads of the predicates in SC are marked for [+mod] in the languages in which the canonical order is A PIPP/... - NP. Then SC is fronted by remnant movement to Spec,ModP checking and deleting the uninterpretable feature [+mod] on Mod. SCs with predicates marked [-mod] stay put and end up as postnominal modifiers, as in (3 b) or (4b) ? (2 1 ) a. [ModP [SC /j A P1 Mod [NP NPj [N [�]]] b. [ModP [SC ti interesling1 Mod [NP somethinglbooki [N [tj] ]] Finally, Q is merged to ModP: this is another locus for linkers, as in French, Basque, and Thai, and provides ultimately for the 'visibility' of NP-intemal quantifiers in the proposition of which the NP is a constituent of.

If Q carries an uninterpretable feature [+q], it has to be checked and deleted by moving the NP containing the Qpron to Spec,QP. Unless the Qpron occupies this prominent position, it will not be v isible for acquiring its scope over the proposition, ultimately blocking quantifier interpretation. We contend that Qprons must surface in positions inaccessible to common nouns because of the properties of scope relations in the language in question. NPs containing common nouns as their heads are of course marked [-q] and thus prevented from remnant movement. (23) a. [oP [NP NPj [N tj] ]k Q [ModP [sc Ii A P]j Mod elk]]] b. [oP [NP somethingi [N tj]]k Q [ModP [SC /i interesting]j Mod [ tk]]] Note that in this approach Qprons are taken to be . nouns and thus distinct from determiners (placed under D). Their complementary distribution is accounted for by claiming that once Q is projected, no D-l�yer becomes possible, because referential interpretation, carried by D, is incompatible 2

Lack of space precludes the discussion of the regularity captured in (Sa-b) or (6b-c), but nonpredicative and/or nonrestrictive prenominal APs must receive an account crucially different from that for predicative/restrictive ones. It may well be the case, for example, that strictly prenominal (nonpredicative, nonrestrictive) APs do without the Mod layer and the NP is a complement to the A head, as was suggested by Borer ( 1 984). Note that we believe SC-movement to be compatible with subsequent N-movement to D as in Longobardi ( 1 994) and oth ers .

Why indefinite pronouns are different 3 1 7

with quantifier interpretation, carried by Q. Consequently, D cannot be merged to Q(P), but ModP or NP are possible options, and of course no preposing of the NP containing the common noun can take place. (24) a. [oP D [ModP [SC ti AP]j Mod [NP NPi [N [tj ] ] ] b. [DP the [ModP [sc tj interesting]j Mod [NP boo� [N [�]]] 4. Conseq uences and further problems

The schematic proposal outlined in section 3 gives us adequate ground to :;; ·aC:;ComnlOOlate the varieties in AP - Qpron structures surveyed above. To begin with the English examples not covered in section 3, since adjectives cooccurring with Qprons must originate in predication (Le., relative clause) structures, only (3 d), (4f), and (5c) can emerge. Neither wil l the nonrestrictive reading in (6c) and the incompatible meaning in (7d) be available. The same requirement accounts for the difference in ( l 2c-d). As � SU1l)poseCl above, ModP is not projected in case of postnominal APs, in words, these adjectives cany a [-mod] feature, as in (3b, d), (4d, f), c) in English, and ( 1 3) in French. In these latter examples the SC stays place, but the Qprons must move to Spec,QP to check [+q] on the Q as in (25), with Q spelled out as de in French. [QP NPi Q [NP Ii N [sc Ii AP]]] [QP rienj de [NP Ii N [sc Ii extraordinaire]]] Finally, the order of postnominal adjectives, as in (1 7), can be accounted by recursion of QP in the small clause in the following derivation. [QP Q [NP N [sc [QP everyonei Q [ModP [sc . . . tall1 Mod [NP tj ]]] [AP present] b. [QP [QP everyonej Q [ModP [sc . . . tallJj Mod [NP tj ]]]k Q [NP tk N [sc tk . [AP present]]]

First everyone tall is generated in the subject of the SC by moving the ' 41djective and then the quantifier in the lower cycle as reviewed above, and this QP raises initially to fonn the relative construction, and then on force of its [+q] feature to the matrix Spec,QP, while present remains in original position. Althoug� this appears to be a satisfactory analysis for :. the possible order, it does not explain why the opposite order is ' , unacceptable with tall sitting in the higher predicate and Mod merged to the matrix NP.

3 1 8 Istvan Kenesei

As more and more data surface from a wide array of languages, such as Turkish and Japanese, more interesting problems related to Qpron AP constructions emerge and call for a more comprehensive analysis. However, it seems that the crucial property of Qprons is their being quantifiers, which makes it necessary for them to be in a prominent position so that they could be visible for taking scope over the proposition they and the NPs containing them are constituents of. -

References

Abney, Stephen. 1 987. The English Noun Phrase in its sentential aspect. PhD dissertation. MIT, Cambridge. Borer, Hagit. 1 984. Parametric syntax: Case studies in Semitic and Romance languages. Dordrecht: Foris. Den Dikken, Marcel, and Pomsiri Singhapreecha. 2004. Complex Noun Phrases and linkers. Syntax 7, 1 -54. Kayne, Richard S. 1 994. The antisymmetry ofsyntax. Cambridge: The MIT Press. Kishimoto, Hideki. 2000. Indefmite pronouns and overt N-raising. Linguistic Inquiry 3 1 , 557-566. Larson, R. and Marusic, Franc. 2004. On indefmite pronoun structures with APs: Reply to Kishimoto, linguistic Inquiry 35, 268-287. Longobardi, Giuseppe. 1 994. Reference and proper names: A theory of N­ movement in syntax and in logical form. Linguistic Inquiry 25, 609-665 . Rubin, Edward 1. 2002. The structure of modifiers. Unpublished draft. Un iversity of Utah.
Seeing the forest despite the tree

: 1. Introduction

In 1 985 Chomsky and May introduced segments and categories into the inventory of Generative Syntax. In subsequent years, the attempts to make sense of this distinction (or to use the confusion to further one's own agenda) have sparked a plethora of brilliant ideas (e.g., Kayne's LeA, to mention just one). Yet, 20 years on, a little demystification seems in order: A s egmented tree is actually a two-tree forest-and that's all there is to it. I .

2. Trees: A reference manual

What is a tree? In our not particularly naturalistic context it is an instance of a fairly simple mathematical object, i.e. a partially ordered set (or poset) (N, [>. ) consisting of a set N of nodes and some antisymmetric, transitive, reflexive order relation2 Do· on N, which obeys two additional A poset q'= (N, [>* ) is a tree if: a. there is a distinguished element TEN with the property 'dn EN. r[>* n, and b. every set S {y I x [>* y l>* z} e N is totally (or linearly) ordered, in the sense that 'dX,Y E S. x l>+y v y l>* x . =

. la) ensures that every tree has a Toot, while ( l b) excludes deviant .. structures like (2). ..

•

• I This squib is a sl ightly revised excerpt of material

I was

working on when [ left

'the field in the late 1 990s. Sorry, Henk-still no natural language examples ... 2 These properties of a relation R are (standardly) defmed as:

(i)

a.

h. c.

xRy /\ yRx -+ x=y 't/x, y, z. xRz /\ zRy -+ xRy

Antisymmetry: 't/x, y. Transitivity:

Reflexivity: 't/x. xRx

320 Hans-Peter Kolb

(2)

a

[>.

is, of course, just the reflexive variety of the dominance relation we all know and love from Syntax 1 0 1 . Many popular geometric relations on syntactic trees are readily stated in tenns of ( 1 ). A few (trivial) definitions are given here for later reference:

(3)

Proper dominance: x 1>+ Y =def x l>* Y 1\ x-:j:. y Immediate dominance: x l> y =def X [> + Y A. 'Vz. X l>* z[>+ y ---+ x=z Sisterhood: x sister-ojY =def x-:j:. y 1\ 3z. z I> x /\ Z I> Y

The irreflexive, or proper, variant of dominance is the irreflexive core 1> + of I> * .3 Immediate dominance (motherhood) is minimal irreflexive dominance. As a rule, I' ll use the notion of dominance ambiguously in what follows if both interpretations are meaningful in the context. It has to be noted, however, that the actual choice of interpretation may have consequences which I will leave to the reader to detennine unless they have a bearing on the argumentation.4 The other textbook relation, precedence, is not an intrinsic part of our definition of trees. On scrutiny this turns out to be true for the trees employed in linguistic theory, too: Precedence is almost invariably not part of the definition of a syntactic structure, but defined on it in various ways: We find extrinsic characterizations such as 'Standard GB' directionality

J

Both and figure in the linguistic literature. My particular choice of fundamental relation is based on convenience, not ontological commitment. Obviously, the notions are interdefinable. 4 By way of a simple example, consider the c-command relation: if defined using the reflexive notion, a. does not c-command itself, yet it does if dominance is taken to be irreflexive.

Seeing the/orest despite the tree 321

parameters or GPSG's LP rules as well as intrinsic ones like Kayne

( 1 994)' s LeA. 5

According to ( 1 ) any set of entities that are all different can serve as the nodebase of a tree. Syntactic nodes, on the other hand, carry a lot of extra information: features of various kinds, maybe barlevels, indices, etc. pp. We can capture this fact by assuming some set of labels L and a labelling function I.:N�L assigning to any node neN a label I. E L . It shou ld be clear, that neither the construction of L, nor the definition of I. is a trivial task-in fact, these two will contain much of the gist of any linguistic theory. Here, however, we will just assume that there are such L, i, getting more specific only where needed. Let us call a tuple (N, [>* ,L,e) a labelled (or syntactic) tree, if (N, 1>* ) is a tree in the sense of ( l ) and L, .e are as above.6 Moreover, anticipating on the terminology commented on in :the next section, we call a linguistically relevant element of a syntactic ' structure a category, assuming the tenn to refer to the pair (0, 1.(0» , :consisting of some node n and its label . 3 . Segments and Categ ories

our linguistic relations are just special cases of the intrinsic geometric ' relations between tree nodes aod our categories are just labelled nodes, the structures considered so far are undoubtedly trees. Accordingly, are dealing with well-defined, straightforward objects the formal ' .c;: of which -as ordered structures as well as in their relation to nrl'\nF'rt'lF totltnal languages (e.g. Davey and Priestley 1990, Ginsburg 1 97 5)-- are 'Well understood. This picture changes dramatically with the introduction of segments and categories by Chomsky ( 1 985), following May ( 1 985), ;reslJlti'[19 in structures like (4).

,� .,,_.

However, including some notion of precedence is no big deal, e.g. following . et al. I 990:443f): __ ___

An ordered tree is a structure �=(T,-<.)� with -<. the antisymmetric, transitive and irreflexive precedence relation on N, with the following additional constraints: a. Exclusivity: \lx,yeN. x-<.y v y-<.x +-+ x y 1\ Y x b. Non-tangling: 'v'w,x,y,zEN. w-<.x " w y " X z � y-
322 Hans-Peter Kolb (4)

0

�8 � I

2

1

3

3

I

4

�

� 6

7

5

/�

5

This move decoupled the linguistic relations on categories from the geometric relations on nodes in a way hard to capture in a single tree. The picture emerging from the sketches in Chomsky ( 1 985) and 7 subsequent papers is as fol lows:

(5)

7

(Chomsky-)adjunction of some a. to some node labelled p does not create a new category label1ed p but rather extends the p-category by a new segment. a. A node n is a segment of category � iff fen) = p and: i . n l> m � Rem) * �, or ii. 3m. n I> m /\ £(m)=� /\ Vo. n I> 0 /\ £(0)=p � o=m b. A category, then, is a maximal sequence of segments, linearly ordered by I> . c. The category a dominates � if every segment of a. dominates 13. (Chomsky 1 993 : I I ) d. The category a contains � if some segment of a. dominates p. (ibid.) e. (l excludes J} if no segment of (l dominates p. (Chomsky 1 985:9 ( 1 7»

We assume that categories are uniquely identified by the labels of their segments. Throughout this paper we adopt a particularly simple version of this assumption, namely that (i) all segments of a category have the same label, and (ii) all distinct categories have different labels. More sophisticated constructions are easily devised.

Seeing theforest despite the tree 323

To see that these definitions are not as innocuous as they look, just sit back for a moment and try to figure out, what is immediately dominated by the category 1 in (4}--and then, which category immediately dominates 6. The answers existing formalizations of G/B theory give to these apparently simple questions are-interesting, to say the least. To mention but two, Johnson and Stabler (1 995) don't provide an intelligible solution at all. 8 Rogers ( 1 994), on the other hand, takes Chomsky's definitions fairly literally. He derives coherent-though not necessarily the intended­ results: Category 1 dominates 3 and 7 immediately, 6 is not immediately­ dominated by anything, neither is category 5. In both cases-and in all other formal treatments I am aware of--one can't help wondering, why the . trivi al relations of (3) suddenly don't seem to hold anymore. The confusion arises from definition (Sc) which generalizes the dominance relation to categories. While this is a widespread and only mild abuse of terminology, in this case it obscures two important facts, namely that:

·

the (,geometric ') dominance relation 1>* between nodes has, in fact, two counterparts in the world of categories, namely the dominance relation I>j (Sc) -which we hereby christen i-dominance (with the i of 'inclusion') - and containment 1>(: (Sd), 9 and . I>j and I> co are not intended as relations between nodes, but strictly between categories (and, maybe, terminals-cf. Chomsky 1 994:26). 10

if l>j and l>c: are the categorial counterparts of 1> * , they should obey same constraints. In particular, they should be antisymmetric, transitive,

Since in their framework it is nodes which immediately dominate, it is unclear the category J d ominates immediately; 6 is immediately-dominated by the segment of 5. Again, it is not clear, whether this means that it is lnllnecliatlelyo·domiIlat��d by the category 5. It most certainly is a sister of category 'xJ--allth01Ugh they are not immediately dominated by the same element (10hnson Stabler 1 995:47f). �' h:KCll11Si(m is just that this requirement, albeit clearly intended in the linguistic literature, is an illterpretation of (5). Fixing the defmitions to actually express it on our standard . . SYntactic tree without loosing reflexivity of is not an easy task. Under the treatment proposed below, however, this problem vanishes without a trace. . .

324 Hans-Peter Kolb and reflexive order relations on categories, subject to ( 1 }-and here the trouble starts: This can 't be true for both in the same tree. " 4. From segmented trees to forests...

If we take this point seriously, it becomes clear, that, with respect to categories, structures employing segments are just a short-hand for pairs of trees < C, f) with I> c the tree-order on C, and 1>; on 1. In a structure q'=(T, 1>. ) comprising only trivial, i.e. one-segment, categories, C =(C, 1>(: ) and 1 =(1, I>j ) coincide and are equal to rr. As Soon as q' contains non-trivial categories, however, their segments will be conflated to one node both in C and I, which will dominate a category a. in C or I iff some or all segments, respectively, dominate some segment of Cl in q: Some examples are in order: (6) a. A simple adj unction structure: tf =

0

c =

/\

2

/\

2

/\

3

II

0

/\ ;1\ 2

5

3

4

5

I =

0

ffi 2

/\

3

5

4

4

One way out of this dilemma would be to drop the requirement that the partial orders and be tree-orders, i.e. conform to ( 1), but this move would not only blur the parallelism between and these relations, it would also complicate the definition of the syntactic notions building on them.

Seeing the forest despite the tree 325

b. Multiple adjunction [= (4 )] : 12 q' =

0

c =

/\ /\ /\ /\ I /\

/\ !l\ /\ I

8

2

2

3

3

6

I

2

6

3

8

5

7

5

6

5

4

0

!l\ �

8

7

3

4

7

1 =

0

5

c. Multiple adjunction and n-ary branching: t[ = 4

/\

A 11\ /\ 2

3

�

11\ /\

12

• • •

/\ � 10

6

/\

6

10

6

2

6

9

/\

4

2

5

0

c=

0

6

3

5

7

9

8

8 J

7 4

5

2

= 3

0 9

10 6

7

8

featuring the only strictly category-based solution to our earlier questions.

326 Hans-Peter Ko/b

The recipe for constructing C and I is very straightforward: Construct C and I by collecting one representative for every category-its highest segment in the case of C, its lowest for 1.13 Restricting the domain of 1>* accordingly immediately yields 1>(: and 1>; , and thus C and 1, respectively. To put it formally: (7)

DefineJc:T-+T and.fi:T-+T as follows:

fc(x)=y such that £(y)=£(x) /\ 't/z. £(z)=f(x) -+ Y 1>* z Ji (x)=z such that f(z)=e(x) /\ 't/y . £(y)=£(x) � Y 1> * z Then C=(C, 1>(: , Lc, lc ) with x

;�. ���!. r���}

LC = L and fC = f rCxL

I=

(I, I>i , Lh £1) is defined correspondingly.

It is an easy exercise to verifY that 1>'(: and I>i (a) are tree-orders in the sense of ( 1 ), and (b) conform to (5d,c), while expressing strictly relations between categories in C an d 1, respectively. All geometric relations based on 1>* like the ones in (3) just carry over, 14 as do linguistically inspired ones like c-command. Ordered trees in

There is one special case to consider, namely if the root node itself is a non­ trivial category. Without special provisions, I would fail ( 1 a) - in this case. To account for this possibility, add to Ta new, distinguished formal node r dominating all nodes in tT. This leaves all relations between the original nodes in orunchanged. Let e(r)=1- to ensure that it can't be referenced by any linguistic principle or mechanism. Since this construction is always possible, we assume without loss of generality in what follows, that the root of Tis a one-segment category. 1 4 This comes as no surprise, since (7) specifies in effect a syntactic interpretation of C and I in or (cf., e.g., Ebbinghaus et a1. 1 996: 1 3 1 ff), i.e., both C and I can be viewed as substructures of tf, as exemplified by yet another rendering of the relations in (4): 13

Seeing theforest despite the tree 327

the sense of footnote 4 do not pose a problem, either: restricting -< in the same way as 1>* yields precedence orders on C and I which impose on the Tots terminal nodes an ordering identical to the one in The construction of C and I provides us with an easy and efficient way of verifying the structural relations between categories on any given rr. Given the considerable confusion that is a recurrent feature of the actual use of the category/segment distinction in the literature, this is of some value in itself. In the next section, however, we will investigate the possibility to tum the relation between q'and (C, 1) on its feet.

Syntactic theory explores (structural) relations between linguistic entities. In the ontological universe of Generative Grammar, categories clearly are such entities, while the abstract nodes of some tree depicting these relations are not. Since C and I are just about categories while q' fosters additional nodes (i.e., contains 'too much' information), the pair (c. 1) is abstract . more fundamental from this point of view. To be able to put this insight to use we have to know how tight the relation between our trees is . So, to what extent can the original q' be rl�T" \II" rI from C and n A geometric tree q = (G, I>G , LQ, £0) derived from C and l is a structure satisfying the following conditions: a. C c G ;;;;2 1 b. I>c . � ' 1> G . ;;;;2 ' 1> j . c. lc C lG � II d.

'V'CE C,XEG . eG (x)=lG (c)

--+

C l>a x

e. 'V'i E I�EG . .eG (x) =.fo (i) --+ X I>G i f. 'V'c,d E C , i j E I. l (c)=l(i) /\ l(d)=.eO) /\ C l>c d 1\ i �� j --+ 3X,Y EG. lG(x)=lG(y)=l(c) 1\ x l>� d /\ y �;; d

however, that-spoken order-theoretically-while C is a sub-(semi-)lattice of u-semilattice t1, 1, albeit a v-semilattice in its own right, is not: only k ;" ;nl�A.c!'F�nlE·c!' order and joins (u). With other words: is much more different from is. As long as precedence is intimately linked to this is all we can hope fOf. The . ;(tClilrtiIlanc;e relations in C and I differ, so the relative precedence order of the non­ iiE':tcmninai categories will be different, too.

328 Hans-Peter Kolb

g. VC E C, i E I. £(c)=i(i) -+ 'v'X E G. c l>a x 1\ xl>a i h. I> G is a tree order in the sense of ( 1 ).

+4

f(c)=f(x)=f(i)

(8a--c) express the fact that both C and I are restrictions of a geometric tree to (representatives of) its categories. (d,e) make sure that these representatives have the right properties, i.e., being the highest or lowest segment of a category, respectively. (f) ensures that adjunctions are actually 16 represented as adjunctions in q. (g), finally, imposes requirement (5b) on q: A category is the maximal, uninterrupted sequence of its segments, linearly ordered by 1>'(; . Unfortunately , (8) defines an infmite equivalence class [q] of structures, with 'Tas one of its members. Note, however, that all members of [ q] are isomorphic with respect to relations among categories. Since we are only interested in these relations, we can safely forget about q' and pick the smallest of them as the representative of all others. 17 Again, efficient algorithms to compute this smallest member of [q] exist. How do our example trees in (6) do with respect to this construction? Their minimal 'geometric trees' equal the original trees except for (6c): (6)

c' ,

T

0

=

�

4

c = I

�

2

�

5

2

�

4 2

1

6

/'--..

5

2

1

�

�

�

3

o

3

I

6

�

9 10

6

7

8

8

6

,� 9

6

'A

10

7

6

16 Without it, an a contained in, but not i-dominated by, � could end up dominated by all segments of P in q, due to the fact that, with respect to categories, . => .

'.

Ac�ally, imposing this condition makes (8g,h) redundant; they follow from the structure of c and land the minimality requirement on q. 17

Seeing theforest despite the tree

329

In the general case, the differences between c:r and q will be confined to categories with more than one adjunct; they will invariably be represented as two-segment categories by q. Yet, to my knowledge, no linguistic princip le or operation makes reference to the relative hierarchical order of • adjuncts to the sam� category, so, again, q'contains too much information. 6. Arbores sani in silva sana

. The confusion arising from the generalization of the geometric notion of . dominance between nodes to the linguistic one between categories evaporates if one takes the standard l inguistic derivation tree q' as a ;shorthand for the twin structure (C, /), representing the containment and i­ dominance relations between categories, respectively. Since in contrast to their counterparts on q; I> c-: and I> i are tree-orders on categories, linguistic notions and operations on C and I can retain a simple, node-based characterization. (C, /), as well as the smallest single tree q that correctly .r�presents its containment and i-dominance relations are easily computed . ·· 80, for all intents and purposes, we can abolish cr and proceed using (C, 1) In general, examining the compatibility of structures independently ... :characterized by some operation or relation can provide a sanity-check on . the descriptive means of syntactic theory building. A quick scan of the literature suggests that even in the era of minimalism such an endeavour may be no lUXUry.

Chomsky, Noam. 1985. Barriers. Cambridge, MA: MIT Press. Chomsky, Noam. 1993 . A Min imalist Program for Linguistic Theory. In Kenneth Hale and Samual Jay Keyser (eds.), The View from Building 20: Essays in Honour o/Sylvain Bromberger, pp. I-52, Cambridge, MA: MIT Press. Chomsky, Noam. 1 994. Bare Phrase Structure MIT Working Papers in Linguistics. pavey, Brian A. and Hilary A. Priestley. 1 990. Introduction to Lattices and Order. Cambridge: CUP. . Heinz-Dieter, JOrg Flum and Wolfgang Thomas. 1 996. Einfiihrung in die mathematische Logik. Heidelberg: Spektrum, 4th edn. Ginsburg, Seymour. 1 975. Algebraic and Automata- Theoretic Properties of Formal Languages. Amsterdam: North-Holland . Mark and Edward P. Stabler Jr. 1 995. Topics in Principle Based Parsing. .. Johnson. . Ms. (January 1 995), Brown University and UCLA. 'Kayne, Richard S. 1 994. The Antisymmetry 0/ Syntax. Cambridge (Mass.): MIT Press.

330 Han�-Peter Kolb

May, Robert. 1 985. Logical Form. Cambridge (Mass.): MIT Press. Partee, Barbara H., Alice ter Meulen and Robert E. Wall. 1 990. Mathematical Methods in Linguistics. Dordrecht: Kluwer. Rogers, James. 1 994. Studies in the Logic of Trees with Applications to Gramm ar Fonnalisms. Ph.D. thesis, University of Delaware.

When to pied-pipe and when to strand in San Dionicio Octotepec Zapotec

I many of the prepositional languages of Mesoamerica, a wh-phrase (can) . n ) occ .. ur to the left of preposition. This is illustrated in ( 1 ) for San Dionicio Zapotec (SDO Zapotec), taken from Broadwell (200 1 , 2005): 2 XhH cim

d Hny Juaany beh' cw? what with PERF-hit Juan dog ' What did Juan hit the dog with?' �i'!i�·��§:;'

u

-

:The derivation of ( 1 ) seems rather straightforward. The wh-phrase fronts to PP, supporting the process that Henk so convincingly argued for in 1 978 book, and pied-pipes the PP to Spec, CP. Aissen ( 1 996) proposes analysis along these lines for Tzotzil. In essence, PP internal wh­ brings the wh-phrase high enough into the PP to enable checking of wh in Spec, CP, through cyclic spec head aweernerlt. Why the P cannot be stranded by extracting the wh phrase from PP remains unclear in this analysis. This squib examines what moves to Spec, PP and what strands when where in SDO Zapotec. SDO Zapotec presents an interesting puzzle, jlllLlStJ�atc:d for postnominal possessors below, but reproducible in other ,environments as well. Possessors are postnominal: 3 ) This squib is dedicated to Henk van Riemsdijk, my first syntax teacher, thesis

', advis()f., and

prepositions.

friend. It could have been

on empty verbs, but it ended up being about

� SDO Zapotec is Otomanguean language spoken in Oaxaca.

Aaron's Broadwell

�aneties

All

of Zapotec presented at the syntax semantic seminar at

2005 . I fol low Broadwell 's glossing conventions, except that

and

data come from

(2000) and his extensive handout on pied-piping in different

com(pletive) as

PERF(ective).

I

UCLA in May p- as poss­

gloss

Broadwell expresses doubts that "a structural

account for the observed variability can ever be presentedU and develops an OT

. account. This squib sketches a structural account. 3 The possessed

glossed

as

NP in alienable possessive constructions is preceded by a prefix x,

poss. The string is in all probability derived from a relative clause

structure containing a nominal small clause, as in Kayne

( 1 994)

332 Hilda Koopman

(2)

x-

peh 'cw Juaimy

poss-dog Juan 'Juan's dog'

When the possessor is questioned, the wh-phrase must appear at the left edge of the DP, and pied-pipes the DP into the left periphery of the clause. wh-phrases obligatorily front to CP. (3)

TiJiJ x-peh'cw cu'a

JuaAny? who poss-dog PERF. grab Juan 'Whose dog did Juan grab?'

Even though the wh-phrase moves to the left edge of the DP, further extraction appears to be blocked as the possessed noun cannot be stranded: (4)

* Tuu cu'a

Juaany x- peh'cw ? who PERF.grab Juan poss-dog 'Whose dog did Juan grab?'

So far, these data hold no surprises. However, when 'whose dog' is embedded in a possessive construction, the possessor must ' strand' the possessed, and surprisingly pied-piping is excluded: (5)

(6)

Tuu x-cutoony

x-miehgw nde'? who poss-shirt poss-friend this ' Whose friend's shirt is this?'

*?T66 x-mtehgw x-cutoony nde'? who poss-friend poss-shirt this 'Whose friend's shirt is this?'

These patterns illustrate the following puzzle. A wh-possesso� must move to the left periphery DP internally (who POSS-friend). It cannot escape to Spec,CP (*who did you see POSS-jriend), but must escape to Spec,DP (who POSS-shirt POSS-jriend). Furthennore, pied-piping is forced for moving to Spec, CP, but excluded when moving to Spec, DP ( *[who Poss-friend] poss-shirt). How can these patterns be made to fall out from a restrictive theory ofUG? (i) «(JeIm x- dog]j [[Johni Isc Jelm;�j

When to pied-pipe and when to strand 333

I will sketch how these patterns fall out from the order of merger, and basic spell-out properties of SDO Zapotec. Two theoretical assumptions play an important role. The first one is that all word formation is syntactic in the sense of Koopman and Szabolcsi (2000). Thus, a bimorphemic wh­ like who is build in the syntax from a wh head and (silent) someone (or whatever the relevant structural parts will turn out to be). (Silent) ;::some�')ne moves as a phrase to Spec, who The phonological form is selected .. on the basis of the syntactic output. The second theoretical assumption concerns what merges where, and departs from the traditional view of how : .:==::: constituents are build up. In particular, a verb does not merge directly with DP, but with NP and all functional categories are merged outside VP (D [vp V[NP]]) (Sportiche ( 1 996, 2002, 2005). The surface constituency of a DP is derived through movement (NP merges with D and VP moves around). PPs are not underlying constituents with P merging with DP, (Kayne, 1994, These views converge in important respects with Williams' s (2003 ) ': R<3Prc�SeJl1ta1tion Theory. In terms of solving the puzzle at hand, this implies just understanding where the wh-head of who/what/which is located ,with respect to other nominal functional categories, but also its location With respect to other heads or constructions, in the case of this squib Ps, complex possessives. I develop these ideas below, starting with PPs. SDO Zapotec is prepositional. A wh ph rase can precede or follow most Ps (Broadwell reports preference for the wh-P order). -

a. XhH dejts zuu beh' cw? what behind stand dog 'What is the dog behind?' b. Dejts xhif zuu beh'cw? behind what stand dog 'Behind what is the dog?' a. Xhii c�n u-diiny Imiimy beh'cw? what with PERF-hit Juan dog 'What did Juan hit the dog with?' b. elm xhil il-diiny Juaany beh'cw? with what PERF-hit Juan dog 'What did Juan hit the dog with' 'Some Ps however can only precede the wh-phrase (dehspuehehs 'after',

' ,�axt 'toward', dantem 'before', and zi 'cy ' like'.

334 Hilda Koopman

If who and what are built in the syntax, their spell-out position minimally reveals the location of the wh-head, a.k.a. the wh-feature. We can thus take the surface position of who and what as an indication of the location of the wh-head, as initial hypothesis. We can construe the difference between the a and b examples in (7) and (8) as one reflecting the 4 height at which different Ps can be merged , with the wh merging higher than the lowest P position. The following order of merger correctly captures that all Ps precede DP complements, some Ps follow wh-words, and some Ps precede them.

(9)

P>wh>P>D

What determines where a particular P can be merged? Broadwell (2005) 5 argues that the native body part Ps are nominal in character: lefs assume that these (as well as with) can be merged lower than wh, in what I will call 6 PN, as well as at the higher P leve1 . For Ps that can never follow wh­ phrases, the lower P position seems simply never available, perhaps for principled reasons: since these are mostly temporal Ps, perhaps they can only be merged at the 'temporal' level, i.e. at a quite high level of representation. This yield the following structures (recall that wh here represents the morphosyntactic wh-feature) : ( 1 0)

[ a. [ something/someone b. [behind/with [ something/someone [ something/someon.e c. [before [P [

wh wh wh wh

[ PN [ [some/one/thing] ]]] ]]] [ behind/with ]]] [ ]]] [

Thus, there is no movement to Spec, PP, but there is movement to Spec, who There is no movement of a whP, but of some smaller phrase that makes

This is consistent with the fact that the internal structure of PPs consists of many layers of functional structure (Van Riemsdijk, 1 990, Koopman ([1 993], 2000)), and the mounting evidence that PPs are not merged as constituents Kayne (1 994, 2000, 2004). S See Brook Lillehaugen (forthcoming) for the behavior of these Ps in another variety of Zapotec. 6 This analysis in essence translates Broadwell's constraints rankings in structural hierarchical terms. 4

When to pied-pipe and when to strand

335

up a whP. The remainder of this squib focuses on the interaction of PN with · the D and wh environmene . Where is the wh-phrase spelled out? Is it sitting in Spec, wh, or does it spell-out the features of the wh-head which agree with the silent phrase in ; Spec, i.e. is it more like an agreeing compiementizer?

[ who

wh [ behind <spell out someone as who if in Spec, wh>

b.

[ someone wbo[ bebind <spell out +wh as who if +animate (via agreement»

There are two arguments in favor of the agreeing head hypothesis as in (lOb). First, even though the wh-phrase precedes the P, it cannot escape from the PP, and pied-piping is obligatory. Xhii cim

b.

u-diiny Juaimy beh'cw? what with PERF-hit Juan dog 'What did Juan hit the dog with?' *Xhii u-diiny Juaany beh'cw cim? ' What did Joan hit the dog with?'

wh-word is an agreeing head, as in ( l Ob), ( 1 2a) follows Straightforwardly. The minimal constituent that must move into the left "· periphery of the clause must include wh, and that constituent includes also i · .ffiicIUides the lower P. ( 1 2b) simply cannot be formed: the wh-head cannot be split from the structure it dominates, i.e. there is no way to strand the complement. Silent someone could in principle extract into the left periphery of the clause since it is a phrase, stranding what with in-situ . . However, someone does not contain the wh-head that the wh-head in the CP periphery needs. Therefore, who or what cannot remain in-situ. Pied-piping ., is thus obligatory. A second argument for treating who/what as an agreeing head comes from the form and behavior of which phrases, which I take to be (minimally) composed of wh +D. Which occurs at the left edge of the DP, ,

·

,7

The surface patterns with respect to the higher P levels, as described in Broadwell (200 1 ) follow, and will be no further discussed in this squib.

336 Hilda Koopman

pointing to wh>O, and its form is homophonous with bare who (tuu) or 8 bare what (xhii) depending on the animacy of the NP .

( 1 3)

Xhii cyaag cu 'a Juaany? what stick PERF.grab Juan 'Which stick did Juan grab?'

( 1 4)

Tuu beh'cw cu'a JuMny? who dog PERF.grab Juan 'Which dog did Juan grab?'

This suggests that all instances of who and what are wh-heads agreeing in animateness with some constituent in their Spec. Interpretive differences follow from the other structural parts that are present. Let me make the derivation of which- forms more precise. Demonstratives in SDO Zapotec are at the right edge, beh 'ew . re 'dog . . . that', and thus trigger pied-piping to Spec, OP. Let us assume that the D determiner which is part of which shares this property with the demonstrative head: it triggers pied-piping to its left, and agrees in animacy with the D. The remnant DP subsequently moves to the Spec of the wh-head, where it triggers agreement on the wh head. ( 1 5) results from the following (incomplete, i.e. VP movements are not shown) derivation: merge NP, merge 0, move NP to Spec, DP, (spec head) agreement for animacy, merge wh, move OP, spec head agreement, spell out {wh, animate} as mu, spell out NP as beh 'ew . .

( 1 5)

.

I � [ [DP [Danimate]j [ whan;mate [ beh'cw tuu who(=which) dog

To form a wh-question, the minimal constituent that c�rries the wh feature fronts to Spec, CPo

( 1 6)

Tuu beh'cw ciJ 'a Juaany? who dog PERF.grab Juan 'Which dog did Juan grab?'

The same holds in Malagasy: iza 'who' and inona 'what'; zaza ;za ' child which'; trano inona 'house which' .

8

When to pied-pipe and when to strand 337

Nothing special needs to be said to exclude stranding of dog, which simply cannot be derived . Which phrases should behave in the same way as who and what with respect to Ps, as they spell out the same head. This seems correct: com bining a (low) P with which phrases leads to the lower P splitting the DP. Pied-piping is forced because what is in the wh head position:

( 1 7)

Xhii CLIO cyaag u- diiny Jm\any beh'cw? what with stick PERF-hit Juan dog 'which stick did Juan hit the dog with?'

Why is DP internal pied-piping excluded? (1 8)

*Xhii cyaag cim u-diiny Juaany beh'cw? what stick with PERF-hit Juan dog 'What stick did Juan hit the dog with?'

Again, this directly falls out from the hierarchy in (9): low (body part) Ps ' are sandwiched between wh and D. The order which stick with simply ::;kC�Ulnlot be fonned from this order of merger: it would require wh to merge ' below the lowest P, contrary to what we have seen. In other words, ( 1 8) is " not a pied-piping problem. , Let us finally return to possessors, which were the starting point of this ' squib. That wh-phrases must occur to the left of the possessed noun reveals a hierarchy wh> D (cf (3) 'who poss.dog). If this structure is further :embedded in a possessive construction, the possessor must extract, , stranding the possessed NP:

�

(19)

Tilil x-cutoony x-miehgw who poss-shirt poss-friend ' Whose friend's shirt is this?'

(20)

TUll x-cut6ony miehgw nde'? who poss-shirt friend this 'Which friend's shirt is this?'

nde'? this

.According to the logic of this squib, it must be the case that the wh-feature of the lower possessor ca� only merge outside the complex possessive , cpnstruction, and cannot merge within. Silent someone moves apparently , long distance as a phrase:

338 Hilda Koopman

(2 1 )

[ someone [ wh [ poss-shirt [ poss friend [s8me8ne ]]]]

This further predicts that the wh-phrase should appear to the left of the low P: (22)

Tuu cun x-cutoony x-miehgw who with poss-shirt poss-friend 'with whose friend's shirt'

We can now answer why (23) and (24) are excluded: x-miehgw x-ciItoony ode'? [[who poss-friend] poss-shirt] this 'Whose friend's shirt is this?'

(23)

*?Tuu

(24)

*x-cut6ony tUu x- miehgw nde'? poss-shirt who POSS-friend this 'The shirt of whose friend is this?'

These strings can only be formed if wh can be merged inside the complex possessive construction, contrary to our earlier conclusion. Thus, these examples are excluded because the wh-feature must be merged outside the complex possessive construction, not because of a constraint on pied­ piping. The question of how this follows from the architecture of the grammar remains an important one for future research. If this squib is on the right track, however, the answer to this squib's title is completely determined by the order of merge and general constraints on movement. References

Aissen, Judith 1 996. Pied-Piping, Abstract Agreemeni a�d F�nctional Projections in Tzotzil. Natural Language and Linguistic Theory 14, 447-491 . Broadwell, George Aaron. 200 1 . Optimal order and pied-piping in San Dionicio Zapotec, in Peter Sells,ed. Formal and empirical issues in optimality theory. Stanford: CSLI. Broadwell, George Aaron. 2005. Pied piping and Optimal order. Handout of syntax and semantics seminar. UCLA. Julien, Marit. 2002. Syntactic head,; and Word Formation, Oxford University Press. New York

When to pied-pipe and when to strand 33 9

Kayne, Richard S . 1994. The antisymmetry of syntax. Cambridge, Mass: MIT Press. Kayne, Richard S. 2000. Parameters and Universals. New York.: Oxford University Press . Kayne, Richard S. 2004 Preposit ion s as Prob es, in A.Belletti (ed.) Structures and Beyond The Cartography of Syntactic Structures, vol.3. New York, Oxford University Press, pp. 1 92-2 12. Koopman, Hilda. 2000. The syntax ofspecifiers and heads. Rout ledge, London. . Koopman, Hilda. 2005. Korean and Japanese Morphology from a Syntactic .: :. Perspective . in Linguistic Inquiry 36.4. Koopman, Hilda and Anna Szabolcsi. 2000, Verbal Complexes. Current Stud ies in Linguistics 34, Cambridge, Mass: MIT Press Sportiche, Dominique. 1997. Reconstruction and constituent structure, GLOW . abstract. Sportiche, Dom in ique . 200 1 Structural Bonds and Strict Locality, Tools in Linguistic Theory # I , Rijksuniversiteit Utrecht. 'Sportiche, Do m in iqu e . 2005. Division of Labor between Merge and Move: Strict locality of Selection and Apparent reconstruction paradoxes. LingBu zz 000 1 63. RiielIlso.1Jk, Henk van. 1 978 A Case study in Syntactic Markedness: the Binding Nature ofPrepositional Phrases. Faris Publications, Dordrecht. : 'RieITlso:uk, Henk van. 1 990. Functional prepositions, in H. Pinkster and I.Genee (cds), Unity in diversity. Papers presented to Simon C. Dik on his 60th birthday. Dordrecht, Foris, pp.229� 1 . 'Willialms, Edwin. 2003. Representation Theory. MIT Press, Cambridge Mass.

.'

Free relatives as light-headed relatives in Turkish Jaklin KornfiIt

1 . Introd uction !

One of the most clear-cut of the properties that are different for Free Relatives (FRs) and regular, headed relative clauses (HRCs) concerns matching effects: In many languages, such as English and German, matching effects are observed for FRs, but not for HRCs. In English FRs, matching is relevant for category selection only, while in German, a language more richly inflecting for case, matching effects in FRs are observable for case, as well: Kandidaten, dem du vertraust ( 1 ) a. Ich nehme den take 1 theacc candidateacc whOdat you trust 'I (wil l) take the candidate whom you trust' The matrix verb checks for Accusative case, while the embedded verb checks for the Dative. In this HRC, the m ismatch with respect to case does not lead to i ll-formedness. The corresponding FR, however, is ill-formed, no matter whether the Accusative or the Dative is realized: ( 1 ) b. *Ich nehme wernlwen

du vertraust take whomdaJwhomacc you trust ' I (will) take whom you trust' (cf. Van Riemsdijk 2000, ex. 20 b)

I

FRs where matching is observed are perfectly well-formed:

I This paper is written in celebration of, and in gratitude for, Henk van Riemsdijk's immense contributions to syntactic theory and to the generative grammar community in so many ways. I hope that these few pages are worthy of him and of his insights into the nature of Free Relatives. I thank Mark Brown and Hans Broekhuis for their help with the formatting of this paper.

Free Relatives as Light-Headed Relatives 34 1

Ich nehme wen du mlr empfiehlst I take whomacc you medat recommend 'I (will) take whom you recommend to me' (cf. Van Riemsdijk 2000, ex. 20a) both the matrix and the embedded verbs check for the Accusative, thus matching is observed, leading to well-fonnedness. Turkish is a language where matching effects are not observed in FRs: [sevgili-m-in e sev-me-dik-ler-in ]-e yardlm lover- } SG-GEN love-NEG-F.NOM-PL-3SG -DAT help et-me-m do-NEG-AOR+ 1 SG ' I don't help who(m) my lover doesn't like' example, the matrix verb checks for the Dative, while the embedded checks for the Accusative. Does this mean that case matching is not a distintcti',{e property of FRs, after all? Or that this property is parameterized, ' ith Turkish (and similar languages) having a negative setting in this espect? While these are certainly reasonable statements to make cross­ (m�tUH;tlcally in terms of a fonnal typology of FRs, they are merely and should be used only when no explanatory account is lable. I will attempt such an account in this squib. My proposal will, in �ss(:mce, be to analyze Turkish FRs such as the one exemplified in (3) not . genuine, true FRs, but rather as what Van Riemsdijk 2000 (citing Citko . 999) refers to as a ' light headed relative' (LHR). Such constructions are jiS!mOled not to be a true FR, either formally headless or with a head 'OCtcuplied by an empty category (whose nature is itself open to discussion), rather to have a "semantically minimally specified head, generally a i:urlonc)Un" (Van Riemsdijk 2000: 1 6). Therefore, they are not expected to exhibit matching effects. This is illustrated by the following German examples:

§�iE;
Ich nehme den, dem du vertraust you trust whomdat I take the-oneacc 'I (will) take the one whom you trust' (cf. Van Riemsdijk 2000, ex. 20b')

342 Jaklin Kornfilt

(5)

Ich vertraue dem, den du mir me you whoma dat 1 trust the-onedat cc 'I trust the one whom you recommend to me'

empfiehlst recommend

It is not evident that the Turkish construction illustrated in (3) is such LHR; it does not exhibit a pronoun. I will argue that a weak pronoun, i.e. a clitic-like element, in fact does occupy the head position of the apparent FR, thus making it into an LHR. The appropriate translation of (3), under this analysis, would be as follows: "1 don't help those who(m) my lover doesn't like". I will further argue that some additional, apparently disparate facts receive a principled explanation under the LHR analysis for Turkish FRs. an

2. Free Relatives in Turkish, their distribution, and their Agr 2. 1.

'Reordering ' between predicate-subject agreement and inherent plural

Note the order of the morphemes between the inherent p lurality marker -IAr and the agreement marker in (6a): (6) a. [Geyen yaz ada-da gor-duk ler im] bu yaz last summer island-LOC see-F.NOM-PL-lsG this summer gel-me-di(-ler) come-NEG-PAST(-3PL) '(Those) who(m) 1 saw on the island last summer didn't come this summer' -

-

This order is unexpected. The expected order would be for agreement to precede plurality, given the corresponding relative construction with an overt head: (6) b. [[Geyen · yaz ada-da ej gor-dug-lim] insan Iarjl last summer island-LOC see-F.NOM-lSG person-PL 'the people who(m) I saw on the island last summer' -

The same applies to (3), which I am repeating as (7), for the sake of convenience:

Free Relatives as Light-Headed Relatives 343

[sevgili-m-in e sev-me-dik-Ier-in ]-e yardtm lover- l SG-GEN love-NEG-F .NOM-PL-3SG -OAT help et-me-m dO-NEG-AOR+ 1 so 'I don't help who(m) my lover doesn't like' The corresponding HRC would be as follows: [[sevgili-m-in ei sev-me-dig-i ] insan-Iar-ai 10ver- l sO-GEN love-NEO-F.NOM-3SG person-PL-DAT yardtm et-me-m help do-NEO-AOR+ 1 SG 'I don't help those people who(m) my lover doesn't like' In all examples, the agreement morpheme is on the embedded verb and expresses person and number features of the embedded subject. The plurality morpheme, on the other hand, expresses the inherent plurality feature of the head (of the RC). Thus, in a FR without an overt head noun, ,if the plural feature of the head is to be expressed at all, it should show up the agreement morpheme and not before. Why, then, is the original order between agreement and plural morphemes not preserved, i.e. why is the shape of the embedded (nominalized) verb in (6a) not as in (6c), and the shape of the . corresponding verb in (3/7a) not as in (7c)? c.

c

*gor-diig-iim-ler *sev-me-dig-i-Ier

I will attempt an answer somewhat later in the paper. It is obvious, in any event, that some sort of reordering appears to have taken place between the agreement .morpheme on the embedded verb and the inherent plurality morpheme of the head of the entire RC construction. At this point, I would like to tum to a special type of relative clause construction in Turkish-a construction that corresponds to infinitival Res in English.

2.2. 'Infinitival ' Relative Clauses

Turkish does not have infinitival relative constructions; th is is illustrated by the ungrammatical (8):

344 Jaklin Kornjilt

(8)

* Ahmet Akgiil-ej

[DP [C p2/DP PROj ei oku-mak] bir kitaPi]

read-INF a book Ahmet Akgiil-DAT ai-d. buy-PAST Intended reading: 'Ahmet bought Akgiil a book to read'

This is a surprising fact, given that infinitival complements with Control semantics (as well as with the syntactic properties of Control) are very productive in Turkish; e.g.:

(9)

Ahmeti [PROi biT kitap oku-mak] isti-yor Ahmet a book read-INF want-PRPROG (Subject Control) 'Ahmet wants to read a book'

( 1 0)

Ahmet Akgiil-ui [PROi sinema-ya git-meg] -e Ahmet Akgiil-ACC cinema-DAT go-INF -DAT zorla-d. force-PAST 'Ahmet forced Akgul to go to the movies' (Object Control)

(1 1)

Ahmet Akgiil-ei [PROi sinema-ya git-meg]-i Ahmet Akgiil-OAT cinema-DAT. gO-INF-ACC tavsiye et-ti recommend-PAST 'Ahmet recommended to AkgUl to go to the movies' (Indirect Object Control)

Given these facts, the question becomes intriguing as to why infinitival clauses in Turkish, which do seem to parallel their English counterparts so closely, don't allow for RCs. (Interestingly, they don't . allow for wh­ questions either; in other words, there are no embed,ded . infinitival · wh­ questions- matrix questions with a wh-element in an .embedded infinitival . clause are well-formed.) I have proposed elsewhere (e.g. Kornfilt forthcoming) that Turkish infinitivals are a subset of a type of nominalized construction, more specifically of the non-factive, homogeneously nominal embedded clauses. .

2

.

"

There is some evidence that such clauses are not CPs, but just DPs-however, this issue is tangential to our concerns here. A brief discussion is offered in the text between examples ( 1 1 ) and ( 1 2).

Free Relatives as Light-Headed Relatives 345

I have claimed that due to their homogeneously nominal categorial status, these clauses are not CPs; rather, they are DPs. Therefore, they have no specifier position that can serve as an ultimate target position for an operator (however, they do have a specifier that can be an escape hatch, thus explaining why they can host a wh-element in a matrix, but not em bedded, wh-question). This, then, explains why Turkish has no infinitival Res. However, Turkish does have a construction which does bear some resemblance to infinitival RCs; the semantics are similar, and although the modifier clause is not an infinitival, it resembles an infinitival clause, because it has no agreement morphology: ( 1 2)

.

Ahmet Akgiil-ej [DP [C pPROj ej oku-yacak] Ahmet Akgiil-DAT read-FUT al-dl buy-PAST 'Ahmet bought Akgiil a book to read'

bir kitapil a book

Comparing the well-formed ( 1 2) with the ill-fonned (8), we see that they are very similar, with the only difference being that instead of the infinitival morpheme on the embedded predicate in (8), we have a (nominal) Future marker in ( 1 2). Indeed, this marker does double-duty as a regular, fully finite future tense marker in root clauses and as a nominalized :factive/indicative clause marker, when such a clause is in the future tense. Since nominalized indicative clauses are CPs (as argued for in Kornfilt '2003 and forthcoming), there is no problem with their hosting a relative 'clause operator. Since these are not, strictly speaking, infinitival relative clauses, let us call them irrealislpotentiality Res. 2. 3.

Irrealis Res cannot be FRs

Let us now return to the issue of FRs and their distribution. We note that t •. hese irrealis RCs cannot serve as FRs. * Ahmet Akgiil-e

[DP [CP PRO ej oku-yacak]] al-dl Ahmet Akgiil-DAT read-FUT buy-PAST Intended reading: 'Ahmet bought AkgUl what (something) to read'

propose the following constraint, to explain the iII-formedness of such examples:

I

346

( 1 4)

Jaklin

Kornfilt

In Turkish complex NPs (i .e. N-headed NP/DPs with overt modifiers and/or specifiers), the N must be overtly filled.

In regular FRs (whose modifier clause is, in fact, the indicative type), the Agr of the modifier clause raises, I propose, into the empty N head position. (For arguments that Turkish FRs indeed have such a head position, cf. Kornfilt 1 995). In irrealis RCs, however, the modifier c lause has no Agr, and thus nothing can raise into the head position. (Notice that Agr in the well-fonned instances is nominal, and thus categorially compatible with the N-head. There is nothing else in the ill-formed constructions which would be similarly compatible with the N.) Note that this analysis also explains the morpheme 'reordering' we saw in the previous section. The nominal Agr element raises out of the modifier clause to the N/D-head position of the head-DP of the entire construction. Since that head hosts the inherent plurality morpheme (in those instances where the head is indeed plural; the singular is not marked morphologically), we see the raised Agr fol lowing the plural marker which is, in the original structure, h igher in the phrasal architecture than Agr. 3 There is one problem, however, that this account faces. This has to do with FRs whose target is a subject. 3. FRs with subject targets

The previous section proposed an explanation for the ill-formedness of irrealis FRs via the constraint in ( 14), and the fact that irrealis RCs have no Agr (nor any other element) that could raise in such a way as to satisfy that constraint. This would predict that FRs can never take any kind of modifier that lacks overt Agr. This is not so. Note that all of our examples so far had non-subject targets. Let us now il lustrate a subject-targeting HRC and then see if it can have a corresponding FR: 3

The constraint I just proposed, as well as the morphological and phrase-structural independence of Agr in Turkish, and the categorially nominal feature of nominal Agr, receive additional support from certain partitive constructions . Unfortunately, constraints of space don't permit me to discuss those constructions and facts here; I refer the reader to Komfilt (200 1 ) and to von Heusinger and Komfilt (forthcoming).

Free Relatives as Light-Headed Relatives 347

', ' ( 1 5)

] insan-Iaril ada-ya gel-en [[ei ge�en yaz person-PL last summer island-DAT see-RELPART. 'The people who came to the island last summer�

Note that here, the embedded nominalized verb has no agreement mo rphology. Elsewhere (e.g. Komfilt 1 99 1 , 2000), I have argued that this :is due to an 'anti-agreement' effect, motivated by generalized binding constraints, yielding a constraint against resumptive pronouns; pro, licensed and identified by overt Agr (in this null subject language), is one of pronoun and thus can function as a (licit or illicit, depending on the syntactic context) resumptive pronoun. Given such lack of overt agreement morphology, we would expect that subject-targeting FRs should not exist. However, they do: [[ej

ge�en yaz ada-ya gel-en-1eri] last summer island-DAT see-RELPART.-PL ' (those) who came to the island last summer'

, We saw previously that the morpheme for the inherent plural of the head is sufficient to satisfy the constraint in ( 1 4), and that overt nominal agreement morphology is one way of satisfying the constraint. What, then, satisfies the constraint in example ( 1 6)? Notice that the nominalization morpheme of subject-targeting Res is :. 'dlitlc�rell1t from that of Res with non-subject targets. In other words, the two of Res differ not only in exhibiting versus prohibiting agreement '<,m.oIl)ho'log�y on their embedded predicates, the shape of those predicates is �1" j::g§§�", ,different, as well. While the predicate of Res with non-subject targets is of ,the same form as any embedded nominalized indicative, the predicate of ""':;; : "�..."....,,� with subject targets has a special shape. What does this shape represent, and why can't we use the general, unmarked indicative form for subject-targeting Res? I have suggested elsewhere (cf. Komfilt 2000, 2004) that this is due to , something quite similar to complementizer agreement phenomena in better­ studied languages such as West Flemish, Bavarian, and French (the que -) qui phenomenon). Thus, the special form of the predicate in subject­ targeting FRs, by virtue of being a nominalized form and in fact by virtue :of being an agreeing nominal form, can raise to the head position of the :FR, thus satisfying the constraint in ( 1 4) and leading to well-formedness. In contrast, the ungrammatical instances of irrealis FRs, i.e. the construction

348 laklin Kornfilt

illustrated in ( 1 3), do not bear any such agreement form that might be able to raise to the head position of the FR. Why, then, does the special nominalized predicate form in subject­ targeting FRs not get reordered with the inherent plurality marker of the head? The answer lies in the independence from each other of syntax and morpho-phonology. The overt Agr in non-subject targeting FRs is not only syntactically mobile, i.e. it can raise, but it is also morpho-phonologically mobile, i.e. it can reorder locally, within the same syntactic position (which is, here, the head N position of the RC). In other words, the syntactic (head) raising within the phrasal architecture makes the sequential ordering possible. What makes this reordering necessary is a morpho-phonological constraint of morpheme orders, such that case morphemes must follow agreement morphemes rather than plural morphemes, when both agreement and plural are realized. (This constraint is evident in this-morphologically very rich-language in general.) Since the special nominalized predicate in subject-targeting FRs is a syntactically agreeing form, but does not belong to the morpho-phonological agreement forms, it does not need to obey the sequential ordering constraint; however, by virtue of being a syntactic agreement form, it is able to satisfY the syntactic constraint in ( 1 4) that requires an overtly filled head noun for the entire Re. As a final observation in favor of the proposed LHR analysis for Turkish FRs, I would like to mention the fact that these constructions are generally interpreted as definite FRs, rather than as universally quantified FRs (cf. Van Riemsdijk 2000:23). The latter interpretation is possible only with additional adverbs and verbal aspe�tual markings that make such an interpretation possible. The unmarked interpretation, especially without such additional markings, is that of a definite FR If the head position is filled with a nominal element, the prevalence of such an interpretation is not surprising. References Citko, Barbara. 1 999. Light-Headed Relatives. In Proceedings of the 23rd Annual Penn Linguistics ColloqUium; University of Pennsylvania Working Papers in Linguistics 6: 1 ; 1. Alexander, N.R. Han and M.M. Fox (eds.); Philadelphia: University of Pennsy 1 vania Heusinger, Klaus von, and Jaklin Komfilt. Forthcoming. The case of the direct object in Turkish: Semantics, syntax and morphology. To appear in Turkic Languages; Wiesbaden: Otto Harrassowitz.

Free Relatives as Light-Headed Relatives 349

. Komfiltt Jaklin. 199 1 . Some current issues in Turkish syntax. In Turkish Linguistics Today. H.Boeschoten and L.Verhoeven (eds.); Leiden: Brill Publishers; 60-92. Komfilt, Jaklin. 1 995 Con straints on Free Relative Clauses in Turkish. In FAS Papers in Linguistics 4, A.Alexiadou, N.Fuhrhop, P.Law and S.Lohken (eds), 36-57. Berlin: FAS. Komfilt, Jakl in . 1 996 Naked Partitive Phrases in Turkish; in Partitives. J.Hoeksema (ed.); Berlin: Mouton-de Gruyter; 1 1 1- 125. Komfi lt, Jaklin. 2000. Some syntactic and morphological properties of relative clauses in Turkish. In: The Syntax ofRelative Clauses. A.Alexiadou, C. Wilder and P .L aw (eds .) ; Amsterdam/Phi ladelphi a : John B enj amins; 1 2 1 - 1 59. Komfilt, Jaldin. 200 1 . Non-specific partitives and the unreliability of specificity markings. Pap er presented at the LSA annual meeting, Washington, D.C. Komfilt, Jaklin. 2003. Subject Case in Turkish nominalized clauses. In Syntactic Structures and Morphological Information (= Interface Explorations; 7).

U.Junghanns and L.Szucsich (eds.); BerlinlNew York: Mouton de Gruyter, 1 29-2 1 5. Komfilt, Jaklin. 2004. Unmasking covert complementizer agreement. Paper presented at the LSA annual meeting, Boston. Komfilt, Jaklin. Forthcoming. Agreement: The (unique and local) syntactic and morphological licenser of subject Case. In Studies on Agreement. ICosta and M.C.Figueiredo Silva (eds.). Amsterdam and Philadelphia: John Benjamins. Riemsdij k , Henk C. van. 2000. SynCom Case 44: Free Relatives. Ms. University of Tilburg.

Is linguistics a natural science? Jan Koster

1 . Language: internal or external?

Although language uncontroversially has internal, psychological dimensions and external, cultural aspects, the focus of modem linguistics in the Chomskyan tradition has been primarily on the internal dimension (I-language, Chomsky 1 986). It has often been seen as part of cognitive psychology, ultimately as a form of theoretical biology. From Lenneberg ( 1 967) to Jenkins (2000), the 'biolinguistic' perspective has been the leading thought. According to this view, a substantial part of language, particularly recursive core grammar, is the abstract description of a ' language organ'. In this view, the human mind and the brain must be identical at some level, witness frequent use of terms like the mind/brain . It is expected (or hoped for), according to the biolinguistic perspective, that ultimately the study of I-language can, at least in principle, be unified with the core natural sciences (biology, chemistry, physics). I will refer to this standard view as linguistic naturalism. Linguistic naturalism is not without critics, particularly outside the community of generative grammarians in the strict Chomskyan tradition. Thus, many continue to adhere to the traditional, cultural view of language, according to which language primarily is part of our cultural record and external to the human mind (E-Ianguage). Another important E-perspective on language is the linguistic Platonism advocated by the late Jerrold Katz and others (see, for instance, ,Katz 1990 and the works cited there). Personally, I think a fully naturalistic approach even to. core grammar is untenable. I am a moderate Platonist and also committed !to ' aoversiori of the traditional view that °language is in the first place" a cultural phenomenon, crucially depending on a supra-individual, external record. I furthermore believe that the mind, unlike the brain, cannot be seen in isolation from the shared, external memory with which the brain lives in symbiosis (see Donald 1 99 1 ). This external, cultural record determines how our biologically given resources are applied. A good example is playing the piano. It is of course based on our biological capacities: tone perception, memory, the physiological and neurological aspects of finger movement,

Is linguistics a natural science? 35 1

etc. Nevertheless, piano playing is primarily seen as a cultural activity. It involves a tradition of musical composition and, crucially� the invention of an artefact, the piano� which allows us to integrate our biological capacities · with the culture we participate in. Clearly � without the invented artefact and tradition, there would not be . anything biological that could be reasonably called 'the biology of piano playing. ' There have been numerous cultures without pianos� aJl with · · people similarly endowed as we from a biological point of view. In general, biology is not culturally transparent. Cultural phenomena heavily depend on our biological capacities, but cannot be 'read back' into our biology. Doing so, was the big m istake of sociobiology and also of its successor, evolutionary psychology, which claims that our biology is transparent with respect to the cultural exigencies of hunter-gatherer societies. As I have argued elsewhere (Koster 1 989), a fully biolinguistic perspective suffers from the same confusion. We can only speak of language because of the invention of cultural artefacts, words, which are comparable to the piano. Only thanks to the interfacing properties of our . cultural invention we can speak of language. Words connect our abstract " computational facilities with our concepts, neither of which have anything to do with language in abstraction from our invented interface elements. As for the relation between biology and culture, I see no logical difference :between playing the piano (and the rest of our culture) and the use of language. Language is of course easier to acquire (no doubt facilitated by evolution) and based on much more accessible structures than playing the :piano, but that is irrelevant for the logic of the situation. All cultural phenomena differ in this respect among each other. Thus, singing a simple song involves much more accessible and universally available biological . capacities than p laying, say, Bach 's Goldberg Variations. In what follows, I will focus on another possibly external and important aspect of language, the Platonic world of universals. Like the argument from culture, the reality of universals leads to the conclusion that · ilinguistics cannot be reduced to, or unified with, the natural sciences. Natural language, then, is E-Ianguage in two crucial aspects: next to our 'mind-external cultural record, it accesses abstract external structures, just ;;;:· ' 1 1 ICP math ematics. · ,2. Universals and particulars

live in a dual world: on the one hand we have the particulars of the :physical world and on the other hand we have the world of general terms,

352 Jan Koster

which determine our thought processes. Even small children understand the distinction 'type-token' as they apply terms l ike 'dog' to each dog, not just to one dog at a particular place or time. Even proper names, like 'Mary', are universals, since we apply the name in a potentially infinite number of circumstances in which Mary appears somehow. The world of our thought is exclusively based on universals, while the physical world consists of particulars. This distinction between universals and particulars has always been one of the core problems of philosophy: how are particulars related to universals? The classical answers were given by Plato and Aristotle. According to Plato, universals (or forms) have an existence independent of particulars and the human mind (universalia ante res). Particulars, in this view, are intelligible to the extent that they can be seen as imperfect instantiations of universals accessed by the m ind. The universal fonns are timeless and necessary and, therefore, even constrain the creator (or whatever determines 'becoming' in our world). Platonism entails a dualistic, two-world ontology. Plato's student Aristotle thought that Plato's theory was not parsimonious enough and that there was only one world. He furthermore criticized Plato for failing, given his two world-view, to elucidate how the two worlds (of universals and particulars) were connected. This second criticism is certainly to the point, but Aristotle's alternative theory was a dismal failure, causing philosophical confusion until the present day. Basically, Aristotle denied the dual nature of our reality, making him, at least in Western philosophy, the father of one-world ontologies. He sought to overcome the perceived problems with P latonism by reinterpreting un iversals as somehow being part of · particulars (universalia in rebus). This, however, does not make sense. Since we are talking about a philosophical debate of over 2000 years, there is no point in even beginning to summarize the objections raised against Aristotelianism over time. So, I will limit myself to a very few personal favorites, often without remembering exactly where I learned about them for the first time. First of all, by making universals part of each particular, the number of universals becomes equal to the number of particulars. This seems to annihilate the concept of a universal somehow. Furthennore, even in the simplest cases universals involve properties that are not instantiated by particulars in any obvious sense. Consider a simple geometric fonn, like a straight line. The length of a straight line can be infinite and therefore be longer than any line found in the physical universe. Similarly, a geometrical

Is linguistics a natural science? 353

straight line has no specific thickness, while each physically realized line has. In fact, the vast majority of mathematical structures has no realization , in the world of physical particulars. The same is true for fictional universals like 'unicorns' . One could of course extend the world of particulars from physical 'particulars to non-physical particulars and claim that universals define a potential world distinct from the actual, physical world. But in that case the , 'argument against Plato's two-world vision would collapse: Aristotle would have to assume more than one world as well. Universals are not part of the physical world but perfect forms that are normative with respect to the often less than perfect physical world. ,Crystals, for instance, show geometrical shapes and are said to be more or , less perfect, depending on the extent to which they approach the ideaL • ,More generally, we can say that universals, in relation to physical objects, are not belonging to the object world itself, but to the meta-world in which things are compared with each other and with some standard external to the objects at issue. Aristotelianism, it seems, is a hopeless conflation of . level perspectives and meta-level perspectives. The intense medieval debate on universals left Western philosophy since the 1 7'h century with a heritage of two major world views: rationalism empiricism. European rationalism, much inspired by Descartes and . . in the philosophy of Kant, can be seen as an epistemologization of Platonism. Classical Platonism is ontological, i.e., it )I'0stulates the reality of universals in a realm independent of the human " mind. According to European rationalism, there is no certainty beyond the }human mind and its categories of understanding. Arguably, also Frege's alleged Platonism is of this epistemological k ind, which was influenced by ' Kant but particularly by the 1 9fh-century German philosopher Hennann ' :Lotze (see Sluga 1 980). However, it should be stressed that European rationalism is not ' naturalistic, in the sense that it explicitly denies that our categories of understanding can be reduced to empirical psychology or biology. Famous , in this respect is Husserl's attack on psychologism in mathematics, which tf,�;::-�w ' :Was inspired by Frege's critique on Husserl's earlier work (see Baker and f@�;Si;!L �Hacker 1 984: pp. 4 1 ff.). It is very interesting from this perspective that �[�IT�;�i�:�" :Chomsky, on the one hand, sees his work in the tradition of European 'rationalism, while, on the other hand, he advocates a naturalistic, �> ;)�::;!{:H,jpsychological/biological approach to language and perhaps to cognition in i: �;,:;,;);, ;:general. {

r¥"m!���f;i,

'

3 54

Jan

Koster

Before going into this matter, I would like to say that, although I accept the critique on psychologism, I do not find said epistemologization of Platonism convincing in the long run. Motivated by a now outdated quest for certainty, it seems to pull the whole world too much within the purview of the human mind. By way of reductio ad absurdum, this easily leads to the megalomanic subjectivism of solipsistic Idealism, at the cost of the common sense view that there exists an objective world independent of the human mind. Whatever consequences were drawn from Descartes' cogito ergo sum. they do not seem to be less hypothetical and remote from absolute certainty than the working hypothesis that a fullfledged world existed before (and will exist after) the short period of time I am aware of it. If we want to be Platonists, I do not see strong reasons to deviate from the original, ontological version. So, where does the current idea come from that human cognition (including language and, perhaps, mathematics) must be studied from a fully naturalistic point of view? In my opinion, it is a residue of empiricism and materialism, which were (and are) ideologically motivated world views, which are often -wrongly- seen as the core of the scientific world picture. For empiricists (and their nominalistic predecessors) the only reality is the reality of particulars, which for them creates 'the problem of universals. ' This is a curious metaphysical tum, because in a sense, if anything, there is a problem of particulars. Universals are more elementary in our experience, because they are the immediate building blocks of our thoughts and our theories. Particulars are not immediately given in experience but, in the physical science�, form the hypothetical reality that Qur thoughts and theories are about. So, what is the status of universals in a metaphysical framework that only recognizes particulars as real? Traditionally, empiricists have seen universals either as convenient fictions or, at best, as somehow derived from particulars. I wil1 argue in a m inute that full naturalism in cognitive science is based on a variant of the empiricist idea that universals can be derived from particulars. The standard approaches to the problem sought to explain the derivation of universals by mysterious processes with names like ' abstraction', 'generalization' or ' induction ' . To say that general terms (universals) are derived by ' generalization' is entirely circular, in the spirit of the doctor in Moliere's Le rnalade imaginaire, who claims that opium causes sleep thanks to the fact that it contains a substance called vi... dormativa ' sleep force' . Since there are no successful ways to derive

Is linguistics a natural science? 355

univ ersals from particulars, empiricists tend to be in denial about universals, calling them 'fictions', etc. I will take the reality of universals for granted here, as do Chomsky and many other current advocates of naturalism in cognitive science. Granting that full reductionism is unlikely, latter day naturalists believe that, at least in principle, it is possible to unify theories about physical reality with theories about our mental capacities. Note, however, that unification in the physical sciences, say between physics and chemistry, involves theories that are all exclusively about particulars. This is an area where empiricists and rationalists agree, because for empiricists particulars are the only reality to begin with, whereas rationalists see physical phenomena also as particulars, be it as instantiations of universals. Full unification of physical and cognitive science would involve something totally unheard of, namely unification of theories about particulars (physics) and theories about universals (for instance, linguistics). It would entail the belief that causal chains of particulars could ; ... . . ....... &<.. the way somehow mysteriously produce the universals of our mind. this 'causal chain' view would be just a variant of empiricist notions like generalization, abstraction or induction and therefore, in my opinion, . doomed to be a failure. In other words, the causal chain (what naturalism comes down to) is the latest guise of the Aristotelian error to see particulars and universals in one plane of reality. Actually, as argued above, thinking and the world of universals are lly different from the physical world of particulars in that . thinking involves a meta-perspective on the physical world. Unification of an object perspective (physical sciences) and our meta-perspective seems impossible and we therefore seem to be destined to remain dualists-in­ practice for ever (see Bracken 2002 for a related view). This dualism-in­ practice should not be confused with classical substance dualism. If there is an epistemological barrier to see the physical world and the meta-world of thought in one plane, the ontological question of substances does not even . 'arise. Given our epistemological limitations to understand the relation between universals and particulars, belief in ful l naturalism in cognition is a form of dogmatic monism and in fact an unprovable metaphysical position, .,... . unsubstantiated assumption about the ultimate nature of reality. This is . o',:: · "' ; . i""-." 'as pointless, I believe, as all other beliefs about the total of reality, like , idealism, or the idea that all is text (certain postmodemists) or . created by God (most religion). If one dogmatically insists on full Jlaturalism, it can only seemingly be upheld by making a distinction

3 56 Jan Koster

between 'problems and mysteries', the mysteries being a term for what is behind our epistemological horizon (Chomsky 1 9 7 5 : ch 4) It seems to me that the relation between universals and particulars is a prototypical example of such a rational mystery. Aristotelianism and its nominalistic and empiricist offspring failed to make universals intelligible 'from below', i.e., by starting from particulars. Platonism sees the relation 'from above', i.e., by assuming the distinct reality of universals and by seeing particulars as imperfect instantiations of these universals. I agree about the distinct reality of universals. But is Platonism successful as a theory about the relation between universals and particulars? U ltimately, the relation is a mystery for Plato as weB, because he only gives metaphorical and mythological hints, like the myth of the copies seen in the cave or the idea that our mental access to universals is a matter of reminiscence of an earlier life. Rationalists criticize empiricists for their failure to account for universals on the basis of particulars. But I think it is often overlooked that the problem situation is symmetrical: there is no successful account of the relation seen 'from above' either. For me at least, Platonism is the most successful perspective on universals, but it leaves it a complete mystery how our brain can access these universals and how they can be recognized in particulars. As soon, then, as one (against the empiricists) believes in the reality of universals, the distinction between 'naturalism cum mysteries' (Chomsky) and 'Platonism cum mysteries' (Katz) should not be exaggerated. I see Chomsky'S naturalism as a kind of naturalized Kantianism, in which the Kantian objections against a naturali�tic approach of our categories of thought are met by the introduction of an epistemological horizon (the 'mysteries' of Chomsky 1 975: ch. 4). But Kantianism, it should be remembered, was the culmination of the century-long attempts of European rationalism to replace Plato' s ontology by . a more epistemological perspective. As argued above, this epistemological tum . was inspired by an ill-conceived quest fot philosophical certainty; ·which, 1 believe is ultimately futile. .

.

. .

3. Concluding remarks Given our limited epistemological horizon, in practice, the main difference between Platonism and naturalism will be that according to the Platonist, the brain somehow accesses the (external) universals, while the naturalist will say that universals are mind-internal and caused by the brain. I doubt

Is linguistics a natural science? 357

if these two positions are empirically distinguishable and, if I am correct, they both run into the barrier that prevents us from seeing universals and particulars in a unified, one-world theory. This leaves us minimally, once more, as dualists-in practice, with no definitive conclusion about the · ultimate nature of our reaJity. If we look at our scientific practice, the dual nature of our reality is confirmed by the existence of mathematics. Not the physical sciences, but mathematics is our most successful form or rational inquiry. The development of mathematics looks like the development of the empirical · sciences in important respects, but is about universals and not about particulars. In this sense, it is a Platonic science and mathematics has, understandably, always been the field of choice of the Platonically inclined. We need mathematics to understand the physical world, not the other way around. Naturalism, in the sense of reduction or unification of mathematics with physics, is not even considered. Why would naturalism be more plausible with respect to linguistic computation than with respect to the patterns of necessity studied by mathematics? Like mathematics, linguistics studies universals, such as our concepts or the recursive structures of grammar ('merge'). Recursion is an instance of self-similarity, a form of symmetry ubiquitous in the extra­ mental, physical world. Patterns of symmetry are the subject matter par of mathematics. Hence, I believe that, as for concepts and linguistics is a form of inquiry in the same league as studying universals rather than the particulars that are the · subject matter of the natural sciences. This view also seems to be confirmed by the actually existing brain At best so far, we are successful with localizations, i.e., by determining the part of the brain where certain computations take place. as we know from computers, indicating where in some hardware ice computation takes place is very remote from unifying the theory of computation and the physics of the hardware. Similarly, it is an error an illusion to believe that the brain sciences -naturalistic theories about particlulars- can explain our thought processes based on the universals by mathematic and linguistics. As in the relation between nathernatlcs and physics, I believe that linguistics is like mathematics in it is more fundamental than whatever the brain sciences come up with . . sties can guide us in understanding where and how the brain �xe(�ut(�S certain computations, but what makes computations computations for that matter, what makes a triangle or a number a triangle or a nUlno(�r) is beyond the possible scope of the natural sciences.

·

358 Jan Koster References

Baker, Gordon , and Peter Hacker. 1 984. Frege: Logical Excavations. Oxford, etc.: Oxford University Press. B racken, Harry . 2002. Descartes. Oxford: One World. Chomsky, Noam . 1 975. Reflections on Language. New York: Pantheon Books. Chomsky, Noam. 1 986. Knowledge o/Language. New York: Praeger. Donald, Merlin. 1 99 1 . Origins of The Modern Mind. Cambr idge, Mass.: Harvard University Press. Jenkins, Lyle. 2000. Biolinguistics: Exploring the Biology 0/ Language. Cambridge: Cambridge University Press. Katz, Jerrold . 1 990. The Metaphysics o/Meaning. Cambridge, Mass.: MIT Press. Koster, Jan. 1 989. How Natural is Natural Language? in J.E. Fenstad et al. (eds. ), Logic, Methodology, and Philosophy of Science VIII. Amsterdam, etc.: Elsevier. Lenneberg, Eric. 1 967 . Biological Foundations of Language. New York: John . Wiley and Sons. Sluga, Hans. 1 980. Gottlob Frege. L ondon, etc.: Routledge and Kegan Paul.

Two asymmetries between Clitic Left and Clitic Right Dislocation in Bulgarian lIiyana Krapova and Guglielmo Cinque

1. Introduction

Despite the substantial identity in syntactic properties between Clitic Left Dislocation (CLLD) and Clitic Right Dislocation (CLRD), and the prospect of deriving the latter from the former (Kayne 1 995, Samek-Lodovici 2005), a prospect which we also find attractive, we point out here two types of contexts from colloquial Bulgarian where the two constructions diverge. I 2. Prepositionless datives ( na-drop )

The first asymmetry between CLLD and CLRD in colloquial Bulgarian invo Ives the so-called na-drop phenomenon (Vakareliyska 1 994), which consists in omitting the preposition na 'to' in front of indirect object DPs • resumed by a clitic in the same sentence ? For most speakers na-drop is possible if the object appears to the left of the verb, typically in sentence­ initial position, which can be identified with one of the available CLLD . positions in this language. According to the results of the test carried out and discussed in Vakareliyska ( 1 994), speakers reject the possibility of na-

I

Kayne's ( 1 995) and Samek-Lodovici's (2005) analyses of CLRD involve leftward movement of the dislocated item (as in CLLD) followed by leftward raising of the remnant IP. Cecchetto's ( 1 999) putative asymmetries between the two constructions, which he takes to argue against Kayne's analysis, do not seem to us to be cogent. Also see Samek-Lodovici's (2005) criticism. If the two . , differences that we discuss here could be attributed to the extra step in the derivation of CLRD and to the different pragmatic conditions associated with it, miliW then the derivational relation between the two constructions could still be : ';iili�' maintained With psy hological and physical perception predicates the omission of na is not ;�;; sensitive to person distinctions, while with all other predicates taking indirect . ' objects it is restricted to first and second persons. This, and other complexities of : ¥{�, the phenomenon, discussed in Vakareliyska ( 1994), are however orthogonal to our . " '. point. . .__.'

:!� .2 �\

�

360 /liyana Krapova and Guglielmo Cinque

�

drop if the indirect ob ect appears postverbally, i.e. in CLRD or clitic doubling (CD) contexts. See ( 1 )-(3):

(1)

(Na) mene sa mi kazali, ce djado Assen e (to) medat areJpl mednl.CL said3pl that grandpa Assen is da moze da gi gleda (CLLD) rabotil na tri mesta, worked in three places to be-able to themacc .cL take-care3sg 'They told me that grandpa Assen worked in three places in order to be able to take care of them. '

(2)

Ne moga dori da mu pomogna *(na) coveka. . . (CLRD ) not be-able)sg even to himdat .cL helplsg (to) man-DEF ' I can't even help the man.'

(3)

Tozi film mi xaresva sarno *(na) mene. (to) me dat this film medat .CL appea hsg only 'Only I like this film. '

(CD)

3 . Indefinite specific DPs

The second asymmetry concerns the possibility of clitic left dislocating, and the impossibility of clitic right dislocating (pace Ivancev 1 978, 1 64), indefinite specific DPs (i.e., indefinites which presuppose the existence of a 4 referent for the DP):

) "The highest over-all tol erance was for sentence-initial na-drop (e.g. [ .. ] mene ne mi xaresva tozi film ' I don't like that ,film'), while, ,as ,e{{pected from the results of the earlier test, sentence-fmal na-drop was found unaccep�ab le by most speakers (e.g. [. .] TozijUm ne mi xaresva _ mene).n (Vakareliyska 1 994, 1 37). 4 To judge from Philippaki-Warburton et at. (2004,982), the same left-right asymmetry with indefinite specific DPs is also found in Greek. See their example (29), reproduced here in (i), as well as their discussion following the example: _

(i)

a.

b.

7·

mu pighenondas sto periptero my going to-the kiosk mu ton sinandisa pigenondas sto periptero going to-the kiosk a fellow studenlatc my himacc met3sg 'I met a fellow student of mine on the way to the kiosk.' ton

sinandisa ena simfititi a fellow studentact

met) sg ena simfitit

himatt

i

Asymmetries between CUtic Left and Clitic Right Dislocation 3 6 1

Edin moj prijatel go vidjaxa da izliza ot xotela. one my friend himacc.cL saw3pl to walk-out3sg from hote1def 'They saw a friend of mine leaving the hotel. ' (Amaudova 2003 , 1 68) <edin moj prijatel> da izliza b. *Vidjaxa go to walk-out3sg saw himacc.cL one my friend ot xotela <edin moj prijatel>. from hoteldef one my friend 'They saw a friend of mine leaving the hotel.' 'Indefinite specific DPs in Bulgarian differ from both non-specific DPs, as ' well as from definite specific DPs, which do not show any such asymmetry: ' the former can never be dislocated, as illustrated in (5), while the latter freely enter both the CLLD and the CLRD constructions without any ;limitations « 6» (cf. Ivancev 1 978, Guentcheva 1 994, Assenova 2002, Arnaudova 2002, 2003): nameri toj, rano iii kiisno (CLLD) one woman will himself heracc ,cL find3sg he, sooner or later 'He will find one woman or other for himself, sooner or later' b. *Rano iii kiisno ste si ja nameri toj (CLRD) sooner or later will himself heracc.cL find3sg he edna zena. 5 one woman 'He will find for himself one woman or other, sooner or later'

a. *Edna zena

ste

si

ja

v

Zenata taka i ne ja nameri toj. and not heracc.CL found3sg he womandef so 'The woman, he did not fmd after all' h. Taka i ne ja nameri toj zenata and not heracc.cL found3sg he womandef so ' He did not find the woman after all'

(CLLD) (CLRD)

� Bulgarian CLLD in this respect differs from Italian CLLD, where even indefinite non-specific DPs can be clitic left (and clitic right) dislocated, as indicated by the " . ;' �"tl (grammaticality ofthe equivalent of (5):

iw

a.

h.

Una donna, prima 0 poi la troveri. Prima 0 poi la troveri., una donna.

362 Iliyana Krapova and Guglielmo Cinque 4. Are the asymmetries real?

In addition to CLLD, the Bulgarian left periphery is known to host more structural positions than the right periphery, each associated with a different construction: Focus movement, Hanging Topic Left Dislocation (HTLD), and the kolkoto do 'as for' construction (Amaudova 2002, 2003 , Krapova 2002, Krapova and Karastaneva 2000,2002, Krapova and Cinque 2005). So, if it could be shown that the options on the left side which are unavailable on the right side (i .e. ( 1 ) and (4)a) are not instances of CLLD but rather of one of the other left peripheral constructions, the syntactic identity between CLLD and CLRD could still be maintained. This however cannot be right, for several reasons. First, the fact that na-Iess datives and indefinite specific DPs cannot bear contrastive focus rules out the focus movement analysis «7a-b» . Secondly, the fact that the same left-right asymmetries appear in an embedded context, renders a HTLD account not viable « Sa-b» in view of the root-only character of this construction. Thirdly, the fact that na-less datives and indefinite specific DPs count as first position occupants w.r.t. Tobler-Mussafia effects (see ( 1 ) and (4)a above), rules out the possibility that they be (reduced) kolkoto do 'as for' phrases, given that, as noted in Krapova and Cinque (2005), the latter never 6 count as first position occupants. See the contrast between (9a) and (9b):

(7) a. *MEN sa mi kazali, ce djado Assen e rabotil me focus are3pl medat.CL said3pl that grandpa Assen is worked na tri mesta. in three places 'To me they said that grandpa Assen worked in three places.' vidjaxa da izliza b. *EDIN MOJ PRIJATEL go one my friendfocus himacc.CL saw3pl to walk-out3sg ot xotela. from hoteldef 'They saw a friend of mine leaving the hotel.'

6

By Tobler-Mussafia effects we mean the fact that Bulgarian clitics cannot be in first position of the clause but must be preceded by either the verb (in which case they are enclitic to it; see (9b)) or by some other constituent (in which case they are proclitic to the verb; see ( 1 ) and (4a).

Asymmetries between Clitic Left and Clitic Right Dislocation 363

(8) a. Kazax ti, ce men ne mi e studeno. said lsg YOUdat.CL that medat not medat.CL is cold 'I told you I am not cold' b. Razbrax, ce i edin tvoj prijatel sa go understoodl sg that also one your friend are himacc.CL pokanili na srestata. invited to meetin �ef ' I understood that they have invited also a friend of yours to the meeting.' pokanixa na srestata a. *Kolkoto do mene# me as for me meacc.CL invited3pl to meetingdef oste vcera. already yesterday na srestata b. Kolkoto do mene# pokanixa me as for me, invited3 pl meacc.CL to meetin�ef oste vcera. already yesterday 'As for me, they invited me to the meeting yesterday already' ' Having thus excluded all possible alternatives, we are left with the conclusion that the na-drop case « I )) and the dislocation of the indefinite SD4f:!czllC DP «4)a) are true instances of CLLD, unavailable in CLRD.

, Arnaudova� Olga. 2002. Clitic Left Dislocation and Argument Structure in Bulgarian, in Proceedings of FASL 1 0, University of Michigan, 23-46 ;f@,,:. ,.Jf\rnlaU(1ova� Olga. 2003 . Focus and Bulgarian Clause Structure. Ph.D. Dissertation, University of Ottawa Assenova, Petya. 2002. Balkansko ezikoznanie. Veliko Tumovo: Faber Cecchetto, Carlo. 1 999. A comparative analysis of left and right dislocation in Romance. Studia Linguistica 53.40-67 Guentcheva, Zlatka. 1994. Thematisation de I 'objet en bulgare. Bern: Peter Lang Ivancev , S vetomir. 1 978. Nabljudenija varxu upotrebata na elena v balgarskija ezik in: Prinosi v balgarskoto i slavjansko ezikoznanie, Sofia: Nauka i Izkustvo, 128- 1 52. , : Kayne, Richard 1 995. Class lectures at Harvard. Krapova. Iliyana. 2002. On the Left Periphery of the Bulgarian sentence University o/ Venice Working Papers in Linguistics 1 2. 1 07-128. Krapova, Iliyana and Guglielmo Cinque. 2005. Clitic Reduplication Constructions in Bulgarian, ms . , Univers itiJ. di Venezia

364 fliyana Krapova and Guglielmo Cinque Krapova, Iliyana and Tsena Karastaneva. 2000. Complementizer positions in Bulgarian, in Bulgarian Studies at the Dawn of the 2r' Century: A Bulgarian­ A merican Perspective. Sofia: Gutenberg, 93- 1 04 Krapova, Iliyana and Tsena Karastaneva. 2002. On the Structure of the CP field in Bulgarian. Balkanistjca 1 5.293-32 1 Ph iIippaki-Warburton, I., S. Varlokosta, M. Georgiafenius, G. Kotzoglou. 2004. Moving from theta-positions: pronominal clitic doubling in Greek. Lingua

1 1 4.963-989.

Samek-Lodovici, Vieri. 2005. When Right Dislocation Meets the Left-Periphery. A Unified Analysis of Italian Non-final Focus, ms., UCL, London. Vakareliyska, Cynthia. 1 994. Na-drop in Bulgarian. Journal of Slavic Linguistics 2

( 1 ). 1 21- 1 50.

On dative subjects in Russian

1. Two hypotheses on the F-coDstructioD

Moore and Perlmutter (2000) discuss a construction in Russian illustrated by the following example: ) Gruzovikam ne proexat' trucks NEG get-throughlNF ' it is not in the cards for the trucks to get through '

[1- 1 ]

I will call the construction of this type the F-construction (F-C; F for . 'fate'). M&P take the F-C to be a main clause whose verb is in the infinitive form and whose subject is in the dative case; in the sense M&P (2000:3 77) define it, it is a dative-subject construction. They claim that "Russian has a true dative-subject construction in which surface subjects are in the dative case." (M&P 2000:373). In contrast, 1 would like to claim that the F-C is an I-construction (I-C) in M&P's terms. M&P (2000: 377) define an I-C as a construction with "a surface indirect object [which] behaves like a subject in a restricted number of ways"; the term I-nominal used to "refer to the dative-marked nominal in it". M&P's position and mine are summarized in the following hypotheses:

.F:D HYPOTHESIS (F:D H): The F-C is a dative-subject construction with the imain verb in the infinitive form .

. .�. I would like to express my gratitude to Maria Polinsky for all the help she offered : l11e while I wrote this note. An earlier version of this squib was written while I was . �, visiting researcher at ATR, Kyoto, Japan, in 2000. I would like to express my gratitude to Yasuhiro Katagiri for providing me with this opportunity. Examples .. from M&P ( 1999, 2000) are identified by the numbers in brackets; ��::!�§(�I:t()mc:LD numerals I and II refer to M&P (1999), and M&P (2000), respectively. The '}��¥,,;j�nmati(:a}ijty judgments of the examples cited from M&P are those given by ·. ·: .:.:,,,,,:cu . •ltilli. l owe grammaticality/acceptability judgments of the other examples to Maria

"il�l'olinSkY.

366 s. - Y. Kuroda F : I HYPOTHESIS (F:I H): The F-C is an impersonal I-C whose main verb is a copula (silent in the present tense) with a dummy subject. 2. Temporal particles in Russian

A peculiar but uncontroversial fact in Russian pointed out by M&P ( 1 999) is that the phonetic forms that function as copulas in the past, habitual past and future tense in third person neuter singular can also function as temporal particles indicating the past, habitual past and future tense reference. That is to say, the word form bylo, byvalo and budet are ambiguous between copulas and temporal particles. This fact is demonstrated by M&P with the fol lowing kind of sentences where those words functioning as temporal particles co-occur with copulas: (2)

Gorod bylo byl okruzen vragom citynom.masc TP wasmasc surroundedMAsc enemYinstr 'The city was about to be surrounded by the enemy'

[1-34a]

Let us call temporal particles that are homophonous with copula forms pseudo-copula temporal particles (henceforth, P-C TP) . 3. Apparent occurrences of temporal particles in F-Cs

Now, consider the following exampl�s: (3 )

uzovikam bylo ne proexat' trucksdat was NEG get-throughinf 'the trucks couldn't get through'

(4)

Emu bylo ne stat' izbrannym kollegam i [I-5b] himdat was NEG becomeinf electedinstr colleaguesins,r ' it was not in the cards for him to be selected by his colleagues'

[I-Sa]

M&P take these sentences as F-Cs with the P-C TP bylo. Were it not for the existence of P-C TP i n Russian, M&P would have to take bylo as a copula in the past tense functioning as a main predicate in these sentences. F -Cs in the present tense such as ( 1 ) above could, and indeed would have to, be taken as having a silent copula as the main verb, for the sake of parallelism with F-Cs in the past tense; one would have to conclude that the infinitive in the F-C is not a main verb, and whatever the right analysis one might propose for the F-C, the F:D H would be fatally undermined. Thanks to the

On dative subjects in Russian 367

existence of P-C TP, M&P were not led to this impasse in spite of sentences like (3 ) and (4). :

4. True occurrences of temporal particles in F-Cs

However, in the presence of apparent counterevidence for the F:D H like (3 ) -(4), the existence of P-C TP is only a necessary but not sufficient . condition for this hypothesis to hold. The occurrence of bylo in (3 ), for example, could indeed be a P-C TP, but the possibility of it being a copula in the past tense has yet to be ruled out. M&P dramatically demonstrate the authenticity of p-c TP by showing ,their cooccurrence with copulas functioning as main predicates in forms as . in (2). The possibility of bylo in (3) being a copula in the past tense would ,be ruled out if it is shown that bylo in (3 ) may not coexist with a P-C TP. . F:D H would predict that the following forms, constructed in parallel to are ungrammatical: Gruzovikam byvalo bylo ne proexat' trucksdat TP was NEG get-throughinf 'The trucks used to be unable to get through' proexat' Gruzovikam byto bylo De trucksdat TP was NEG get-throughinf 'the trucks were about to be unable to get through' these forms are acceptable! 2 These sentences demonstrate that bylo in F-C like (3) is a copula as decisively as (2) demonstrates that bylo in (2) not a copula. Negative particle

ne

in F-Cs

a second argument in support of their claim that bylo in (3 ) is a P-C TP, not a main verb, M&P refer to the positioning of the negative particle In Russian sentence negation, the negative particle ne precedes a tensed iredicat:e. But ne follows bylo in (3). This ordering is taken by M&P as �vidlell(�e for by/o in (3 ) n ot being a past tense copula, but a P-C TP. M&P state that "in fact, temporal particles cannot host negation":

." A small pause is needed between the two bylo and the frrst one needs to carry nigh pitch in (6).

368 S.- Y. Kuroda

(7)

*Gruzovikam ne by10 proexat' trucksdat NEG TP get-throughinf ' it was not (in the cards) for the trucks to get through'

[1-28a]

I would l ike to maintain that this argument for the F:D H does not hold either. 5 1 Evidence for a negated copula . .

in F-Cs

The behavior of the negative particle in the F-C is very complicated. I am not in a position to provide a satisfactory full description and account of it. Here I must restrict myself to two remarks that are sufficient to cast doubt on the validity of M&P's argument for the F:D H based on the negation facts. To begin with, the syntax of Russian negation is in fact compatible with the assumption that bylo is a tensed copula and proexat' is an embedded infinitive in (3). The reason for this is that in Russian the negative particle precedes the infinitive in the embedded negative clause. If bylo in (3) is a copula, as I claim it is, (3) is an affinnative sentence (the main verb is not negated) with a negative infinitive embedded in it; a structurally more faithful translation of (3) in pseudo-English would be: (3 ' )

for the trucks there was no getting through

Secondly, if we assume, as I do, that bylo in (3) is a copula, it is evident that the negative of (3) is as given below: ( 8)

Gruzovikam ne bylo ne proexat'

And indeed (8) is grammatical and means what one would expect it to: (8')

it is not true that it was (meant) for the trucks not to get through

that is (8")

it is not true that it was (meant) for the trucks to get stuck. 3

Thus, it is clear that bylo in (3) can be negated and is a copula.

3

l owe

the grammaticality judgment and the translation to Maria Polinsky.

On dative subjects in Russian 369 5.2.

Negation and the pragmatics ofthe F-C

M&P mark (7) as ungrammatical. But perhaps the fact of the matter is that (7) is not readily acceptable, rather than ungrammatical. The F-C NP copula [ne VINF.] pragmatically entails the expectation that NP V holds. Thus, (3) accompanies the expectation that the trucks generally get through, and asserts that (the fate is that) contrary to this expectation the trucks couldn't get through. (8) accompanies the same expectation as (3) and denies what (3) asserts: the trucks indeed could get through as one would expect. In contrast, then, (7) would deny (9) below (9) Gruzovikam bylo proexat'

and would, as (9) should, accompany the expectation that trucks generally do not get through, or get stuck. This would not be the unmarked expectation one entertains with trucks. Thus, in order to accept (7) as well as (9), one is required to contrive special contexts to entertain such an uncommon expectation, and that would perhaps contribute to the unacceptability judgment for (7) when given without context.4 Be that as it may, given the grammaticality of (8), the quest for the . cause of a star given to (7) by M&P is deemed moot insofar as OUf present 'concern of detennining if bylo in (3) is a copula or a P-C TP . . 6. The F-C is not a dative-subject construction

From the considerations in the previous two sections, I conclude that the F­ C is a sentence with a copula as its main predicate and with a verb in the infmitive as the predicate of an embedded clause. The subject of this infinitive verb would then be an empty category controlled by the sentence­ Initial dative nominal. The F-C thus appears to have the following structure: (1 0)

NPi-dat Copula [PROj Verbinf . . ..]

Note that the main verb of this construction, a copula, is empty in the present tense, as copulas in Russian are generally silent in the present tense. 4 Thus, (9) is unacceptable unless there is high pitch on byla. l owe this account of the nature of unacceptability of (7) to Maria Polinsky. Compare It was the fate of the truck to get through and It was the fate of the trucks not to get through. The latter doesn't sound as odd as the former.

370 S.- Y. Kuroda The copula in ( 1 0), nonetheless can be assumed to he the third person singular, since it takes the third person singular form in the future and the habitual past and the singular neuter form in the past tense. The sentence-initial nominal can be any person, gender and number . This means that the sentence-initial dative nominal fails to agree with the predicate, and hence it cannot be a dative subject in the sense M&P wish to understand this term . What, then, is the subject of this verb? It must be an expletive empty subject; the F-C is an impersonal construction. And indeed the F-C satisfies a diagnostic test for the impersonal construction discovered by Perlmutter and Moore ( 1 999). For the interest of space, I have to leave aside details of this claim. I only mention that the following contrast provides a relevant data for the diagnostic test: (1 1) ( 1 2)

*ctoby gruzovikam in order trucksdat

byt' ne b�nf neg

proexat' go-throughinf

ctoby gruzovikam bylo ne proexat' besubj neg go-throughinf in order trucksdat 'in order for the trucks to be unable to go through'

I inv ite the reader to construct a proper argument for the test by consulting P&M ( 1 999). 7. The main concern of Moore and Perlmutter

So far I have been concerned exclusively with the analysis of the F-C . I would now like to tum our attention to the concept of dative subject, which is the main concern of M&P (2000). M&P claim that they "show[ed] that Russian has a true dative-subject . construction in which surface subjects are in the dative case';' . I-nominals' failure to behave like subjects in some respects "canno� - be attributed to their dative case because true dative subjects� als o 'itt · the -;: dative case, behave like subjects in every respect. We conclude that dative subjects and I-nominals instantiate distinct constructions which must be recognized as such in syntactic typology and syntactic theory". (M&P 2000:373) It is well-known that the nominative subject of a finite clause in Russian is converted to the dative case when the clause appears as an infinitival subordinate clause, as i llustrated by the following examples:

On dative subjects in Russian 3 7 1

( 1 3)

cto by nam uexat' na vokzal ... in-order USdat go- outinf to rail way-stat ion ... ' in order for us to go (out) to the railway station . .. '

( 1 4)

ctoby im nacat' rabotat' odnim in-order theydat begininf workillf alonedat 'in order for them to begin to work alone'

[11-26]

[II-44b]

According to M&P's definition of subject-predicate agreement, the dative nominals in these infi nitival clauses may be said to agree with their infinitival predicates. 5 It is not difficult to see that such dative nominals can be said to raise and be controlled. Thus, they qualify as dative-subjects as M&P define this term . Thus, the existence of desired dative subjects in Russian can be shown even without mentioning the F-C. In that case, whether the F:D H holds (as M&P claim) or not (as I do), it must be taken as i rre levant to the main concern of M&P (2000), the existence of dative subjects, as they define the term. In fact, not only do we not need the F-C to make M&P's point, but the F-C itself could not serve as evidence for M&P's objective, even if M&P's analysi s were correct. For, according to M&P, the F-C is an infinitival clause used as an independent clause; the infinitival clause used as an independent clause takes on a particular functional load of expressing a modal type of meaning attributed to the F_C.6 Under those circumstances, the dative nominal of the F-C cannot raise: by definition, the F-C cannot be embedded and hence its subject can never be in the context for raising. For the same reason, the dative nominal of the F-C cannot be controlled . To be fair, I also cannot say that the dative nominal of the F-C FAILS to raise or to be controlled. The matter of the fact is that proper environments for raising and control fail to occur. Thus, the dative nominal of the F-C could be said to qualify as a dative subject by default. However, it could not serve as positive evidence for the existence of dative nominals that behave like subjects in a l l respects. F-Cs figure prominently in the course of the presentation of M&P's arguments as examples of dative-subject constructions. But interesti ng

,5

M&P formulate subject/predicate agreement in tenns of unification: "A predicate must unify with its subject in person, number, and gender". (M&P 2000:396) Since . infinitives have no agreement features, they agree with any nominals vacuously. 6 l owe this clarification of M&P's position to John Moore.

372 S. -Y. Kuroda

though they are for their own sake, F-Cs are superfluous for the stated objective of M&P (2000), even if M&P's analysis of the F-C were correct. 8. Summary and conclusion: two conceptions of dative subjects

Dative subjects in Russian infinitival clauses can be accounted for by their correspondence to nominative subjects in corresponding finite clauses, whatever mechanism one might devise to account for this correspondence. To see the significance of this point, observe that nominative subjects in finite clauses are converted to for-phrases in the corresponding infinitival clauses in English. The correspondence between nominative subjects and the dative subjects in infinitival clauses in Russian is comparable to this correspondence between nominative subjects and for-phrase subjects in infinitival clauses in English. Naturally, these non-nominative subjects in infinitival clauses in Russian and English are expected to behave like subjects due to these correspondences with nominative subjects in finite clauses. They are 'transforms' of nominative subjects of a canonical sentence type. We have here an issue belonging to the morphology of transformed subjects. Likewise, the genitive case-marking of gerundive clauses in English belongs to the same problem area. In contrast, the problem of I-nominals, in M&P's sense, concerns nominals masquerading as subjects; this is an issue orthogonal to the morphology of transformed subjects. We are in a problem area where such matters as canonical vs . non-canonical sentence types are at issue. Sentences with I-nominals can deviate from the canonical sentence type insomuch as they do not display nominative subjects of the canonical sentence type in a form visible at the surface level; yet I-nominals behave l ike subjects, though only in a restricted number·of ways. It is conceivable that we have another noncanonical sentence type where a dative nominal IS a surface subject, that is, a nominal that does, unlike I-nominals, pass all diagnostic tests for surface subjects. Such ·a subject also determines a noncanonical type of main clause. In order to distinguish such dative subjects from dative transformed subjects, we can use the familiar term dative quirky subject. The existence of transformed dative subjects is not a matter worthy of particular attention for the typology of canonical and noncanonical sentence types. That they take the dative case is a problem of the morphology of transformed subjects. In contrast, whether quirky dative subjects exist or not is an issue that relates to the typology of canonical and noncanonical sentence types.

On dative subjects in Russian 373

M&P "conclude that dative subjects and I-nominals instantiate distinct constructions which must be recognized as such in syntactic typology and syntactic theory". (M&P 2000:373) This statement is certainly true. But what M&P have argued in connection with the F-C does not get to the heart of the matter and does not jibe well with this statement. To begin with, they argue that the dative nominal of the F-C instantiates a dative subject, but in fact, as I have shown, it instantiates an I-nominal . And they do not make the important distinction between dative quirky subjects and dative · · transformed subjects. Russian provides evidence for the dative transformed subjects: Russian infinitives take dative nominals as subjects. The above ',conclusion' by M&P would be almost a truism if it is meant to explain dative transformed subjects. But the dative nominal in the F-C does not • instantiate a quirky subject, or, for that matter, any subject; rather, what it �n fact instantiates is an I-nominal, an ironic twist in the present context with M&P's conclusion above. Dative quirky subjects may still exist in Russian, but as the F-C does not provide evidence for it, we for now need to look for evidence in other :ilangl11��es such as Icelandic to ascertain the viability of the concept of .... I1' ...Ii'" dative subject. 7 "

",,',�l'J.V" A �,

John and David M. Perlmutter. 1 999. Case, agreement and temporal particles in Russian infmitival clauses. Journal a/Slavic Linguistics 17.2, 2 1 9246. 'M4[)Or,e. John and David M. Perlmutter. 2000. What does it take to be a dative subject? Natural Language and L inguistic Theory 1 8: 373-4 1 6. iZ:'; F'erllmu1tter, David and John Moore. 1 999. Syntactic universals and /anguage­ particular morphology: Russian impersona/s. Handout, the 73rd Annual Meeting of LSA.

7 · For the literature on other languages that are thought to have quirky subject constructions, see M&P (2000:407).

O n the nature of case in Basque: structural or inherent? Itziar Laka

1 . Introduction

1

In this squib I question the hypothesis that case is structural in Basque (Ortiz de Urbina 1 989 , Fernandez 1998 Laka 1 993 among others). I claim that case morphology is best explained as inherent (Chomsky 1 98 1 , 1 995); as a direct reflection of thematic role. This view, originally suggested in Levin ( 1 983), has as a consequence that this grammar is not ergative in the sense of Bobalj ik (1 992), Bittner and Hale ( 1 996), Fernandez ( 1 998), or Laka ( 1 993, 2000). 2. Case Theory and Structural Case

The main empirical motivation for keeping Case Theory separate from Theta Theory in the Principles-and-Parameters model (Chomsky 1 98 1 , 1 986) i s the fact that case and theta role are not equivalent phenomena: an argument bearing a theme theta role can surface as nominative or as accusative ( l �b), and conversely, nominative case can be assigned to either an agent or a theme ( l b,c). ( 1 ) a. She sank. b. A storm sank her. c. A storm arose. In the Principles-and-Parameters (P&P) model : of grammar, nominative and accusative case are assigned · to NPs that,oc,cupy, lsp�dfic, positions in the syntactic structure at the level of S-structure:' nominative is assigned by Infl to NPs in the specifier of IP, and accusative is assigned by V to its complement/sister NP (Chomsky 1 98 1 , 1 986). This is why nominative and accusative are called structural cases: they are assigned to whichever NP sits in the relevant structural position, regardless of the theta role it is assigned at D-structure, and regardless of what category assigns that theta I

Research funded by the ESF, Eurocores programmes, OMLL (MCYT: BFF2002-

1 0379-E).

On the nature ofcase in Basque 3 75

role. P&P also differentiates the notions of structural and inherent case: stru ctural case is dissociated from theta role assignment, but inherent case is indeed equivalent to it, and can only be assigned by a theta role assigner . to its assignee. Inherent case then correlates with thematic assignment, and . therefore, there can be no dissociation between theta role and inherent case. ,this distinction between Case Theory and Theta Theory on the one hand, . and between structural and inherent case on the other, is maintained in the Minimalist Program (Chomsky 1 995: 1 1 4). What varies with respect to are the structural positions where cases are checked: nominative and " a(�cusat:'lve case checking involve Spec-head relations with Agrs and Agro respectively (Chomsky 1 995: 1 73). 3. Nominative and ergative grammars

Given this account of structural case, and following Bobalj ik ( 1 992), !:�;::C:hOlns�� ( 1 995 : 1 76) assumes that the different case patterns found in inative versus ergative grammars arise from the choice of case�he:ck]m ' g position in intransitive sentences: if Agrs is kept active and Agro the nominative/accusative case pattern arises, and if Agro is kept and Agrs inert, then the ergative/absolutive case pattern arises. In this of nominative versus ergative case patterns, the case names rzon'lln4'ltll'e and ergative tum out to be descriptive labels corresponding to structural case checked in Spec-Agrs, and the case names accusative absolutive correspond to the structural case checked in Spec-Agro. Both .param.etr:'lc choices involve structural case, and therefore it is predicted that USSioclatlons of the sort illustrated in ( 1 ) are to be found also in ergative ars. This is indeed the case, as we see in the following examples Burushaski (2) and West Greenlandic (3) (from Manning 1 996): a. Ne hir-e phal6 b6k-i. seed man themasc erg pl abs SOW3sg.masc.S 'The man planted the seeds,' llir yalt-i. b. Ne themasc manabs yawn.Pret.3sg.masc.s 'The man yawned. ' -

Oli-p neqi neri-vaa. Olierg meatabs eatind.tr.3Sg.3 sg '0 Ii eats meat. '

376 Itziar Laka h.

sinipoq. Oli Oli abs sleepind.intr.3Sg 'Oli sleeps.'

Examples (2a,b) illustrate the case of an agentive NP ne hir 'the man', bearing ergative (2a) and absolutive (2b), and (3a, b) illustrate the same dissociation for the agentive NP Olio If this account of ergativity is correct in its general outline, as I assume, both nominative and ergative grammars are minimal parametric variations within the general mechanism of structural case assignment. 4. Case morp hology in Basque: unaccusatives I f we consider the grammatical equivalents of ( 1 ) in Basque, we find no direct evidence that case and theta role are dissociated (4);

(4) a. Txalupa hondora-tu da. is boaldetabs sin�rf ' The boat sank.' h. Ekaitz-a-k txalupa hondora-tu duo has stormdet.erg boaidet.abs sinkperf 'The storm sank the boat. ' c. Ekaitz-a sor-tu da. stormdet.abs ariseperf is 'A storm arose' In these examples, there is no manifest dissociation between case and theta role: the NP txalupa 'the boat', which is a theme in both sentences, bears absolutive case regardless of whether the sentence' is intratlsitive, as in (4a), or transitive as in (4b). On the other hand, the . WP ekaitza 'the storm' changes case as it changes theta · role: in (4b) ,if is �n', agent,. and carries ergative case, but in (4c) it is a theme, and it carries :absolu�ive case. Given the general account of nominative ' versus ergative' grammars outlined in section 3, the data in (4) are easily accounted for: in ergative grammars absolutive case is the only available case in intransitive sentences, and thus the account predicts the assignment of absolutive in unaccusative predicates like (4a) and (4c), as it predicts the assignment of ergative case to the agent of a transitive predicate like (4b). So far, then, the data in (4) are uninformative as to the nature of case in Basque: the pattern that arises is

On the nature ofcase in Basque 377

the one predicted by most accounts of ergativity (Bobaljik 1992, Bittner and Hale 1996 Fernandez 1 998, Laka 2000, Nash 1995),

" ,5. Transitive unergatives Unergative predicates constitute a very relevant piece of data in order to " find instances of case/theta role dissociation in ergative languages: if the subject carries absolutive case, then we have a clear instance of case/theta role dissociation, and we can conclude that case is structural (2h), (3b). " ,Consider the equivalent sentences in Basque: a. Gizon-a-k aharrausi egi-n duo mandet.erg yawn dOperf has 'The man yawned. ' h . Oli-k 10 egi-ten du. o lierg sleep dOimpf has ' 0 Ii sleeps.' (5a,b) are intransitive, then the ergative case on the subject NPs is not Uribe-Etxebarria ( 1 989), Bobaljik (1993), Hale and Keyser ) and Laka ( 1 993) have argued that (Sa, b) are in fact transitive i:SE:nt(mc:es:' in (Sa) there are two argument NPs, one is gizona 'the man' and " other one aharrausi 'yawn' , The verb egin 'to do' assigns a theme theta .. 'the to aha"ausi 'yawn' and an agent theta role to gizona 'the man'.
unergatives are not syntactically equivalent to (2b) or (3 b), :U;W'helre an agentive NP carries absolutive case, As first discussed in Levin ,, ), the ari-progressive construction looks like a candidate for such a case/theta role dissociation:

378

Ilziar Laka

(6) a. Gizon-a aharrausi egiten ari da. making PROG is Mandet.abs yawn 'The man is yawning.' b. Oli 10 egiten ari da. Oliabs sleep making PROG is 'Oli is sleeping.' Compare (6a, b) with (5a,b): no ergative case is assigned in the progressive sentences. The ari-progressive has been argued to be an antipassive by Postal ( 1 977)� if this were the case, it would constitute direct evidence that case is structural in Basque because antipassives in ergative languages are in this respect the mirror image of passives in accusative languages: a passive results in the assignment of nominative to a theme, and an antipassive results in the assignment of absolutive to an agent. However, as argued by Hualde and Ortiz de Urbina ( 1 987) and Laka (2005), this construction is not an antipassive: ari is an unaccusative verb, meaning 'to be engaged (in)'. This unaccusative verb assigns a theme role to its NP argument and it takes a locative-like predicate that can (but need noti be an infinitival clause. In short, the syntactic structure of sentences like (6a,b) is biclausal, and it contains two verbs, as shown schematically in (7) (recall that Basque is head-final):

Once we consider the syntactic structure of the ari-progressive, we can see that the apparent dissociation between 9ase and theta role vanishes. There are two verbs in this structure: (a) the unaccusative verb ari (Vari) in the main clause assigns a theme theta role to its subject (NPj); this verb also takes a PP ([pp]) headed by the locative postposition -no The complement of the locative postposition is a nominalized clause ([NP]) headed by the nominalizer -ie, which takes the infinitival sentence headed by the embedded verb (V). (b) the embedded verb (V) heads the nominalized infinitival clause, and it assigns theta roles to its arguments within its own clause. The external argument of this infinitival clause (PROj) is controlled by the subject of the main clause (NPj). Crucially, the matrix subject that surfaces with absolutive case (NPj) is not an argument of the embedded verb, but of ari 'to be engaged in ' . This progressive construction, then, is 2

For details on this and other progressive constructions, see Hualde and Ortiz de

Urbina ( 1 987) and Laka (2005).

On the nature of case in Basque

3 79

. . ;'essentially the same as (4a,c): an instance of absolutive case borne by a ,theme NP. A grammatically minded translation of the sentences in (6) into ,En glish makes this point clearer: (6a) is akin to 'the man is engaged in ';;" vawm·l nll" and (6b) literally means COli is engaged in sleeping' . For the . ' urposes of this squib, the relevant conclusion to be drawn from the study the progressive is that gizona 'the man' in (6a) and Oli in (6b) are not ;;ex.amlpH�s of agent NPs marked with absolutive case. Rather, they are ; instaJnCf�S of themes bearing absolutive, which we find happens consistently, far, in the grammar of Basque. There is no dissociation between case and role in the progressive either. Intransitive unergatives

Let us now consider instances of unergative predicates that show no signs transitivity. The following data are well known problem cases for all ac:counlts of case in Basque that involve the notion of structural case (Ortiz Urbina 1 989, Bobaljik 1993, Laka 1 993, Fernandez 1 998 among others): Klara-k ondo eskia-tzen duo has Klar8erg well skiimpf 'Klara skies well.' b. Erloju-a-k ondo funtziona-tzen duo has Watchdet.erg well workimpf 'The watch works well.' sentences in (8) are instances of unergative predicates (Levin 1 983, 2003 ), but unlike the light-verb type unergatives we have �oi1,sid,ere:d in (5), here there is no sign of transitivy, that is, there is no object that could activate the assignment (or checking) of lbSlolultlv'e case, which would in tum permit the assignment (or checking) ergative case to the agentive NPs. Recall that in structural-case accounts . ergativity, assignment of absolutive case is required whenever ergative is assigned. In fact, as we saw in (2b) (3b), ergative grammars :;s�'ste:m(llti·i cally assign absolutive case to agentive arguments of intransitive : s.entences. That is the reason why (8a,b), the grammatical equivalents of ::{2b) and (3 b) are problematic cases: their subjects should display absolutive case, not ergative. The set of intransitive unergatives is relatively small in the lexicon of .l:Sasqtle (see Alberdi 2003 for a thorough estimation). Since Sarasola 1 977, ithas been claimed that: (a) all intransitive verbs borrowed from romance

3 80

Itziar Laka

in recent times are like (8) regardless of their conceptual structure; (b) all verbs like (8) are borrowings from romance, and (c) these verbs act like nominative/accusative exceptions in the grammar. Alberdi (2003) argues in detail that there is no empirical basis for such claims; his overview of recent verb-borrowings and their case frames shows that: (a) when an unaccusative verb is borrowed from romance, there is no ergative case on the subject as shown in (9a), and only those borrowed verbs that are unergative behave like (8). As for (b), not all intransitive unergatives are borrowings (9b): (9) a. Gizon-a errendi-tzen da. surrenderimpf is. mandet 'The man surrenders.' b. Eguzki-a-k disdira-tzen duo shine impf SUndet.erg has 'The sun shines.' c. *Eguzki-a-k itsaso-a disdira-tzen duo has shineimpf SUfidet.erg se3det Intended meaning: 'The sun shines the sea.' The unergative verb disdira(tu) '(to) shine' in (9b) is not a borrowing and yet its agentive argument displays ergative case, although the sentence is truly intransitive (9c). These predicates do not constitute a 'borrowed' exception in the grammar; they are unaccounted for within a structural case approach, but are straightforwardly explained if case is inherent. 8. An Inherent case approach to Basque

Most data on case assignment in Basque are not informative on the issue of whether case is structural or inherent in this language: both structural and inherent approaches can explain the case patterns tliat arise in this grammar (sections 4 and 5); there are no instances where case morphology and theta role are clearly dissociated, which would render the inherent-case hypothesis untenable. Linguists familiar with Basque are aware that truly intransitive unergatives require special provisos or must otherwise directly be acknowledged as flat-out counterexamples to the general case assignment mechanism. A solution to this long standing problem is to tum the tables around and rescue Levin' s ( 1 983) original proposal that case is a D-structure phenomenon in Basque (or, in updated tenns, that case is inherent, theta-related). A detailed analysis of the ari-progressive shows

On the nature of case in Basque 3 8 1

.

. that i t does not pose a problem for this hypothesis, an issue that Levin ( 1 983) wondered about. Once we conclude that progressives do not break the general pattern of case and thematic role correlation we find in this . grammar, we can look at truly i ntransitive unergatives as the crucial evidence that case is inherent. This approach can also account for the fact that case is freely assigned in infinitivals, and could explain why Basque stands as one of the few (if not the only) ergative grammars with no �tipassive, and no true case-split, both common features of ergative 3 . languages, not to be expected if case is indeed inherent.

Alberdi, Xabier. 2003 . The transitivity of borrowed verbs in Basque: an outline. In: Bernard Oyhar�abal (ed.) Inquiries into the Lexicon-Syntax Relations in Basque. Supplements of the Anuario de Fil% gia Vasca "Julio de Urquijo "

XL VI. 23-46. , Gontzal. 2002. Split ergativity in Basque: the Pre-Basque antipassive­ imperfective hypothesis. In: Folia Linguistica Historica XXI/I -2, 3 1 -97. ?B.ittnc�r" Maria and Ken Hale. 1 996. Ergativity: towards a theory of a heterogeneous class. In: Linguistic Inquiry 27-2. aljik, Jonathan D. 1992. Nominally Absolutive is Not Absolutely Nominative. Bob .. In: J. Mead (ed.), Proceedings of the Eleventh West Coast Conference on Formal Linguistics WCCFLl l, Stanford, California: CSLI Publications.

Distributed by the University of Chicago Press. �ik, Jonathan D. 1 993 . On ergativity and unergative predicates. In: Jonathan Bobaljik and Collin Philips (eds.), MITWorking Papers in Linguistics 19: Papers on Case andAgreement IL 45-88. Department of Linguistics and . Philosophy, MIT, Cambridge, MA. :Ch.omJsky, Noam. 1 98 1 . Lectures on Government and Binding. Dordrecht: Foris. ,i:pllornsk:y, Noam. 1 986. Knowledge of Language: Its Nature, Origins and Use, . . New York: Praeger. 't.:l1lomsky Noam. 1 995. The Minimalist Program. Cambridge: MIT Press. Beatriz. 1 998. Egiturazko Kasuaren Erkaketa Euskaraz, Bilbao: Servicio de Publicaciones de la UPV EHU Ken and Samuel Jay Keyser. 1 993. On Argument Structure and the lexcial expression of syntactic relations. In: Ken Hale and Samuel Jay Keyser (eds.) The Viewfrom Building 20. Cambridge: MIT Press. -

1;

.

The data and generalizations in ihis squib hold of western dialects but not of all �astern varieties. As suggested in AIdai (2002), some eastern varieties might well .. �(r'ergative' ; Oyhar�abal (1993) shows that eastern Basque has absolutive subjects in intransitive unergative predicates; space prevents us from discussing them here.

382 ltziar Laka Hualde, Jose Ignacio and Jon Ortiz de Urbina. 1987. Restructuring with ari. In:

Anuario del Seminario Julio Urquijo 2 1 : 425-452 . Hualde� Jose Ignacio and Jon Ortiz de Urbina. 2003 . A Grammar of Basque. BerlonlNew York: Mouton de Gruyter. Laka� Itziar. 1 993 . Unergatives that Assign Ergative, Unaccusatives that Assign Accusative. In: Jonathan Bobaljik and Collin Philips (eds.), MITWorking Papers in Linguistics 18: Papers on Case andAgreement /, 149- 172. Department of Linguistics and Philosophy, MIT, Cambridge, MA. Laka, Itziar. 2000. Thetablind Case: B urzio's Generalization and its image in the mirror. In: Eric Reuland (ed.) Arguments and Case. Explaining Burzio IS Generalization. 1 03 - 1 29 Amsterdam/Philadelphia:John Benjamins Publishing Company. Laka, Itziar. 2005 . Deriving Split ergativity in the Progressive: the case of Basque. In: Alana Johns, Diane Massam and Juvenal Ndayiragije (eds.) Ergativity: Emerging lssues. 1 73-95 DordrechtlBerlin: Springer. Levin, Beth. 1 983 . On the Nature of Ergativity. PhD Dissertation, MIT. Manning, Christopher. D. 1996. Ergativity. Argument Structure and Grammatical Relations. PhD Dissertation Stanford University. Nash, Lea. 1 995. Portee argumentale et marquage casuel dans les langues SOV et dans les langues ergatives: l' exemple du georgien., PhD dissertation, U niversite Paris VIII, Paris. Ortiz de Urbina, Jon. 1 9 89 . Parameters in the Grammar of Basque. Dordrecht:Foris. Oyhar�abal, Befiat. 1 993 . Inherent and Structural Case Marking: Ergaccusativity in Basque. In: Jose I. Hualde and Jon. Ortiz de Urbina (eds.) Generative Studies in Basque Linguistics. 209-42 Amsterdam : John Benjamins. Postal, Paul M. 1 977. Antipassive in French. In: Linguisticae Investigationes 1 ,

333-74. Sarasola, Ibon. 1 977. Sobre la biparticion inicial en el am'llisis en constituyentes. In: Anuario del Seminario de Filologia Vasca Julio Urquijo (ASIU), XI, 5 1 -

90.

Uribe-Etxebarria, Myriam. 1989. On noun incorporation in Basque and some of its consequences in the phrase structure. University of Connecticut, Storrs, ms.

Examining the scope of

Principles-and-Parameters Theory David LeBlanc

Having come to linguistics from an external field (computer science), my work has tended to revolve around how the tools of infonnation processing can be applied within linguistic theories and what affect these tools may ' have in how such theories are postulated. I This has been the traditional approach of cross-disciplinary researchers, whether they are speculating as to, e.g., how a limited-depth stack could account for center-embedding restrictions (Church 1 980), how unification could be used in feature checking (Pollard and Sag 1 994), how a distributed representation of information could account for the U-shaped learning curve (Rummelhart and McClelland 1 986, Davis 1 987), or how a statistically-influenced learning mechanism could account for gradual learning phenomena . lanc 1 995, Yang 2003). The influence of linguistic theory upon other ,' , " however, has been limited in recent years. This is somewhat surprising, particularly given the external influence of linguistically­ centered ideas in the past. Chomsky's early work on the classification of , (i.e. the Chomsky Hierarchy - Chomsky 1 956) had a major imJ;lact on fonnal language theory within computer science and the concept , deep structure (Chomsky 1 965) has motivated researchers in a variety of to look beyond surface phenomena for unifying hidden structures. It be argued that this lack of influence is the result of the complexity of m()Oelrn linguistic theory, making the field more opaque to researchers not nnrnerseo in its minutia and restricting its observations to a small core of ,:CleOic:ate:d followers. However, no matter how complicated, at their base all .PTlop()sals in the field are grounded in the core concepts of innatism and : gelnenitivity - concepts that may exist in other fields of endeavour, but are ",,," r:!lI rp, ll,, sourced back to linguistics. , . ,. ' , . As I have now left the world of linguistic theory to re-enter my original : , field of study, I am left to wonder what concepts and constructs I take with " }ne from linguistics that can be applied to my new topic of research,

j 'This work was

fundedt in partt by NSERC. I should also like to thank Henk van

Riemsdijk for his support and patience during my fonnative years. He should not, howevert be held responsible the outcome - he tried his best, but I am stubborn.

L

3 84

David LeBlanc

human-computer interface design. The transition from computer science to linguistics was much clearer - computer science gives to one a set of general modeling tools that can be applied to pretty much any system whose functioning can be described. The transition back to computer science from linguistics is much less clear, given that much of linguistics is theory-internal.

There

are,

however,

base

concepts

concerning

how

language information is acquired and represented that I hope to apply to my new research - specifically, the concept of innatism in general and the concepts of principles and parameters in particular. The idea that there is an innate basis for behaviour is not specific to generative theory - indeed, the entire nature/nurture debate is centered on the extent of innate information we are born with and its effect upon our development. But generative theory is far more descriptive than most in postulating an actual mechanism for integrating innate knowledge into the development process. The idea that we are born with more innate information than we need for our specific environment (i.e. the entire content of parameters) and then 'choose' only that which is relevant (i.e. set the parameter) is a highly specific claim. This idea provides the basis for a formal

description

of knowledge

of language

as parametric

values

corresponding to the mechanisms determining language behaviour for a given phenomenon. Generative grammar stipulates that there are a relatively small number of possible behaviours for a specific phenomenon (the parameter values), all of which are innately specified. The task of the language acquisition device is then simply to choose from this defined set the correct behaviour (i.e. parameter setting) that corresponds to the language of exposure, integrating this behaviour into the adult grammar to the exclusion of competing behaviours. This is. a fundamental idea that can take from linguistics and apply to other fields - if I

can

I

identify nOD­

linguistic behaviour that exhibits the same properties as language in this respect. The field of human-computer interface design may seem like an unlikely field to try and apply the concepts of innatism and parametric variation. Interfaces are often designed either in upon

the

individual

preferences

of the

an

ad-hoc way dependant

designer

him/herself or

in

accordance with a list of company-wide standards based on past experience of what has worked for a particular interaction environment. There is, however� more theoretically-driven work that addresses fundamental issues of interaction - specifically, how identifiable groups of people prefer to have their infonnation presented to them based upon what kind of cultural environment they were raised in . This work,

based

in the field of

The scope ofPrinciples and Parameters theory 385

communication theory, attempts to identify culturally-dependant variables (i .e. parameters) and to tie specific individual behaviour (with respect to infonnation exchange) to the corresponding cultural environment of the individual. Just as linguistic parameters define the possible behaviours for a linguistic phenomenon and then try to relate them to the language of exposure, communication parameters define possible behaviours governing the exchange of infonnation and try to relate them to the culture of exposure. The question is, are these postulated communication-based parameters valid and, if they are, are they innately specified the same way linguistic parameters are and are the proper behaviours for the culture of exposure 'selected' in the same way as linguistic parameter settings are? Communication theory is a relatively new field that has not attracted the same rigorous study that linguistic theory has been subjected to. It is . based heavily on studies of groups that share a common orientation (usually . a common workplace, job or field of study) hut that exist in different cultural environments. Participants are asked questions to ascertain their 'attitudes and preferences with respect to different phenomena related to how they interact with others. From this data researchers then attempt to . correlate these attitudes and preferences with the cultural environment of subjects. For examp le, in the most famous study of this kind, Geert ( 1 997) surveyed workers for IBM corporation who did similar ts;tecle 1FId[o . IBM is a very large corporation with branches throughout most of the so Hofstede was able to obtain a large sample of data from many �u. .......& ... .1&.. cultural groups. This group was particularly valuable, Hofstede arJlmes. because the workplace environment of the sample groups was very :OUJIUI'�1 (at the time IBM was famous for its own rigid company culture that ;aplplle�d to all branches), meaning that the results should be dependant upon the prevailing culture of the host country - the workplace t:'lI lhl1rp being remarkably similar across all groups. Hofstede then looked clusters of like attitudes between the groups and attempted to correlate to characteristics of the prevailing cultural environments. In doing so identified five parameters of cultural variation and the effects of these in subject attitudes. These parameters are: power distance, ;c<JtUecti'v ism vs. individualism, femininity vs. masculinity, uncertainty and long- vs. short-term orientation. Other researchers have to this list of possible parameters or have made competing proposals; instance, Hall ( 1 969) add.s speed of messages, sense of space, sense of . Jime (polychronic vs. monochronic) and context of information flow. Once loci of variation have been identified, researchers can then JUt1eml1>t to predict how a culturally-based attitude should affect one's

3 86

David LeBlanc

preferences as to how information is exchanged and use this to guide in the development of a computer system interface. For example, the variable power distance is a measure of the relationship between two persons having a disparity in power. In a high power distance culture, power structures tend to be more hierarchical and the difference in status and salary between superiors and subordinates large - such differences are more readily accepted and even expected than in a low power distance culture. In a low power distance culture, power structures tend to be flatter with more equality and interchangability between superiors and subordinates members of the society are considered to be more equal regardless of their position. Having identified this variable as being significant and correlated to the prevailing environmental culture, this variation can then be used to guide the design of an interface. For example, Marcus (2000) uses the power distance variable to propose that in a high power distance culture one should design the interface to include, e.g., more appeals to symbols of authority and information presented in a highly structured way with tall hierarchies. In a low power distance culture one should design the interface to include, e.g., more appeals to symbols from the user's immediate environment and information presented in a more exploratory fashion with flatter hierarchies. One could also design an interface that includes both approaches to information presentation and then adapts itself automatically to the cultural background of the present user. We now have two theories that appear, at least on the surface, to share common properties. Generative theory notes that sim ilar linguistic behaviours appear in a number of different languages. If this behaviour is common to all languages then we call. it a principle, if it is a behaviour shared by only some languages then we call it a parameter setting. Parametrically determined behaviour is related to a trigger phenomenon an external source of information embedded within the language of exposure that 'selects' the proper behaviour from · . a ' limited set of possibilities. Likewise, ' communication · theory:: ':i�entiftes' common behaviour across different (cultural) ' environments. . While it does not seek to identify behaviours that are universal to all cultures, as the researchers are specifically looking for variation, this does not mean that they do not exist. For instance, it is well know that the human visual system reflexively responds to movement - a universal characteristic of all humans. Such an observation is commonly used in interface design to catch the attention of the user with blinking text or icons. Those behaviours which are not universal but are present in a number of culturally disparate groups are identified and compared to the behaviour of other groups, identifying a

The scope ofPrinciples and Parameters theory 387

limited number of options that characterize the overall phenomenon - an attem pt to parameterize the phenomenon. These behaviours are then linked to the cultural input the learner is exposed to - identifying triggers in the culture of exposure. Although the jargon associated with each field differs, the intent is similar. Given the surface (and certainly gross) similarities between the two phenomena, can it be said that communication theory defines a bounded system of principles and parameters that determine information presentation preferences? Almost certainly not. Communication theory is, in its present form, manifestly incomplete and haphazard in its description of cultural variation. Only gross loci of variation have been identified and the amount of study that has gone into investigating the proposed variables is woefully inadequate. It is very likely that the theory is looking at too high level of behaviour and should instead be looking at more basic more closely tied to cognitive machinery. However, the . J.la.lUU,,",�� are intriguing - and, in the end, both relate to basic cognitive :p.hc�nofme:na. As a strong innatist, I am open to the possibility that each se),arate cognitive phenomenon could have a separate and different learning - but that is not the preferred explanation. Is the idea of learning based on innately specified variation a general mech�ml:sm? I would like to think that, in some way, it is. It is not ridllculOllS to me to suppose that there only a few innately determined ways people react to differences in personal power, or divide up their time, fee l part of a group. Communication theory, at a gross level, is doing early generative theory did - looking for clusters of similar behaviour different groups and trying to trace that behaviour back to a similar in the environments su�h groups were raised in. nfortunately, it is doing so w ithout an established theoretical framework ! ;PfC)Ce:ss that is bound to lead to incomp lete andlor incorrect hyp othes es . a framework is being attempted and in doing so useful observations being made - an intriguing start for someone steeped in the more : ;<1c�velo�)eCl linguistic theories. So, this is what I take back with me from my time in linguistics. A 'p�lra(iigJm which states that behaviours wiII be common across disparate and that such behaviour can be traced back to our cognitive �gi.a�chine)y. A theory which states that variation in behaviour will be limited that a given behaviour can be traced back to the presence of specific ' stimulli in the prevailing environment. This theoretical framework provides . me with a tool for analyzing cogn itively- rooted behaviour in general, looking for intra-group similarities tied to common stimuli. With such a

3 88 David LeBlanc

tool 1 can seek to identify the possible variations for a given behaviour, variation that can be accommodated in a particular interface design. With knowledge of the underlying basis of the behaviour, I can probe the user for their reactions to specific stimuli and automatically adjust the interface style in accordance with the predicted preferences. Communication theory in itself is an interesting field in which I hope to pursue further study. But for now, the gross observations that it provides are enough to design innovative adaptive interfaces for a global environment. What 1 have acquired from my study of linguistics is really the same as that which I acquired from my study of computer science - a set of basic tools for looking at the world and explaining observed behaviours. Are these linguistic tools general and can they be applied to the specific field I am now engaged in? That is the question I am presently addressing. Regardless of the outcome, I will not say that my time in linguistics has been for naught - the intellectual rigor and depth of analysis within the field have set a high theoretical benchmark for me (beyond the 'does it work' criteria common in my field), a standard to strive for regardless of the applicability of any specific linguistic theory or tool. For this is will always be grateful. References Chomsky, Noam.

1 956.

Three models for the description of language, IRE

Transactions on Information Theory, 2, 1 1 3 - 1 24.

Aspects of the Theory ofSyntax. MIT Press . Proceedings of the 18th annual meeting of the ACL. Davis, Henry. 1 987. The acquisition of the English auxiliary system and its relation to linguistic theory. Doctoral dissertation, UBC. Hall, Edward. 1 969. The Hidden Dimension, Doubl eday and Company, Chomsky, Noam. 1 965 .

Church, Kenneth. 1 980. On Parsing Strategies and Closure.

Incorporated.

Hofstede, Geert.

1 997.

Cultures and Organizations: Software of the Mind,

McGraw- Hi l l, New York.

LeBlanc, David. 1 99 5 . A n activation model of parameter setting. In Maaike Verrips and Frank Wijnen. (eds .) Approaches to parameter setting. Amsterdam series in child language development 4. Marcus, Aaron . 2000. Cultural Dimensions and Global Web User-Interface Design: What? So What? Now What? In Proceedings of the 6th Conference on Human Factors and the Web. Pollard, Carl & Ivan A. Sag. ( 1 994). Head-Driven Phrase Structure Grammar, University of Chi cago Press and CSLI Publications, Chicago and Stanford.

The scope o/Principles and Parameters theory 389 Rummelhart, D.E. and J.L. McClelland. (1 986). On learning the past tenses of English verbs. in Parallel distributed processing. MIT Press, Cambridge, MA. . Yang, Charles. (2003). Knowledge and Learning in Natural Language. Oxford University Press, Oxford.

Clitics and adj acency in Greek P Ps Winfried Lechner and Elena Anagnostopoulon

1. Introduction l

This squib explores a class of phenomena which intersect two central determinants in Henk van Riemsdij k's research: prepositions and adjacency. In the first domain, Van Riemsdijk ( 1 978, 1 990) established that the empirical study of PPs can contribute important theoretical insights into the organization of various subcomponents of natural language syntax, including the theories of phrase structure, locality and selection. In the second area, this time starting from an investigation of theory-internal desiderata, Van Riemsdijk ( 1 993, 1 998) extensively argues for a re­ initiation of the concept of adjacency into the grammar. The two groups of studies are also informed by different perspectives on syntactic theorizing, foregrounding the empirical and the theoretical aspect, respectively. As has been persistently stressed by Van Riemsdijk, adequate analyses can only arise if both aspects are combined in such a way that the importance of strong empirical generalizations is not sacrificed at the cost of claims based on ephemeral theoretical assumptions. Following this methodological guideline, the present squib elaborates on some factors determining the internal structure of intransitive PPs in Greek. As will be made explicit in section 2, Greek PPs display an interesting clustering of properties, which can be distilled into two generalizations describing correlations among word order, cliticization and selection. These correlations support the view that (intransitive) prepositions in Greek come in two different flavors, which differ in that one group emulates the kind of syntactic ('special '; Zwicky 1 977) cliticization properties typically associated with the IP domain (Anagnostopoulou 1 994, 2003 ; Terzi 1 999; among many others), while the other behaves as if it would host phonological ('simple'; Zwicky 1 977) clitics of the type usually found inside DPs (Nespor and Vogel 1 986; Alexiadou and Stavrou 2000 and others). The specific analysis of the contrast between syntactic vs. phonological clitics to be advanced makes reference to linear adjacency. Section 3 outlines a movement analysis for syntactic cliticization. Finally, I

We would like to thank Artemis Alexiadou for discussi.on.

Clilics and adjacency in Greek PPs 391

section 4 demonstrates that the data can be used to adjudicate between two different analyses of terminal displacement in PPs, eliciting evidence in support of a conservative head movement account, and against remnant movement. We would l ike to thank Henk for his inspiring and exemplary ability to dem onstrate that good taste prevails in life as well as in linguistics.

1. 1. Prepositions in Greek Greek prepositions fall into two distinct classes (see Theophanopoulou­ Kontou 1 992, 1 995; Fykias 1 994, 1 995; Terzi to appear) . Transitive prepositions require the presence of a DP complement carrying accusative . · tase, and split into two types, illustrated in ( 1 ) and (2). The two groups are A'�I''''·lrn inated by their variable ability to combine with other prepositions. ' While the functional, light prepositions in ( 1 ) (Riemsdijk 1 990) may also as prepositional complements (see ( 1 b» , the lexical, heavy elements (2) cannot be selected by other prepositions (see (2b» : Functional (light) prepositions: se ' to, in ', apo ' from , by', me 'with', gia 'for' . s- * (to spiti) tolin-the house 'to/in the house'

Me *(ton Petro) with the Peter 'with Peter'

gia *(ton Petro) for the Peter 'for Peter'

mesa s-to spiti inside in-the house ' inside the house'

mazi me ton Petro together with the Peter 'together with Peter'

anti gia ton Petro instead for the Peter 'instead of Peter'

Lexical (heavy) prepositions: pros 'towards', kata+Acc 'according to', mechri ' until, up to', os 'until, up to', eos 'until, up to', isame 'up to ' , san 'like'. pros *(to spiti ) towards the house 'towards the house' "'konta pros to spiti near towards the house

kata * (ton Petro) according the Peter 'according to Peter' *simfona kata ton Petro (cf. simfona me ton Petro) according according to the Peter (according with P.)

Unlike their transitive counterpart s , intransitive preposition s, exemplified by (3), may optionally also occur without an object.

392

Winfried Lechner and Elena Anagnostopoulou

(3) Intransitive prepositions a.

b.

c.

mesa(s-to spiti) inside in-the house ' inside the house' meta (apo to fagito) after (from the dinner) 'after dinner' iper (tu Petru) in favor of (the Petergen) 'in favor of Peter'

mazi (me ton Petro) together with the Peter 'together with Peter' prin (apo ton Petro) before (from the Peter) 'before Peter' kata (tu Petru) against (the Petergen ) 'against Peter'

kato (apo to trapezi) under from the table 'under the table'

enantion (tu Petru) against (the Petergen) 'against Peter'

In this squib, we focus on restrictions that affect intransitive prepositions, some of which have gone unnoticed so far. More specifically, intransitive prepositions (INPs) can be further classified according to whether they partake in one of three alternations. The resulting taxonomy lays the foundation for two new correlations to be introduced in section 2.2. First, some prepositions, such as the ones in (4a), allow their complements to surface as clitics, while others, some of which are listed under (4b), only license non-reduced DP objects: (4) a.

b.

mesa-tu inside clgcn 'inside him/it' * meta tu after cigen

mazi-tu together cIgco 'together with him/it' *prin tu . before cIgco

iper-tu in favor cIgcn 'in favor of him/it' *ektos tu except cisco

enantion-tu against cIgen 'against him/it'

Second, certain prepositions display variable categorial selection, as witnessed by the fact that they may combine · with . a 'nominal or with a prepositional complement. This group, exponents ofwhich are prov ide d in (Sa), contrasts with INPs th�t only tolerate DP comp.lerne.nts, as illustrated . . . . . . :. . . by ( 5 b).

Clities and adjacency in Greek PPs 393

(5 ) meta apo to fagito after from the dinner 'after dinner'

meta to fagito after the dinner 'after dinner'

ekso apo to spiti outside from the house 'outside the house'

*ekso to spiti outside the house

A third and final parameter divides prepositions according to admissible word order patterns. While the objects of the INPs in (6a) may be separated from their heads, the prepositions in (6b) require that their complements �urface in a position right-adjacent to the selecting category.

.2.

me ton Petro mazi with the Peter together 'together with Peter'

s-to spiti mesa in-the house inside ' inside the house'

apo to trapezi kato from the table under 'under the table'

*apo to fagito meta from the dinner after

*apo ton Petro ektos from the Peter except

*tu Petru enantion the Petergen against

A taxonomy ofINPs

three criteria introduced in section 2. 1 . generate a matrix with nine Three of these cells are actively employed by Greek, resulting in a tax�Jnomy which assigns INPs to one of three different categories. The first which includes the prepositions in (7), admits cliticization, selects for complements and tolerates object preposing, as specified in (8):

mazi 'together', mesa ' inside', ekso 'outside', pano 'above', kalo 'below' , konta 'close', makria 'far' , giro 'around' , dipla 'near', brosta 'in front', piso 'behind' a.

Properties of Class I INPs: Cliticization is possible: mesa tu inside itlhim

394

Winfried Lechner and Elena Anagnostopoulou

b. Optionally subcategorize for PP complements introduced by se, apo, gia, me: mesa sto spiti inside in-the house 'inside the house' c. Allow preposing of PP complements: mesa kimithike sto spiti siept3rd sg in-the house inside 'He/she/it slept inside the house' The second group, represented by iper 'for, in favor of , kata+GHN 'against' , and, enantion 'against 2 is characterized by the clustering of properties in (9). Class II prepositions differ from their class I counterparts in that they do not combine with PP complements and fail to license postpositional word orders: (9)

Properties of Class II INPs: a. Cliticization is possible: tu kata against him b. Subcategorize for DP complements tu/*sto Janni kata against the/to-the Janni 'against him/against Janni' c. No preposing of PP complements kata *miIise tu Janni spoke3rd sg the Jannigcn against 'He/she spoke against Janni'

The members of the final group of INPs, exemplified by meta 'after' , prin 'before" ektos 'apart' and anti ' instead , 3, share with class n the strict serialization requirement (see ( l Oc» . However, unlike other INPs, class III heads possess a variable subcategorization frame (see ( l Ob» . In addition, their complements resist cliticization (see ( l Oa» .

2

(because of) also belongs to this category, even though its DP

Eksetias

complement is obl igatory.

J

Just like

eksetias, anti must occur with

preposition.

a complement, i.e. it is not an intransitive

Clitics and adjacency in Greek PPs 395

Properties of Class III INPs: a. No cliticization: *meta tu (cf. mesa tu from class I) after it/h im h. Subcategorize for DP or for PP complements introduced by apo and gia: meta (apo) to proino after from the breakfast 'after breakfast' c. No preposing of PP complements: meta4 *efige apo to proino left3rd sg from the breakfast after 'He/she/it left after breakfast' •.. Tl1leolretlcal I) ,

it is possible to extract two correlations linking cliticization, selection and freedom of word order from the tripartite itwwn.omly of INPs above. On the one side, complement preposing appears be blocked whenever the preposition can in principle combine with a OP lement. On the other side, all and only those INPs tolerate cliticization select their complements from a single category (i.e. PP or OP, but not ): INP may in principle select for OP complement � object preposing impossible b. INP selects one category only (either PP or OP) � cliticization possible foHowing section presents the first steps towards an account of the effects which define INPs. As i t turns out, the correlations in

��I1I�tF�rm I O

� Some infonnants judge the string to be acceptable. These speakers construe meta

,.

a comparative temporal adverb that takes the apo-phrase as its comparative �;compllemlent (cf. after lunch vs. (no) later than lunch). As expected, adding a "se��on.d temporal modifier leads to ungrammaticality if the object precedes the INP (ib» . a. b.

chthes meta apo to proino yesterday after from the breakfast 'yesterday after breakfast' *chthes apo to proino meta yesterday from the breakfast after

396 Winfried Lechner and Elena Anagnostopoulou

( 1 1 ), which seem intuitively rather unnatural in the first place, will not be

part of the final analysis, though, and will be replaced by more substantive generalizations.

2. Analysis

It is well-known that cliticization is not a uniform process, but appears in different guises depending on several poorly understood factors (Van Riemsdijk 1999). This distinction is also attested in Greek, where cliticization of objects arguably involves a syntactic dependency - possibly movement - which preposes the clitic to the left of the verb, subject to syntactic locality (Anagnostopoulou 2003). In contrast, DP-internal cliticization inserts the weak pronoun in a position immediately to the right of the head noun (or adjective; Nespor 1 999; Alexiadou and Stavrou 2000). This adjacency requirement represents a typical hallmark of phonological cliticization. We suggest that cliticization with INPs of class I and II instantiates these two different strategies, respectively. To be precise, class I INPs will be assumed to behave akin to verbs, triggering syntactic cliticization, whereas INPs that belong to class II are nominal in nature and involve phonological clitics. At the moment, we do not understand why INPs are grouped in the way they are or whether the selectional differences can be independently motivated. However, as will be explicated below , adopting this specific view offers a promising new perspective on the clustering effects of section 2.2. To begin with, cliticization with class I INPs resembles indirect object cliticization in the extended verbaIIIP-domain, which in Greek implicates genitive morphology. As a syntactic procedure, cliticization is contingent on movement. We adopt two assumptions, the second of which can ultimately be eliminated. First, class I prepositions bear an edge feature (e.g. EPP) that activates SpecPP as a landing _site (Van Riemsdijk 1 978). Second, cliticization affects the whole complement of the preposition . . Since the complement is itself realized as a PP, the CIitic serves as a proform for a PP, and not a DP. It is for this reason that Greek displays contrasts such as ( 1 2): ( 1 2) a. mesa tu inside it b. *mesa s-to/tu insi de to-itAcJtUGen

Clitics and adjacency in Greek PPs 397

Thus, as schematized in ( 1 3), cliticization with class I INPs consists in movement of PP2 to the edge of the containing PP (SpecPP 1 ). Then, P I rai ses to the left of PP2, landing in a higher FP and restoring the original . word order P I "PP2:

[FP P I °

[PP I [PPI PP2Clitic [PPI PP2Clitic

P I o(Edge feature] PP2Clitic] => (movement of PP2) [P I O tpp] ] => (movement of P I O) tpp]]] [tPID

. The premise that class I prepositions are endowed with an edge feature, which opens an 'escape hatch' for extraction (Riemsdijk 1978), .furthermore correctly leads one to expect that the complement can be . preposed (see (8c) above), accounting for the second defining property of class I items. Thus, the analysis links the availability of cliticization an d . postpositional orders to a common factor, viz. the presence of an (active) specifier of PP. It was pointed out that cliticization affects PPs, but not their DP ' CCtmlllelme:nts. This restriction can be made to follow from the assumption the light prepositions that may head complements of class I 'PI'ep()sH:torls (se 'to, in', apo 'from, by', me 'with' and gia 'for' lack an feature. s As a result, the DP in ( 14) cannot raise to SpecPP2 and DP­ ization, which is contingent on movement, is blocked.6 [PPI P l o [PP2 P2 ° DPClitic]] *[PP I P I ° [PP2 DPClitic P2 °] ]

=>

(cliticization of DP)

Cliticization is also attested with class II prepositions, which h i �·tnrll' !:lI I ", originated in the formal katharevousa register. [n contrast to class I counterparts, class II INPs employ a cliticization strategy Har to the one characteristically associated with the nominal domain. In this sense, Greek class II INPs are more nominal in nature, while class I .

:rr Fil:nctional

pOs behave in this respect like the overt complementizer that, which ;been analyzed as a functional preposition and whose presence also inhibits nm'em.ent and/or extraction for adjacent categories such as subjects (Emonds ; Van Riemsdijk 1 990). Moreover, the idea that functional categories are for extraction dovetails with the view sometimes found in the . ' '. " :literature that definites, which are opaque for extraction, are DPs, while indefinites, 'Yhich tolerate extraction, are represented as (lexical) NPs (Longobardi 1 994). () �Ph()nological cliticization - see below - is never possible with accusative DPs in Greek.

398 Winfried Lechner and Elena Anagnosropoulou

INPs form a natural group with verbal predicates. Thus, cliticization inside DPs can be interpreted as the result of a local, phonological process which is subject to adjacency (on adjacency see Van Riemsdijk 1 993 , 1 998; Lechner 200 1 , a.o.). It is for this reason that the DP-complements of c lass II INPs may be cliticized (kata tu 'against it'), but cannot be preposed ( * tu Janni kata 'the lannis against'). Class II prepositions also mimic nouns in that they do not support syntactic cliticization. On the present conception, this deficiency can be tied to a common property of class II and nominal heads, both of which appear to lack an edge feature. Their complements are therefore not allowed to target SpecPP and SpecDP, respectively, and fail to satisfy the movement criterion for syntactic c liticization. Interesting independent support for such a uniform analysis of class II INPs and nominal heads comes from locality effects in DPs. More specifically, the hypothesis that Greek DPs lack an edge feature contributes to a new understanding of a puzzle in the analysis of DPs related to thematic restrictions on subextraction. Greek represents almost the mirror image of English in that it limits subextraction to possessors (as has been discussed extensively in the literature since the seminal work of Horrocks and Stavrou 1 987), whereas themes7 (in non-derived nominals) may not be moved:

( 1 5)

Tinos ekapses to biblio whose bumhnd sg the book 'Whose book did you bum'

( 1 6)

ekapses to biblio Jia pio thema about which topic burnt2nd sg a book '*Which topic did you bum the book about'

This curious property can �e directly derived from)he . d�ficiency analysis and the plausible additional assum ption that possessors', which serve the function of external arguments, are like subjects in that they are generated in the specifier of the minimal ly containing functional projection. It follows now that possessors, which originate in SpecDP, may escape the DP (sec ( 1 5» , as they satisfy the requirement that extraction necessarily proceed

7

Probably

discussion.

English is like Greek in this respect. See e.g. MU ller ( 1 995) for

Clitics and adjacency in Greek PPs 399

through a specifier even without movement.s On the other hand, the absence of an edge feature in Greek DPs ensures that categories which are generated in lower positions (e.g. themes) can never be attracted into the escape hatch SpecDP, which in tum blocks extraction in cases such as ( 1 6). In sum, class II INPs share two substantive characteristics of DPs: lack edge feature and phonological cliticization. The present account an ·· · of correlates these two properties by making syntactic cliticization dependent on movement. Members of the third class of INPs never cliticize (see (l Oa)), while their complements alternate freely between PPs and DPs (see (l Ob» . Finally, the absence of preposing (see ( I Dc» indicates that class III INPs lack an edge feature. On current views, at least OP complements should be able to undergo phonological cliticization to their host, though (but see rn. 6). Thus, the third class of INPs at first s ight appears to fall outside the present analysis. However, as will be demonstrated below, a slightly more �abstract parse for cases involving apparent DP complementation does not . remove this complication, but also strengthens the overall consistency ·of the account. Class III heads do not fit well in the taxonomy of lNPs for a second �;re�lSOJrl. they form the only group which does not subcategorize for a :uniquely specified category.9 But this ambiguity can be resolved once a .. slightly more abstract representation is admitted. Specifically, we suggest that class III INPs invariably select for PP complements, but that the head . this complement can be phonetically unrealized. As an additional ad'vatltaJge, this analysis offers an explanation for why prepositions that belolllg to class III do not permit phonological cliticization. As shown by (1 7), the c litic tu is separated from the contentful P-head meta by a zero preposition, violating the adjacency requirement on weak pronouns and their hosts, and can therefore not cliticize.

* [PPI meta [PP2 po tuClitic]]

. , The actual word order (to biblio tu Petru, i.e. D NP - Poss) has to be derived by further movement of NP and D across Poss. Note that a similar movement . 9peration is called for in the derivation of mesa tu, where the INP crosses over the .. . , clitic (see ( 12a) and ( 1 3»). Thus, a single additional assumption covers constructions. For the standard view according to which Greek DPs have edge . atures, see Horrocks and Stavrou ( 1 987) . . fe .. . 9 The ambiguity of class I is different in nature, as class I heads come in a transitive and an intransitive version. -

400

Winfried Lechner and Elena A nagnostopoulou

On this interpretation of the data, zero prepositions disrupt the relation between the c1itic and the host, suggesting that the calculation of adjacency does not only refer to phonological information, but is also sensitive to silent, yet syntactically projected, nodes. Recapitulating, lNPs in Greek can be categorized by the use of two properties - selectional criteria and the presence vs. absence of an edge feature - yielding the matrix in ( 1 8): ( 1 8)

Class I: Class II: Class Ill:

Edge feature +

Complement PP or 0 OP pp

Syntactic cliticization is restricted to 'verbal' prepositions of class I. These prepositions project an edge feature whose presence can also be detected in the ability of the prepositional complement to undergo preposing. Class II and class III elements constitute the group of what has been called nominal prepositions and do not sponsor juxtaposition. Even though both groups may in principle partake in phonological cliticization, structural adjacency limits the latter process to class II contexts. Returning finally to the two correlations formulated in ( 1 1 ), it is obvious - as was already foreshadowed in 1 .2 - that they only reflect a conflation of epiphenomenal properties. The analysis for a good reason neither recognizes a connection between potential selection properties and preposing (as encapsulated in ( 1 1 a» , nor a close relation between the number of altemative subcategorizatio� frames and cliticization (see ( l Ib» . 3. Complex alternations and head vs. remnant movement ..

As exemplified by ( 1 9), some class I prepositions enter into alternations with a synonymous variant (see ( l 9b» in which the light preposition (apo . 'from') is reduplicated in a position immediately preceding the contentful head (pano 'above'): ( 1 9) a. pano apo to trapezi above from the table 'above the table' b. apo pano apo to trapezi from above from the table 'above the table'

Clities and adjacency in Greek PPs 40 1

This final section presents an analysis of the doubling construction that also generates an argument in support of a head , and against a remnant movement account of the phenomenon. Assuming that the simple and the doubled variants are systematically related, preposition doubling of the type attested in (19) can be seen as the 10 result of multiple Spell-Out of a head-chain, as made explicit by (20). In . (20), movement of apo to the left of pano supplies the input to the . phonological component:

[PPI pano [PP2 apo [DP to trapezi]]] (20) a. b. [FP apo [PPJ pano [PP2 apo [DP to trapezi]]]]

=>

(movement of apo)

:Doubling is not an isolated phenomenon in Greek. For some rather poorly understood reason, Greek allows mUltiple Spell-Out of heads in a number . of well-defined contexts. For instance, the simple, adnominally modified DP in (2 I a) alternates with the doubled version in (2 I b) (on determiner 'SDlreaam!J! see Alexiadou and Wilder 1 998 among many others): to megalo trapezi the big table 'the big table' b. to megalo to trapezi the big the table 'the big table' .De:le��atlmg the details to future research, the mechanisms responsible for multiple Spell-Out in (2 1 ) are arguably sufficiently similar to the ones at " work in (20) in order to justify a common analysis of the two constructions. 00 this view, multiple spell-out of both copies of apo in (20b) directly feeds preposition doubling. 1 J )0 It is immaterial for present purposes whether the head in (20b) reaches its

.lt1;;:-&�g; derived position by movement, or whether apo is merged or generated twice. Both

options are compatible with the analysis presented below. U Similar doubling effects have been observed with prepositions in German. Oennan da 'there' is a prepositional proform, which fuses with in ' in' to drin :'therein' . In (i), the preposition in accordingly surfaces twice, in in and in drin. in der Kiste drin in the box there-in ' in the box'

402 Winfried Lechner and Elena Anagnostopoulou

The construction is of theoretical relevance inasmuch as it poses a challenge for recent attempts to eliminate head movement (Hinterholzl 1 997; Koopman and Szabolcsi 2000; Mahajan 2000; MUller 2004; Nilsen 2003). In particular, adherents of this school of thought, according to which displacement of tenninal symbols is to be reanalyzed as the product of possibly multip le remnant movement, need to find a derivation for ( 1 9b) which yields the correct surface order from the source in ( l 9a) without invoking displacement of heads. However, neither fronting of PP2, as in (22), nor extraction of the DP followed by remnant movement, as in (23), results in the desired output order. (For the sake of the argument, we adopt the liberal assumption that any part of the apo-PP can be spelled-out twice.) (22) a [PPI pano [PP2 apo [op to trapezi]]] b. *[[PP2 apo [op to trapezi]] [PPI pano tppz] c. Spell-out: *apo to trapezi pano (apo) (23) a. b. c. d.

�

[PPI pano [PP2 apo [op to trapezi]]] � pano to trapezi] � [[op [PP2 apo top]]] [PPI [[PP2 apo top] [PP I pano [[oP to trapezi] tpP2 ]]] Spell-out: *apo pano to trapezi (apo)

(movement of PP2)

(movement of DP) (movement of PP2)

Alternatively, remnant movement analyses might also posit the derivation in (24), which is like (23) but involves an additional movement step that generates an occurrence of apo (marked by underlined) i n the correct location inbetween pano and to trapezi: (24)

[ [PP2 apo top] [PPI pano [[PP2 apo top] [[oP to trapezi] [PP2 apo top]]]]]

Such a derivation suffers from two deficiencies, though. F irst, it is not clear why only the first two copies of apo can be pronounced .. Whereas o� the head movement account, dqubling tracks the n\!lllber of syntactic copies, no such simple procedure ' can be defined for remnant movement analysis. Second, remnant movement is usually assumed to be determined by the following algorithm (see e.g. Nilsen 2003):

1 . Move a to the left of its dominating host p: 2 . Merge new category y: 3 . Move p over the newly merged category y:

[po a ]

[a-[13 tal] [1 [a- [13 tal ] ] [[ P fa ] [1 [a til ]]] =>

Clitics and adjacency in Greek PPs 403

Hence, the apo-PP in (24) (PP2, which instantiates p) must immediately move across the newly merged category pano (1 in (24», and cannot land in an intermediate position in between pano and the moved category to trapezi, as in (24). Unless substantial changes are admitted, the system does therefore not sanction the alternative output (24). The doubling construction may also be targeted by cliticization, 'yielding the string apo pano tu 'from above it' . Adopting the analysis of section 3, the derivation proceeds as in (25). To begin with, apo raises into a higher functional head, resulting in doubling (see (25b» . Next, cliticization relocates PP2 into SpecPP I (see (25c» . Finally, pano undergoes local raising to a higher functional projection GP, yielding the output apo pano tu in (25d). (25) a. [pp) pano [PP2 apo DP]] :::::> (movement of apo) b. [FP apo pano [PP2 apo DP] ]] ::::::) (cl iticization PP2) [PPI c. [FP apo ::::::) (movement ofpano) [PPI tu pano tpP2]] d. [FP apo [GP pano [PPI tu tpP) tpP2 ]]] . Two notes are in order at this point. First, observe that on current assumptions, cliticization targets PPs. It follows that the alternative, illicit forms in (26), which implicate DP-cliticization, cannot be produced by the analysis. C learly, this represents a desirable result. �(26) a *apo pano tu apo . b. *apo pano apo tu

.

,Second, pano in (25) needs to satisfy the head movement constraint. This requirement dictates thatpano must not cross over apo in the higher FP, but must choose its landing site inbetween apo and the clitic tU. 12 Thus, the unattested serialization *pano apo tu can be excl uded on principled • · grounds.

''.i The derivational step in (2Sd) presents an instance of tucking in (Richards

1997)

· \vith head movement. We assume that apo can evade the HMC when crossing over "pano either because the relation between the two copies does not involve ; :movement (see fn. 1 0), or because the HMC is only defmed for empty categories.

Note that tucking in does not apply when XPs move across higher thematic ··· ;> specifiers (Anagnostopoulou 2003 ; Rackowski 2002), just as apo does not tuck in below (base-generated) pano in (25b). The claim that (2Sd) respects syntactic locality (the HMC) is also supported by the observation that resumption by , pronouns in many contexts salvages otherwise iIl-fonned movement relations :

404

Winjried Lechner and Elena Anagnoslopoulou

Again, it can be demonstrated that the remnant movement analysis fails to provide the correct results. Crucially, if cliticization applied prior to movement, at the step documented in (27a), the predicted output word order would be pano tu. This string is also attested, but not the one the analysis is intended to generate. (27) a. b. C.

[PP2 apo OP]] [PPI pano [PPI pano [O P [PP2 apo tDP]]] [[PP2 apo tDP] [PP I pano [ OP tpPl]]]

=> =>

(movement of OP) (movement of PP2)

Moreover, cIiticization cannot apply after remnant movement has been completed, at step (27c), because the process would have to target a OP instead of a PP, in contradiction to an empirically well-supported generalization about INP internal cliticization. Among others, admitting OP cliticization in (27) would incorrectly lead one to expect that fonns such as *pano apo tu should be part of the grammar. Hence, the remnant movement account does not offer a descriptively adequate alternative vehicle for expressing restrictions on serialization inside Greek INPs. References

Alexiadou, Artemis and Melita Stavrou. 2000. Adjective-Clitic Combinations in the Greek DP. In Clitics in Phonology, Morphology and Syntax, ed. by Birgit Gerlach and Janet Grijzenhout, 63-84. Amsterdam: Jolm Benjamins. Alexiadou, Artemis and Chris Wilder. 1 998. Adjectival modification and multiple determiners. In Possessors, Predicates and Movement in the DP, ed. by Artemis Alexiadou and Chris Wilder, 302-332. Amsterdam: John Benjamins. Anagnostopoulou, Elena. 1 994. Clinc Dependencies in Modern Greek. Doctoral Dissertation, Salzburg: Salzburg University . Anagnostopoulou, Elena. 2003. The Syntax ofDitransitives: Evidence from Clitics. Berlin, New York: Mouton de Gruyter. Emonds, Joseph E. 1 985. A Unified Theory of Syntactic Categories. Dordrecht: Foris Publications. Fykias, Ioannis. 1 994 . Ot 1tpOO£'tlKt� cppa(J£l� <m1 Ntn EUllvlKTt. In Studies in Greek Linguistics. Proceedings of the J 5,h Annual Meeting of the Department of Linguistics, 352-359. Thessaloniki: University of Thessaloniki. Fykias, Ioannis. 1 995. The case system in Modem Greek: evidence from PPs. In Proceedings of the 2nd International Conference on Greek Linguistics, ed. by Gaberell Drachman et aI., vol II: 439-448. Graz: W. Neugebauer Verlag GmbH.

Clities and ac!jaeency in Greek PPs 405

HinterhoizI, Roland. 1 997. A VO-based Approach to Verb Raising. In Proceedings ofNorth East Linguistic Society, ed. by Kiyomi Kusumoto, 1 87-202. Amherst: University of Massachusetts, GLSA. Horrocks, Geoffrey and Melita Stavrou. 1987. Bounding theory and Greek syntax: evidence for wh-movement in NP. Journal ofLinguistics 23: 79- 1 08. Koopman, Hilda and Anna Szabolcsi. 2000. Verbal Complexes. Cambridge, Mass .: MIT Press. Lechner, Winfried. 200 1 . Structural and Linear Adjacency. Presentation at GLOW 24, University of Porto, April 8- 1 0 200 I . gobardi, Giuseppe. 1 994. Reference and Proper Names. Linguistic inquiry Lon 25.4: 609-666. Mahajan, Anoup. 2000. Eliminating Head Movement. GLOW Newsletter 44: 4445. Miiller, Gereon. 1 995. A-bar syntax. A study ofmovement type. Berlin, New York: Mouton de Gruyter. Muller, Gereon. 2004. Verb second as vP-first Journal of Comparative Germanic Linguistics 7: 1 79-234. · Nespor, Marina and Irene Vogel. 1 986. Prosodic Phonology. Dordrecht: Foris Publications. . Nilsen, 0ystein. 2003. Eliminating Positions. Doctoral Dissertation, Utrecht: OTS. Rackowski, Andrea. 2002. The Structure of Tagalog: Specificity, Voice, and the Distribution ofA rguments. Ph.D. Dissertation, MIT. ! Richards, Norvin . 1 997. What moves where in what language. Ph.D. Dissertation, MIT. . Riemsdijk, Henk van. 1 978. A Case Study in Syntactic Markedness. Dordrecht: Foris Publications. Riemsdijk, Henk van. 1 990. Functional Prepositions. Unity in Diversity. Festschrift , for Simon DUe, ed. by Harm Pinkster and Inge Oenee, 229-24 1 . Dordrecht: Foris Publications. Riemsdijk, Henk van. 1 993. Complements. Adjuncts and Adjacency in Phrase Structure. In De la musique a la linguislique: hommages a Nicolas Ruwet, ed. by Liliane Tasmowsky and Anne Zribi-Hertz, 498-5 12. Ghent: Communication and Cognition. Riemsdijk, Henk van. 1 998. Head Movement and Adjacency. Natural Language . and Linguistic Theory 1 6.4: 633-678. �iemsdijk, Henk van. 1999. Clitics: A state-of-the-art report. In Clilics in the Languages of Europe, ed. by Henk van Riemsdijk, I -30. Berlin, New York: Mouton de Gruyter. . Terzi, Arhonto. 1 999. Clitic combinations, their hosts and their ordering. Natural Language and LinguistiC Theory 1 7: 85- 1 2 1 . Arhonto. To appear. Locative prepositions, predicate inversion and full interpretation. In Proceedings of the 1 7'h International Symposium of theoretical and applied linguistics. Thessaloniki: University of Thessaloniki.

406 Winfried Lechner and Elena Anagnostopoulou Theophanopoulou-Kontou, Dimitra. 1 992. Ot
A minim alist program for parametric linguistics? Giuseppe Longobardi

1. Parametric linguistics

l

Parametric linguistics may be viewed as a distinct subfield of generative grammar, or more generally of the abstract biolinguistic approach developed in the past decades of research, since at least Chomsky ( 1 955, 1 965 ) and Lenneberg ( 1 967). It presupposes the crucial result acquired through this approach, namely that human language comprises a rich system of invariant innate knowledge, and concerns itself with the further problem of a formal and principled theory of grammatical diversity, a privileged testing ground for theories of the interaction between biologically shaped structures and culturally variable information. 2 Such a subfield is minimally defined by three fundamental questions: a. What are the actual parameters of UG? b. How do parameter values distribute in space and time? c. What is the form of possible parameters? Question ( l a) has been addressed in a rather inductive form by the 'hundreds of case studies proposing all sorts of morphosyntactic parameters . over the past twenty years. Longobardi (2003) observed that hardly any 'significant module of grammar, small though we may take it, has so far :�ttained a degree of parametrization with any pretension of typological exhaustiveness. To remedy this situation empirically it was suggested that 'linguists, rather than just existentially asserting the presence of certain parameters, should explicitly aim at a 'modularized global parametrization', i.e. at universal restrictive hypotheses on how widely Jimguages may vary in very circumscribed grammatical modules. �< '

§gg;"

1

I am indebted to Michael Arbib. Hans G. Obenauer. and the participants in the Conference on Parametric Variation (September 2005) organized by ..-........... .. Holmberg for inspiring discussions and comments on the topics of this squib. Longobardi (2003) for some more extensive remarks and methodological . proposals.

.Nt�wc:astle

}:,Cf.

408 Giuseppe Longobardi

Question ( 1 b) is of a different nature and opens the way to the fascinating unification of two traditionally different domains, historical linguistics (with some connections reaching into paleoanthropology) and the cognitive approaches of formal grammar. This question has begun to be explored in Guardiano and Longobardi (2005), Gianollo, Guardiano, and Longobardi (2004), and will not be pursued in this paper. Thus, the present note will deal with question (l c) and how its study may renovate our understanding of question ( 1 a) and, indirectly, ( l b) as well. The only well known restriction proposed on the format of parameters is so far the conjecture, stemming from Borer ( 1 984), that parameters are always properties of functional heads of the lexicon. Fundamentally accepting this insight as a point of departure, in what follows I will suggest the possibility of a more articulated restrictive theory and point out some of its desirable consequences. 2. Parameters: problems and methods

In order to fruitfully address ( 1 c) it may be useful to raise some more particular guiding issues, such as the following evolutionary ones, and try the adequacy of the proposals against them:

(2) a. Why is there so wide grammatical variation (so many parameters)? b. Why is there grammatical variation (Le. parameters) at all?3

Longobardi (2003) noticed that these problems have become more acute precisely with the development of par�etric approaches. Thus, the classical generative model (e.g. Chomsky 1 965, call it the Aspects model for convenience) viewed the Language Acquisition Device (LAD) as consisting of a set of universal principles ' (Universal grammar, UO) and of an Evaluation Metrics for grammars,' valuing grammars freely constructed by the language l Jeamer within the i :lJounds " posed py the Universal Principles. Within such a model, the existence of variation is potentially explained in terms of minimization of genetic endowment; for the 'freedom' of variation allowed can be construed as the complement of the principles made available by UG : the more such restrictive principles, the more the number of languages prohibited. Therefore, the width of variation could be due to the fact that the amount of universal restrictions 3

The relevance of the latter question was first raised to my attention by Richard Kayne (p.c.) some years ago.

A minim alist program/or parametric linguistics? 409

made avai lable by human nature is limited by a sort of 'memory load' constraint on genetic transmission of cognitive information, a conceivable 'economy' condition on the architecture of the LAD, active through evo lutionary history. In the later Principles-and-parameters (P&P) model, LAD consists of a UG with both universal principles and parameters: = Principles + Parameters. Open parameters at So, closed parameters at Ss

ua

In principle, then, grammatical variation is also innately given (exhaustively . given, at the appropriate level of idealization), under the form of a presumably finite amount of discrete possibilities. Variability is already present at the initial state of the mind So in the form of open parameters, actual varieties are represented by closed parameters at the steady state Ss. In this model the existence of variation is hardly explained, and certainly . cannot be explained by the previous line of reasoning: for lim iting the amount of transm ittable genetic information, i.e. the size of the LAD, should presumably reduce the number of possible parameters as well, therefore increase, rather than decrease, the degree of invariance of the i'language faculty observable across individual languages. This puzzle will constitute a good start for an investigation of I:estrictions on parameters. Another prerequisite to attempt a restrictive . , of parameters is a sufficient collection of structured data about :,N�lfi�lti(Jln and a way of representing such information in a perspicuous form . approximate this objective realistically, it has been suggested in (2003) to adopt the strategy of Modularized Global P arametrization, alluded to above. Trivializing matters a little, this method , can be summarized in the following formula: studying relatively many : parameters across relatively many languages in a single module of ' . grammar. Considering a certain number of parameters together is obviously necessary to attempt any generalization; observing more than just a pair of :cp:ntr.astl·m g languages for each parameter is required of a theory with some of typological completeness; and concentrating on a single ::mc)(iule makes the enterprise more realistically feasible but also allows one to' explore a major formal feature of parameter sets, as already emerging from the works of Fodor (2000), Baker (2001 ), and Guardiano and , J�ongobardi (2005), namely their widespread interdependence (cf. ' � :�mmediately below). The MGP method seems thus to be an appropriate compromise between depth and coverage. Following this method, Gianollo,

4 1 0 Giuseppe Longobardi

Guardiano, and Longobardi (2004) have set up a grid of 49 parameters affecting the internal structure of Oetenniner Phrases and tentatively stated their values in 23 languages and their partial dependencies, i.e. the absolutely pervasive situations where choosing one of the two values of a parameter neutralizes the relevance of the valuing of another parameter. All such parameters could be formulated as binary and their values have been marked in the adopted formalism as + and -. When the value of a parameter depends entirely on the value of (one or more) other parameters it is marked with a 0 (cf. Gianollo, Guardiano, and Longobardi 2004 for further details). This approach and formalism produce parametric grids summarizing large amounts of empirical information and theoretical hypotheses, highly valuable for further speculation on the theory of parameters itself. 3. Parametric minimalism

On the grounds of the empirical material so collected it becomes possible to address question ( 1 c) and, more generally, raise an issue like (4): (4)

Can we subject parameters and their fonnats to minimalist critique?

Consider now that, as a first approximation, it turned out that at least 4 1 out of the 49 parameters of OP-internal structure investigated in Gianollo, Guardiano, and Longobardi (2004) can be reduced to a form which falls into one of only 4 abstract parameter schemata, listed below, where F and X,Y are variables over features and categories: _

(5) a. Is F, F a functional feature, grammaticalized? b. Is F, F a grammaticalized feature, checked by X, X a lexical category? c. Is F, F a grammaticalized feature, spread on Y, Y a lexical category? d. Is F, F a grammaticalized feature checked by X, strong (i.e. overtly attracts X)? By ' grammaticalized' in (Sa) it is meant that the feature must obligatorily occur and be valued in a certain structure, e.g. definiteness i s obligatorily marked in argument DPs in certain languages (say English), not in others (say Russian). This does not mean that even the latter languages cannot

A minimalist program/or parametric linguistics?

411

have lexical items occasionally used to convey the semantic meaning of definiteness (presumably demonstratives can convey such a meaning in every language), but in this case the feature 'definiteness' would be regarded as a lexical, not a grammatical one. (Sb) asks whether a certain feature requires establishing a relation with a specific (optionally or obligatorily present) category in the structure, creating a dependency (acts as probe searching a certain syntactic space for a goal, in Chomsky's 200 1 terminology). (Sc) asks if a feature which is interpreted in a certain structural position has un interpretable occurrences, depending in value on it, on other categories. Finally, (Sd) corresponds to the traditional schema inaugurated by Huang ( 1 982), asking whether the dependency mentioned in (Sb) involves overt displacement of X, i.e. remerging of X next to F, or not. I will leave the question open, for the time being, whether, considering other domains, a fifth recurrent parameter schema needs to be added to such formats, namely one asking whether a certain category or feature may be phonetically null in a certain situation (e.g. the well known 'null pronoun' schema, as inaugurated by Taraldsen 1 978 and especially refined !in Rizzi 1 986); it is possible that, in some cases at least, this sort of parametrization is reducible to environmental factors and combinations of other parametric schemata above, therefore does not instantiate a separate schema. Let us then suppose, very speculatively, that these are the only possible :'core' parameter schemata: from this approach it already follows that . certain conceivable types of variation are excluded; there follows, e.g., a conclusion with far-reaching consequences, such as (6): The locus of interpretation of each grammatical feature is universal, not parametrized exemplify the functioning of the parameter schemata system, consider the answers (7a-d) provided, on the grounds of the evidence tentatively built into the specific parametric grid of Gianollo, Guardiano, and Longobardi (2004), to the four respective questions asked in (5) once the ·:values for the variables over features and categories have been fixed:

.To

4 1 2 Giuseppe Longobardi

(7) a. b. c. d.

For F = definiteness, X = demonstratives, Y = relative clauses Yes: Italian, No: Russian Yes: Italian, No: Greek Yes: Wolof, No: Italian Yes: Bulgarian, No: Norwegian

This way, parameter schemata of the type of (5) derive actual parameters, which can be literally constructed out of functional features, lexical categories, and parameter schemata, and set under usual assumptions. If this approach is correct in its essentials, it becomes unnecessary to suppose that the initial state of the mind consists of highly specific parameters, but just of an incomparably more restricted amount of parameter schemata, which combine with the appropriate elements of the lexicon (features and categories) under the relevant triggers in the primary data to both yield the necessary parameters and set their values for each language:

(8)

Principles&Schemata model: UG = principles and parameter schemata. Parameter schemata at So, closed parameters at Ss

It is then conceivable that parameters which seem to end up being set on 0 (according to the formalism of Gianollo, Guardiano and Longobardi alluded to above) in a particular I-language have actually never been present at any state in the mind attaining that I-language. The enormous number of possible core parameters depends, in principle, on the more limited numbers of functional features F and of lexical categories X, Y, combine� with the tiny class of parameter schemata. Notice, however that it is not necessary that all parameter schemata are realized for every possible functional . feature . and all potentially relevant categories: specific principles ; of UG might forbid variation of an a priori admitted format for · parti�ular combinations of features and categories. Descriptively claiming� : fOf; example, as is often done, that the so called EPP feature is 'universally strong' amounts to preventing a widespread schema of variation among languages from determ ining differences as to whether the Spec of T is overtly filled or not. 4.

The issues of variation

Accepting the Principles&Schemata model immediately determines the possibility of huge arithmetic simplification in the primitive axioms of the theory of grammatical variation: exactly l ike parameters were adopted

A minimalist program/or parametric linguistics? 4 1 3

(also) as cross-constructional generalizations, significantly reducing the amount of apparently atomic points of variation, parameter schemata, in the intended sense, are more abstract, cross-parametric entities, allowing . further simplification of the set of primitives. This begins to provide a sensible answer to problem (2a), because the amount of variation itself to be explained is drastically reduced: it will be sufficient to justify the . existence of a certain parameter schema through justification (e.g. reduction to 'virtual conceptual necessity', in Chomsky's 1 995 sense) of a single parameter of that schema, in order to explain the possibility (ultimately, the . ' evolutionary rise) of the whole fam ily of parameters of the same format. But such an approach already relieves the burden of the explanation for •... the very existence of language diversity (issue (2b» as well : for, within the osed model, variation could largely be explained as in the first, pre­ .��;; prop �'�;:;;� P&P, model . As we have just noted, once the introduction of a parameter schema into the language faculty is justified (e.g. evolutionarily explained, perhaps reducing it to conditions of efficiency on language transmission use) for one case, it will be adm itted and cause proliferating potential var'lat'lon for all possible combinations of relevant entities of the lexicon \(te:atulres and categories). This, unless a further particular principle of UG i:tmJhill>.ts certain types of variation: in other words, once a schema has .en1:ere:d UG, then reducing variation essentially requires adding to the size LAD, exactly as in the Aspects model. The kind of explanation in terms 'economy of ua size' usable in that model can therefore be reproposed the Principles&Schemata approach. Of course, in order for a full minimalist program to be pursued within approach it is necessary to show each of the parameter schemata to be �;'illdi:spe:ns;able. i.e. reducible to virtual conceptual necessity, or at least to be related to architectural/computational properties present in biological systems. This whole, crucial, part of the program cannot be seriously addressed . , especially within the narrow limits of the present work, Only some ¢xemplification of the required direction of research can be provided. For example, (5a) could be motivated again by 'economy' constraints performance (no language could grammaticalize the full set of ,: :cofl(;enrable functional features) to be spelt out by specific research; (5c) �,j,iC�luld perhaps be ultimately related to an acquisitional strategy of formal of morphological content under the pervasive diachronic of categorial shift or reanalysis (say, of a lexical item from a , class where the occurrence of certain features is interpretable to another one . 'Where it is not).

,_"

4 1 4 Giuseppe Longobardi

Although this sketch is very preliminary, it should provide a reasonable idea of the heuristic power of the program advocated here and of the perspective of applying Chomsky's ( 1 995, 2005) groundbreaking minimalist suggestions to the domain of grammatical variation. References

Baker, Mark. 200 1 . The Atoms of Language. New York: Basic Books. Borer, Hagit. 1 984. Parametric Syntax. Dordrecht: Foris. Chomsky, Noam. 1 955. The Logical Structure ofLinguistic Theory. Ms . (published in 1 975, New York: Plenum). Chomsky, Noam. 1 965. Aspects of the Theory of Syntax. Cambridge, Mass: MIT Press. Chomsky, Noam. 1 995 . The Minimalist Program. Cambridge , Mass: MIT Press. Chomsky, Noam. 200 1 . Derivation by phase. In Ken Hale: A life in language, ed by Michael Kenstowicz, 1 -52. Cambridge, Mass: MIT Press. Chomsky, Noam. 2005. Three Factors in Language Design. Linguistic Inquiry .

36. 1 : 1-22. Jan et D. 2000. Setting syntactic parameters. In The Handbook of Contemporary Syntactic Theory, ed. by Mark Baltin and Chris Collins. Cambridge, MA and Oxford, UK: B lackwell 730-767. Gianollo, Chiara, Cristina Guardiano, and Giuseppe Longobardi. 2004. Historical

Fodor,

,

Implications of a Fonnal Theory of Syntactic Variation. Paper presented at DIGS VIII, Yale. In press in Proceedings of DIGS VII! ed. by Stephen Anderson and Dianne Jonas. Oxford: Oxford University Press. Guardiano, Cristina and Giuseppe Longobardi. 2005. Parametric Comparison and Language Taxonomy. In Grammaticalization and Parametric Variation, ed. by Montserrat BatIlori, Maria-LluYsa Hernanz, Carme Picallo, and Francese Roca, 1 49- 174. Oxford: Oxford UniverSity Press . Huang, C.- T. James. 1 982. Logical Relations in Chinese and the Theory of Grammar. D octoral dissertation, MIT, Cambridge Mass. Lenneberg, Eric H. 1 967. Biological Foundations of Language. N ew York: John W il ey & Sons. Longobardi, Giuseppe. 2003 . Methods in Parametric Linguistics and Cognitive History. Linguistic Variation Yearbook 3, 2003, 1 0 1 - 1 38. Ri zzi, Lui gi 1 986. N ull Objects in Italian and the Theory ofpro. Linguistic Inquiry .

1 7.3 :501-557.

Taraldsen, Knut Tarald. 1 978. On the NIC, vacuous application, and the thal-t filter. Ms. MIT, distributed by Indiana University Linguistics Club, Bloomington IN .

A syntactic approach to negated focus questions in Bulgarian Krzysztof Migdalski

1. Introduction

. This paper discusses placement of the interrogative complementizer Ii with respect to negation, clitics, and finite verbs in Bulgarian. This is a problematic issue in the syntax of Bulgarian, which has given rise to a lot of controversy in the literature. According to many previous analyses, the position of Ii in the clause structure is detennined by its prosodic requirements. I wil1 point out shortcomings of the accounts that rely on prosody and will demonstrate that the placement of Ii is syntactically constrained. Properties of Ii

.

is an interrogative particle found in most Slavic languages. In Bulgarian be preceded by many different constituents, both heads and phrases . Rudin ( 1 986:64) shows that when the preceding word is a verb, the . whole sentence is interpreted as being questioned. ·U ' '''' <:I''

a.

Kupil Ii e vestnika? bUYpart.M.sg Q beaux.3sg newspaper-the 'Did he buy the newspaper?' b. Pitaxa me kupil Ii e vestnika asked3pl meacc bUYpart.m.sg Q beaux.3sg newspaper-the 'They asked me whether he (had) bought the newspaper' (Rudin 1 986:63)

when Ii is attached to a non-verbal constituent, the scope of .>mten·oglatlC)O falls on this constituent. In such cases Ii licenses a focus feature on the fronted element in Spec, CP (Rudin et al. 1 999).

:0' .See Franks (2005) for a comprehensive overview of syntactic and semantic ·· properties of Ii in Bulgarian.

4 1 6 KrzysztoJMigdalski (8) a.

Kiistata Ii namerixte (vie)? house-the Q found2pl you (Rudin 1 986:64) 'Was it the house that you found?' b. Prez gardinata Ii xodese? through garden-the Q walked2sg 'Were you walking THROUGH THE GARDEN?' (Rudin et aI. 1 999:546)

Moreover, Ii can be used for questioning single words in isolation, which m ight be heads and XP's. (9) a. Az Ii

I Q 'Me?' b. Na masata Ii? on table-the Q 'On the table?'

(Rudin 1 986:65)

Li is an enclitic, so it needs a phonological host to its left; but it must also

precede all the other clitics when it clusters together with them. The ordering in which it occurs is indicated in the template in ( 1 0) and in the examples in ( 1 1 ). ( 1 0)

Ii > Neg > AUX (except 3sg) > OAT > ACC > REFL > AUX3 sg (cf. Franks and King 2000)

go? ( 1 1 ) a. Davate Ii mu give2pl Q him dat.cl., itaec.CL 'Do you give it to him?' h. Dali Ii ste mu go? givepart.pl Q beaux.2pl himdat.cl., itacc.cL 'Have you given it to him?'

(Hauge 1 999: 1 1 7)

However, the sequence of clitics is disturbed in the presence of negation, which may attract a single clitic to the position in front of Ii. As shown in ( 1 2), clitics are arranged hierarchically with respect to each other, and only the highest clitic in the cluster may be fronted. Hence, it is possible to front the accusative clitic go (cf. ( 12a» only in the absence of the dative clitic mu (cf. ( 1 2c» . If both pronominal clitics are present, only the dative one may be preposed (cf. ( l 2b» . Moreover, it is impossible to front more than one

����:,

�l£

Negatedfocus questions in Bulgarian 4 1 7

:���, elitic (cf. (l 2e» . Apart from pronominal clitics, auxiliary clitics and non­rd �.t�i:) elitic finite verbs can be moved as well (cf. ( 1 2e,f) . However, since the 3 :;" person singular auxiliary cJitic follows all the others in the cluster (cf. the ' , template in ( 1 0» , it may not be preposed across the pronominal clitics (cf. (12g» .

b. c. d. e. £

g.

Ne go Ii iskate NEG itacc.CL Q want2pl 'Don't you want it?' Ne mu Ii go dadoxte? NEG him dat.cL Q itacc.CL gave2pl 'Didn't you give it to him?' *Ne go Ii mu dadoxte? NEG itacc.CL Q him dat.CL gave2pl *Ne mu go Ii dadoxte? Ne iskate Ii? NEG want2pl Q 'Don't you want to?' Ne ste Ii mu go dali? NEG beaux.2pl Q himdat.cL itacc,CL givepart.pl 'Haven't you given it to him?' *Ne e Ii mu go dal? NEG beaux,3sg Q himdat.cL itacc.CL givepart.M.Sg (Hauge 1 999: 1 08; Svetoslav Marinov, p.c.)

a;"lSoJlIle previous accounts

fact that the ordering of clitics can be disturbed in the presence of ' ne�,atl(m has been a long-standing problem of Bulgarian syntax and has -ece:lve:a a number of different analyses. For example, Rivero ( 1 993) claims in the context of neutral yes-no questions, such as ( 1 1 a and b), the verb as a head to C to host the clitic Ii, which may not appear in the initial position. In negated clauses, however, the verbal head may . ' move to C, because negation is a barrier for the movement. .'ucms1eqllen:t1y, there is no host for pronominal clitics. In order to save the ;fle:rivj�.ti()fl Ii must lower ('hop') in the structure. It adjoins to I, and as a it follows the clitics, suc� as mu in ( 1 3), but precedes the finite verb.

41 8

( 1 3)

KrzysztoJ Migdalski Ne

mu Ii izpratix himdat.cL Q sendpast.)sg 'Didn't I send him a book?'

NEG

kniga? book

(Rivero 1 993 :5 73)

Izvorski et al. ( 1 997) notice empirical problems with Rivero's account. First of all, negation cannot be a barrier for all verbal heads, because non­ clitic finite verbs may move in front of Ii (cf. ( 1 2e» . Second, Rivero's analysis predicts that if there are several clitics in a negated question, all of them will precede Ii, given that Ii must lower and adjoin to I. In reality this never happens, and Ii always appears after the first clitic, irrespectively of the number of clitics that are present in the cluster. ( l 4) a. Ne

go Ii e himacc.cL Q beaux.3sg 'Didn't he see him?' b. Ne mu Ii go NEG himdat cL Q itacc.CL 'Didn't you give it to h im?' NEG

.

vizdal? seepart.M.sg dadoxte? gave2pl

(lzvorski et at. 1 997: 1 9 1 )

According to lzvorski et ai, the crucial property of Ii i s that it requires a phonologically overt host capable of bearing stress to its left. Therefore, they suggest that if a host of this type is missing, Ii will undergo prosodic inversion and will cliticize to the right edge of the first word that bears stress. The process is claimed to take place at PF.

( 1 5)

I

+

(Izvorski et al. 1 997: 1 93)

Li izpratix mu kniga h imdalcL book Q sent 1 sg 'Did I send him the book?' Negation does not carry stress on its own in Bulgarian, so it is not a suitable candidate for giving support to Ii. However, it has the property of shifting stress over to the word that follows it, even if it is a clitic. Hence, in negated clauses Ii will move across the first clitic to its right (bold faced in ( 1 6» , which will carry stress in this context.

Negated/Deus questions in Bulgarian 4 1 9

( 1 6)

Ne

mu

Ii

go dadoxte? itacc.cL gave2pl 'Didn't you give it to him?'

NEG himdat.cL Q

(Izvorski et al. 1 997: 1 93i

Without appealing to a mechanism of prosodic inversion, Boskovic (200 I ch 4.3. 1 . 1 ) analyzes the data under a 'scattered deletion' approach. What is important for his account is the fact that Ii must occur after the first stressed constituent in the clause. He argues that the string ne go vidja in ( 1 7) forms a complex head, which arises through adjunction of the pronominal clitics and negation to the verb. vidja] [Nego NEG himacc.cL saW2/3sg ' He/she Iyou didn 't see him'

(Boskovic 200 I : 208)

Next, the complex head left-adjoins to Ii. [ne go vidja+li] He ge 'Iidja .

results in a phonologically infelicitous structure, because ne assigns to the word that follows it. In the case at hand, the clitic go is strf�sse�<1. The problem is that vidja is stressed, too, because it is a lexical with its own lexical stress. However, Ii must occur after the first stre�sSf�<1 word in the clause, while in ( 1 8) there are two stressed elements it. Boskovi6 argues that the derivation is rescued by a ph(]lnOllogllcal filte(i, which enforces spell-out of the lower copy of the verb rather than the head of the chain. [ne go ¥i6ja+li] ae-ge vidja this way the phonological requirements of both the negative marker ne the interrogative complementizer Ii are met. '

< A similar account of the facts, which also relies on prosodic inversion, is

res€mte:d in Franks ( 1 998) and Rudin et a1. ( 1999). Acc:ordling to Boskovic, syntactic movement is evaluated by a phonological filter. may enforce pronunciation of a lower of copy of a movement chain if the of the head of a chain results in a violation of phonological COIDStraults. See Boskovic (200 I) for an in-depth description of his pronounce-a­ copy account of cliticization in South Slavic.

420 Krzyszto/ Migdalski 4. An alternative account

All the previous approaches to the interaction between Ii and negation indicate that Ii may be preceded by clitics and finite verbs in the presence of ne. (20) a. Ne

go

Ii

iskate want NEG itacc.cL Q 2pl 'Don't you want it?' b. Ne iskate Ii? NEG want2pl Q 'Don't you want to?'

(Hauge 1 999)

However, a major flaw of the prosodic analyses is that they do not take into account the fact that neither the I-participle nor the subject may precede Ii in this context4 • (2 1 ) a. *Ne

kupil

NEG bUYpart.M.Sg

Ii

e

beaux.pres.3sg

Q

b. Ne e Ii kupil knigata? ' Hasn't he bought the book?' (22) a. *Ne

az Ii

NEG I

Q

b. Ne sum Ii az? 'Not rne?'

knigata? book-the (Svetoslav Marinov, p.c.)

(sum)? beaux.1 sg (Svetoslav Marinov, p.c.)

This is quite striking, given that the non-negated variants of (2 1 a) and (22a) are completely grammaticaL

4 The only syntactic account of

Ii placement in the presence of negation that

aware of is due to Tomic ( 1 996), who argues that the excorporates from the clitic cluster and raises

'ne

I

am

+ first clitic' sequence

as a unit in order to

lend support to Ii.

However, she does not mention the way this sequence patterns in the presence of the subject and the I-participle

(ef. 2 1 and 22). She also does not provide a

syntactic motivation for this movement.

Negatedfocus questions in Bulgarian 42 1

(23) a. Kupil

Ii

e

bUYpart.M .sg Q beaux.pres.3sg 'Has he bought the book?' b. Az Ii? I Q 'MeT

knigata? book-the

Likewise, the sentences in (2 1 ) are completely acceptable if a past tense form of a verb is used instead of the I-participle. Moreover, there is no phonological reason why (2 1 ) and (22) should be excluded. The stress patterns that these examples would potentially show should be the same as in the constructions with past tense verbs or clitics (cf. ( 1 6» . In principle, negation may place stress on the I-participle or the subject, so there is no reason why this option should be ruled out. I conclude that the contrasts indicate that the process is constrained $yntactically, rather than prosodically. The impossibility of the subject or the I-participle insertion between negation and Ii is due to the phrasal status pf these elements (cf. Migdalski 2006, where I claim that the I-participle undergoes phrasal movement in Bulgarian and Serbo-Croatian, and that it targets Spec, TP). I propose that the h ighest head available in the structure is attracted by negation and subsequently, the 'negation+head' sequence into Us. Each of the attracted elements has a focus interpretation, whereas Ii is the spell-out of the focus feature. I suggest that elements check focus via adjunction into Ii. Since negation is a head, may only attract XO elements, such as finite verbs and clitics. It does not amract I-participles or subjects, because they are XP's. As an example, I ........ .. ,. .... "" the derivation of ( 1 2b) in (24).

I suggest that clitics and finite verbs are incorporated into negation, because they

a prosodic word together. More evidence for this assumption comes from ilel�DO··L;noaIl· an, which is closely related to Bulgarian. In Serbo-Croatian c1itics always appear in the second position. In view of this requirement, negation the verb vidim in (i) must have incorporated into each other forming a single because they may be followe
(*ga)

NEG himacc.CL

'1 don't see him'

vidim

seepres. lsg

ga himacc.CL (Rivero 1 99 1 :338)

422

KrzysztofMigdalski

(24) a. Ne mu dadoxte? Ii go NEG himdat.cL Q itacc cL gave2pl 'Didn't you give it to him?' b. hp h dadoxte;] [vp [v t; ] mu go ]] c. [TP h muj + gOk + dadoxte;] [vp [v t; ] t.i tk ]] d. [NegP [Neg ne+m�;1 [TP h tj + gOk + dadoxte;] [Vp [v t; ] t- t k ]]] e. [c p [c m + Ii] [NegP [Neg tm] [TP [T tj + gOk + dadoxte;] [vp [V t; ] �i t k ] ]]] .

The derivation proceeds as follows. Since the verb dadoxte is tensed, it must raise to TO in order to check a [Tense] feature (cf. (24b» . Pronominal cHtics must be adjacent to the verb in Bulgarian, which I assume means that they raise from their argument positions within the VP and adjoin to the verb in TO (cf. (24c» . NegP is merged immediately above TP. Negation in NegO attracts the highest head available below it in the structure, which is mu in (24), and incorporates into it. Finally, Ii is merged as the C head. Li i s a spell-out of a Focus feature; besides it is also an enclitic, so it needs a phonologically overt element to its left. The complex head ne+mu raises out of Neg and left-adjoins to Ii in C (cf. (24e» . Thi s results in the focus question Ne mu Ii go dadoxte?

References Boskovic, Zeljko. 200 1 . On the Nature of the Syntax-Phonology Interface: Cliticization and Related Phenomena. Amsterdam : Elsevier. Franks, Steven. 1 998. Clitics in Slavic. Paper presented at Comparative Slavic

Morphosyntax Workshop, Bloomington.

Franks, Steven. 2005 . Another Look at Ii Placement in Bulgarian. Ms. Indiana Univers ity . Franks, Steven, and Tracy Holloway King. 2000. A Handbook of Slavic Clitics:

Oxford Studies in Comparative Syntax. New York: Oxford " University Press. The Word Order of Predicate Clitics in Bulgarian . .Journal

Hauge, Kjetil Ra. 1 999.

o/Slavic Linguistics 7 :9 1 - 1 39 .

Izvorski, Roumyana, Tracy Holloway King, and Catherine Rudin. 1 997. Against Li lowering in Bulgarian. Lingua 1 02: 1 87- 1 94. Migdalski, Krzysztof. 2006. The Syntax of Compound Tenses in Slavic. Ph.D. dissertation. Tilburg University. Rivero, Maria-Luisa. 1 99 1 . Long Head Movement and Negation: Serbo-Croatian vs. Slovak and Czech. The Linguistic Review 8:3 1 9-3 5 1 . Rivero, Maria-Luisa. 1 993 . Bulgarian and Serbo-Croatian Yes-no questions: yo-raising to Ii vs. Ii hopping Linguistic Inquiry 24:567-575. -

.

Negated/oeus questions in Bulgarian 423

Rudin, Catherine. 1 986. Aspects of Bulgarian Syntax: Complementizers and Wh­ Constructions. Columbus: Slavica. Rudin, Catherine, Christina Kramer, Loren Billings, and Matthew Baennan. 1 999. Macedonian and Bulgarian Ii questions: Beyond Syntax. Natural Language and Linguistic Theory 1 7:54 1 -586.

Tomic, Olga. 1 996. The Balkan Slavic Clausal Clitics. Natural Language and Linguistic Theory 1 4: 8 1 1-872.

The case of midpositions Natasa Milicevic

The syntax of adpositions and their lexical vs. functional status in grammar has been a long-standing topic in linguistic research. As shown by Van Riemsdijk ( 1 987, 1990), spatial and iocational adpositions can be successfully treated as members of the same lexical category. Their surface pre/post-positioning with respect to the nominal argument (as is the case in Dutch), stems from the type of projection a P element instantiates, and not the optionality of the head-complement ordering. Thus, the functional p shells share/inherit the categorial features of their lexical complements, which is in accordance with the usual assumptions about phrase structure. The (at least apparently) DP internal (mid)positions, however, can be a challenge in this respect. The status of the midpositions in partitive constructions (a picture of John, a cup of tea) for example, could be thought of as semi-lexical. It is not clear whether the cases of DP internal P should be treated as base-generated inside or outside DP. I Here I will point to another option: a strictly NP internal P. This is the case of midpositions represented in ( l ). The interesting problem they pose has been somewhat neglected in the literature. At first sight they seem to instantiate a fairly simple coordinated structure. I will examine one approach that follows thi s idea. It is Postma's ( 1 995) account of this type of expressions in Dutch. I will try to show that there are several reasons why, at least in the case of English constructions of this type, such a solution is not very rewarding and I will try to sketch out the basic assumptions and gains of an alternative one. ( 1 ) a.

h.

Grammatical Models Group solved problem after problem . He was buying book by book until he collected the whole set.

I Cf. the account offered by Kayne (2002) for the cases of English partitive

French de

of and

The case ofmidpositions 425

The boldfaced strings in ( I ) represent a constituent out of which no movement is possible. This is only one of the properties distinguishing this construction from the one featuring partitive of(a picture ofMary): 2 (2) a. Who was John admiring a picture of t? b. Money John has lots of t. (3) a. * What did they solve problem after t? b. * Problem they solved after problem t. In addition to the impossibility of extraction (2), there are other structural idiosyncrasies to be accounted for: 3

(4) a. No extraction from the DP argument. b. The lexical reduplication is obligatory (the two Ns are identical) c. Both Ns obligatorily occur without determiners ( *a book by book, *the problem after problem) d. Both Ns are obligatorily singular ( *10 solve problems after problems) e. Modification is h ighly restricted ( *old book after book, *book after old book, '!old book after old book) Semantically, the structures in ( 1 ) are peculiar in the following ways:4 The thematic relation between each of the nouns and the verb is identical b. The reference of the construction is plural c. The (mid)position itself seems to contribute the distributive reading to the interpretation of the whole, similar to the quantifiers each or all a.

· As Postma observes, the structure in question (henceforth NpN) exhibits

crosslinguistic productivity. This is supported by the examples in (6) from

examples in (2a,b) are taken from Kayne (2002). ,Postmla also states (4b-d) for Dutch and the generalization in (4e) is stronger since modification seems to be possible in the relevant Dutch examples. (5a,b) are reported by Postma for Dutch and he refers to the quantificational ;;,: TreatdlIllg obtained in these constructions as universal distributive and! or collective . . reading. .

426 Natasa MiliceviC

Dutch « 6a-b), adopted from Postma's original proposal) and Serbo­ Croatian (6c-d). zit raam aan raam een meisje. In de rosse buurt In the red-light district sits window to window a girl ' In the red-l ight district there is a girl at every window.' b. Ik vroeg het deur aan deur. I asked it door to door c. Resavali su problem za problemom. solvedM,pl AUX problemacc after problem instr 'They solved problem after problem. ' d. Jeo je jabuku po jabuku. ateM,sg AUX appleacc by appleacc 'He was eating apple by apple.'

(6) a.

This fact alone is a good argument against treating the NpN as merely idiomatic or 'frozen', without special syntactic significance. Focusing on the properties (5b,c), Postma aims at the theoretical solution for the puzzl ing syntax/semantics interface mechanism, which relates the phrase involving singular Ns and its plural+distributive interpretation. Th is leads him to propose the underlying coordinate structure in (7).

[op DO [pp [NP PROj hand ] [ p, P [NP PROk hand]]]

(7)

He argues that Dutch idiomatic expressions involving the overt coordinator

en (8) fall into the same type of bare Ns coordination. The interpretative differences (the lexical content of the 'n ouns in the NpN is available for interpretation) are, on this account, attributed to , the nature of the P conj unctor, which is, unlike en, a governing conjunctor.

(8) a.

huis en lijk; man en �uis , ', . house and body; man and mouse 'everybody' b. [DP D' DO [&P [NP PROj man ] [&' & [NP PROk muis]]]

There are several reasons, however, to reconsider the offered solution. One lies in the 'productivity of NpN' argument itself. Namely, it seems that it is not without significance that the NpN structure is the productive one, unlike the N en N found in Dutch. It could be the first indication that what resulted in the N en N limited stock of expressions in Dutch is the result of some

The case ofmidpositions 427

sort of mimicking process, based on the output of a truly productive, and structurally different NpN. With this in mind the same argument could be used for drawing a distinction between the coordination and NpN. Second, it is an attractive and plausible idea that quantification is configurational, which would relieve the lexicon, and consequently syntax, of the redundant feature specification. Although Postma offers arguments against treating these adpositions as lexically distributive, it is not clear why it should be less stipulative to assume special, governing conjunctors. Another argument can be made against the type of projection in which a DP is the extended projection of a P-coordinate structure given in (7). It is difficult to see how this type of projection complies with the Economy of Representation, or any similar principles proposed in other theories to account for the lexical vs. functional projection relation within a phrase. According to the Categorial Identity Thesis of Van Riemsdijk ( 1 990), the proposed underlying structure [OP [PP]] is not the most plausible one. To treat the PP phrase as a special coordination phrase is not enough since, . such a phrase typically coordinates maximal projections, which is not the case here. From the syntactic point of view the projection of the OP shell in this context is highly unlikely and empirically unsupported. The DP is never . realized overtly in any of the languages mentioned above. The postulation of DP, therefore needs more justification, or should be abandoned in favor of a theory that captures the complementary distribution of the D head and . the midposition in NpN. Another empirical point can be made with regard to the restricted modification stated in (4c). The coordination account is based on a much stronger claim, which entirely excludes th� possibility of modifying either of the conjuncts. This, however, is not true for English (9a) or Serbo: Croatian (9b): a. Grammatical Models solved problem after damn/ancient problem . b. Resavali su prokleti/prastari problem za problemom .s solvedM.pl AUX damn/ancient prohlemacc after problemins,f 'They were solving problem after damn/ancient problem.' This alone presents problems for the theory referred to so far, since the conjunct should be a bare nominal in order to be considered part of

:; .��eC(md . .5

The reason why adjectives in the Serbian example occur to the left orN I exceeds my present intentions and present an interesting and wide research topic.

428 NataSa Milicev;c

ZS (zero-semantics). The relevant interpretative properties of such an N can only be derived at the D level):

[DP PROkQ DO [pp [NP PROj a hand ]

( 1 0)

[pI p o in [NP tk

11

hand]]]

The (predication) relation between the bare nominal (obligatorily stripped of its morho-syntactic properties) and PRO in (7) and ( 1 0) is the on e between a lexical item and an abstract object it denotes. PRO is a default interpretable slot responsible for the assignment of lexical meaning to the inserted lexeme. If PRO moves out of the domain of Lexical Closure, the assignment is not possible. On Postma's account, movement of the second PRO is required for full interpretation. PROj is co-indexed with the chain created by movement, and this is the syntactic configuration within the DP domain that allows/leads to simultaneous interpretation of PRO elements.6 The problem is that in English, the second N, which has not bee n assigned lexical meaning is the one that can be modified. Furthermore, the fact that modification of the first N is dispreferred or completely ungrammatical points to the conclusion that the ellipsis within the coordinate structure (out of its first conjunct) is not the right solution for the problem either? The pattern we get . is consistently reverse from the expected eliding strategy within NPIDP: She was carefully rearranging Cdamn/ancientl*oldl*black and white) picture after (damnJancientPold/*black-and-white picture on the shelf. b. She was carefully rearranging picture after damn ancient picture of John. c. *She was carefully rearranging damn picture of John after picture

( l l ) a.

As ( 1 1 b) exemplifies, the occurrence of an adjective can facilitate further modification of the second noun, which strongly defies the idea of the rigid 6 Postma points out that in the clausal domain, simultaneous interpretation

is

characteristic of parasitic gap constructions, negative concord, and the list-reading of mUltiple WH elements.

7

The reduction of the coordinate structure typically involves ellision of

material in the second conjunct, as in (ib). Example (ia) is ungrammatical on reading were

(i)

a.

b.

sad refers both to mother and/ather.

His sad mother and father were silently passing by. *His mother and sad father were silently passing by

the the

The case ofmidpositions 429

modification restriction and the kind of lexical insertion assumed in the theory above. The constraints observed in ( 1 1 a) on the choice of a particular type of adjective stem from the semantic/pragmatic reasons, which can be observed also in ( 1 2): ( 1 2) a. ??He bought a picture after buying another old/black and white picture. b. He bought a damn black and white picture after buying another one of the same kind. The reason for the extreme oddity of ( 12a) is a discourse -related phenomenon. Example ( 1 2b) shows that the modification of the NP in the adjunct clause, such that it refers back to the object already present in the discourse (of the same kind), is possible. It seems that the semantic mismatch has to do with the type of adjectives used to refer backwards to . the (properties of) another member of the same set of entities. In the first attempt to characterize the type of infelicitous adjectives with the more permissive, second N in ( 1 1 ) or (1 2), we can turn to the fmdings of the cartographic approach to the licensing of adjectives. Following Cinque ( 1 994) and, in particular, Scott (2002), we will classify them as 'comment' adjectives. They express the attitude of the speaker and not the inherent properties of the modified noun. The degraded status of old in comparison with ancient is still curious. It should be noted at this point that the direct correspondence between the semantic content of the . ·adjective and its syntactic slot (the right feature content of the related functional head) is difficult to achieve. This is, for instance, what divides adjective pairs ancient and old in Scott's account of the adjective ordering English. Although they apparently fall into the same age class, their syntactic behav ior indicates a distinction (for details see Scott 2002). It prompts the proponents of the fixed ordering theory to assume that sometimes the same, or similar, lexical representations of adjectives (just like some adverbs in the clausal domain) can inhabit different functional projections. This idea relativizes the notion of 'ordering' to a large extent, : :artd leaves the possibility open that many adjectives denoting physical ; ;, (objective) properties can at the same time be subjective/comment-like. . In the example of ancient vs. old in the context in (1 1 ), ancient is the ;more emphatic one, and it intuitively qualifies as the comment-type more ,, : �than the objective age type of adjective. The same difference holds of many ; other strong/weak pairs of adjectives (gorgeous / handsome, amazing /

430 NataSa Milicevic

good). In each case the strong one is the one more acceptable in the NpN. Therefore, we can maintain the initial insights about their meaning: they are comment-like modifiers. Whatever their crucial semantic property is, We make the choice of modifiers in NpN avoiding the objective, lite ral meanings, or reference to unambiguously defining semantic features of objects. This conclusion will suffice since the restriction on N modification observed does not seem to follow from any theory-internal constraints on the size of the NP. The peeled off functional projections would be by definition the higher ones, which is not the case. In the Cinquean representation of the DP structure comment adjectives reside high in the space between the head and the DP. However, the infelicitous ones in NpN are the low ones. Recalling what the examples in ( 1 2) shows we will conclude that the adjective selection is restricted independently, and that adjunction to NP is allowed. Finally, the need to account for the fact that the adjectives generated with the N in the second conjunct are interpreted as modifying the meaning of the whole (both Ns) leads us to the alternative proposal of the NpN structures, and that is the SC analysis. To be more precise, the following structure can be related to SC structure featuring complements of verbs, given in (13):

( 13) a. I considered him as a fool. b. She took him for a serious person. The boldfaced strings are also sm�l1-clause constituents in which the prepositional elements (as and/or) head a functional· projection (cf. Starke 1 995), which hosts the object adjecent to the verb in its Spec and a nominal in its complement position. As and/or belong to the semi-lexical subset of adpositions, which are selected for by the dominating lexical verb of the type in (13). The sketch of the related NpN structure is given. in (14): ( 14)

[NP N

[FP problem

[FO after] [NP damn [NP problem]]]]

The property (Sa) follows from the given structure. The verb in (1 ) assigns a unique role to the NP argument that entirely contains another NP (another instantiation of the same entity set). The observed formal properties of this NP can also be more successfully captured on this approach. Both properties (4b) and (4c) can quite plausibly faU out of the nature of FP. It is obviously in the complementary distribution with the higher functional

The case ofmidpositions 43 1 projection, which hosts determiner and quantifier-like elements). A plausible assumption is that the embedded FP does the job of the D° earlier in the derivation, and the D° is never instantiated. In other words, the interpretative properties on an NP encoded by DO can be expressed through the embedded predication. One can assume that midpositions are inherently quantificational/ distributive (and in that sense in complementary distribution with DO). Whether such a stipulation is attractive and eventually justifiable or not does not depend on the basic structure itself. It would be also interesting to derive this special semantic property of midpositions from the same cognitive source we appeal to when describing the meaning of lexical adpositional heads. In other words, the way adpositions in their lexical guise relate concrete objects, can in a way give us a clue of why some of them are good candidates for the NpN construction. Maybe the thematic asymmetry (Figure/Ground relation) we encode syntactically is blurring the basic (inherent) meaning (type of relation) rendered by the adposition. After all, the function and meaning of adpositions in the sentential context depends on the properties of the event they modify. The Figure/Ground distinction between the two arguments of the lexical adpositions is dependent on their structural environment and the lexical properties of other elements in that environment (the notion of movement encoded by the verb, for example), as well as extra-linguistic factors. Assuming this lexical ·

underdetermination of after or by could help us understand their functional role in another, DP or NP domain. We may think of them as capable of

defining the semantic type of set-intersections within this domain. We may

start from an assumption that after and by denote a complete overlap of the two sets, in the steps of ordering its two atomic parts, temporally or .. . .. spatially (the ordering fashion remaining idiosyncratic to each of them). If we can determine the set-membership of a noun, indicating the complete

· overlap of two sets through a predication relation of an SC type (instead of using· articles or quantifiers), wouldn't the reduplication (identical Ns .. occurring in the terminal string of NpN) be the only possible choice for the lexical insertion? These are the deep questions far beyond the scope of this ·

. ;paper, but still relevant. The point of these few highly speculative suggestions is to show it is not necessary to abandon Postma's idea of the simplest and most parsimonious theory of the syntax/semantics interface in . order to accept the analysis proposed here. The SC analyses have been proposed for the nominal domain before. Bennis et al. ( 1 998), for example, use it to account for another type of

432 NataSa Milicevic

midposition. This is the Dutch een N van een N construction. The underlying structure they offer for the example ( l Sa) is given in (ISb): (IS) a. een beer van een man 'a bear of a man' b. [DP [FP beerj [F ' [FO van+een] [xp man [X' Ii Ij ]]]]] The incorporation of the spurious article (head of XP) into the higher head FO leads to the inverted order of the two nominals. They relate the derivation of the surface order of nominals in this case to the predicate inversion in the clausal domain. The same type of formative (small clause) had been used by Stowell ( 1 991) to account for this phenomenon. The function of van is of the same type as the function of F in a clause. It extends the functional domain of the predicate and accomodates the inversion of arguments. If we compare the function of the midposition in NpN ( 1 4) with the function of van in (15) we will see that the NP small clause head of ( 1 4) does not have this potentiaL The true correlate of ( 1 4) in the clausal domain is indeed ( 1 3) where the functor (as,/or) is lexically selected by the (non-copular) verb and where no inversion is possible. This is the point where the real work on all the correlations and implications for the two domains (clausal and nominal) only begins. The suggested account for the midpositions of the NpN seems to be a promising starting point for the considerations of many issues I have only touched upon. The full account of the DPINP internal P e lements (midpositions) will be a significant contribution to the understanding of the category of adpositions, and revealing with respect to the properties of the domain s they operate in. .

References '

Bennis, Hans, Norbert Corver, and Marcel den' Dikk¢n.' 1998� , Predication in nominal phrases The Journ'al ojComparative Gehnanic Li�gubftics, 1: 85-117, Kluwer Academic Publishers. Cinque, Guglielmo. 1994. On the Evidence for Partial N-Movement in the Romance DP in G. Cinque, J. Koster, J.-Y. Pollock, L. Rizzi and R. Zanuttini (eds.) Paths towards Universal Grammar: Studies in Honor of Richard S. Kayne, 85-110. Wash ington, D.C.: Georgetown University Press. Kayne, Richard S . 2002. On Some Prepositions That Look DP-Internal English of and French de Catalan Journal of Linguistics, 1, 71-115. Postma, Gertjan. 1995b. Zero Semantics: A Study of the Syntactic Conception of Quantificational Meaning, Phd Dissertation, University of Leiden, HIL.

The case 0lmidpositions 433 Riemsdijk, Henk C. van. Foris.

1987. A Case Study in Syntactic Markedness.

Dordrecht:

Riemsdijk, Henk C. van, 1990. Fuctional Prepositions in H. Pinkster and l. Genee (eds.) Unity in Diversity, 229-241. Dordrecht: Foris. Sco tt, G.-J. 2002. Stacked Adjectival Modification and the Structure of Nominal

Phrases in G.Cinque (ed.) Functional Structure in DP and IP (Cartography of Syntactic Structures, Oxford Studi es in Comparative Syntax). Oxford University Press. Starke, Michal. 1995. On the Format for Small Clauses In A. Cardinaletti, M.T. Guasti (eds.) Small Clauses (Syntax and Semantics 28).237-269 Stowell, Tim. 1 991. Determiners in NP and DP, in K. Leffel and D. Bouchard (eds.),

Views on Phrase Structure 1,37-56. Kluwer Academic Publishers.

Quechua P-soup Pieter Muysken

One of the first papers I read when arriving in Amsterdam in 1972 to study linguistics was Henk van Riemsdijk's Dutch P-soup, an initial attempt to decypher the bewildering, mysterious, and seemingly contradictory patterns and arrays of adpositions and particles in Dutch. The paper has always stuck in my memory because it showed how a theoretical analysis could help clarify a wealth of chaotic observations. This topic remained at the centre of Henk' s attention, resulting of course in the 1978 book, one of the classics of the generative syntax of Dutch (Van Riemsdijk 1978). Together with verb clusters, verb second, and the mysterious element er, Dutch adpositions have remained one of the striking contributions of Dutch to the encyclopedia of odd but important grammatical facts in the languages of the world that Henk knows and loves. His current work with Martin Everaert on the syntax companion can be seen as an attempt to saveguard this treasure house for future generations of linguists. Most of what I know about syntax comes from classes and discussions with Henk, and I was fortunate to later collaborate with him on the theory of categories. Long before the Principles-and-Parameters model came in vogue in the early 1980s, Henk had already drawn attention to, and underlined the importance of, cross-linguistic research. In this spirit I offer the following thoughts about Quechua adpositions. Current work on various Amazonian languages shows that it is important to separate the lexical notions of noun and verb from the grammatical notions of argument and predicate 01an Gijn2005, Haude 2005), and hence the need to distinguish between syntactic and lexical categories. However, the same point can be made regarding Quechua adpositions, the topic of this squib: lexical and syntactic categories do not necessarily match. To put it bluntly, the syntactic category in Quechua of head of a PP, an adpositional phrase, desperately seeks lexical instantiations. While Dutch has a rich inventory of adpositions (see e.g. Zwarts 1997), Quechua does not have any. This does not mean that there are no PPs, just that these are not headed by elements from a corresponding

Quechua P-soup 435 lexical category. In the Quechua of Ecuador (Muysken 1 977), we have the following types of PPs: I (1)

[Pedro-ta yalli] yacha-n-mi. exceed know 3_AF Peteracc 'S/he knows more than Pedro.'

(2)

[Misa khipa-ta] shamu-nga-mi. Mass backacc come3.FUT-AFF 'Slhe will come after Mass.'

(3)

[kan-rayku] shamu-n-mi. you-because COme3-AFF 'S/he comes because of you.'

In ( 1) the head of the adpositional phrase yalli 'exceed' is a verb. This lexical item normally appears with person and (optionally Tense Mood Aspect) marking. We know that it is a verb because it assigns accusative case, a property exclusive of verbs. However, verbs ordinarily never occur without some kind of inflection. The structure in (1) resembles that of serial verb constructions in Niger Congo and Caribbean Creole languages. However, the phrase [Pedro-La yalliJ has the PP property of being freely movable to other positions in the clause in Quechua, something which does not hold for verb phrases. In (2), the head khipa 'back' is a noun. We know this because it optionally takes accusative case, structurally assigned to elements (objects, manner and time adverbial phrases, secondary predicates in small clauses) in the Quechua verb phrase. Case can only appear on morpho-lexically [+N] elements in Quechua. Accusative is generally present with direct objects, and less frequently on time and manner expressions. Again, the PP [Misa khipaJ can be moved in the clause, but it cannot be split up. Even though it is a noun, khipa takes a complement, something which ordinary nouns do not do in Quechua. , In (3), finally, -rayku 'because of is traditionally viewed as an oblique :case affix. There are mono-syllabic and bi-syllabic case affixes in Quechua:

. 1

Abbreviations: ace: accusative case; AFF: postmanteau form for third person future tense.

affIrmative evidential;

3. FUT:

436

Pieter Muysken

(4) a. -ta -man -wan -pi -paq b. -manta -kama -rayku

accusative dative / directional comitative / instrumental locative benefactive / genitive ablative until because of

It is tempting to treat the elements in (4a) as case markers and those in (4b) as (morphologically dependent) postpositions. Three things would speak for this: first, lexical roots tend to be poly-syllabic in Quechua (the handful of exceptions include the verbs ku- 'give', ni- 'say', and ri- 'go', and the noun nan 'path, road'), and grammatical elements mono-syllabic. Second, the elements in (4a) tend to have a more grammatical, abstract meaning than those in (4b). Third, in relative clauses where the relativized element is not the subject of the relative clause, the case markers in (4a) can generally be dropped, yielding phrases like [the house that I lived] or [the knife that I cut]. This does not hold for (4b). We may assume that a grammatical element like case can be left out, but not a lexical postposition. Nonetheless, there are good reasons not to take this road. First, the relation between poly-syllabicity and semantic concreteness is only partial. The element -manta is easily as abstract in meaning as the element -paq, for instance. Furthermore, -paq and comitative -wan cannot be omitted in relative clauses: [the girl I bought for book *(for)] and [the man I want to the movies *(with)]. Third, -manta se�ms composed of -man and -tao In other Quechua languages we also have -pi-ta and -paq-ta. This suggests that at least -manta is best analyzed as a complex case marker rather than a postposition. Finally, it· is odd that a lexical· class should be morphologically dependent in Quechua. It could· be.· thought that. the postpositions in Quechua are clitics,but c1itics inthis:language generally show a much more free behavior. . Altogether, my own inclination at this point, in the absence of other arguments, is to treat the elements in (4a) and (4b) as all belonging to the same class of case markers, with gradual rather than absolute differences between some of the elements in the two subsets. For (3) this would imply that the PP has a lexically empty head licensed by the presence of the case marker. What to make of all this? In Lefebvre and Muysken (1988) we developed an analysis in which: (a) the features [±N] and [:tV] where .

Quechua P-soup 437

weighed in the sense that a + value was infonnative, and a - value was not; (b) syntactic projections could be less infonnative, more neutral, than their lexical heads, but not the reverse; (c) folJowing lackendoff, N = [+N,-V], V = [-N,+V], P = [-N,-V], and A = [+N,+V]. These 'assumptions were productive in allowing us to explain the complex distributions of genitive and accusative case markers in Quechua nominalized clauses. In addition, we got the result for free that PPs in Quechua (recall, [-N,-V]) could be headed by either nouns ([+N,-V]) or verbs ([-N,+V]). However, the analysis also had some drawbacks. First of all, it predicts that adjectives ([+N,+V]) can head both noun ([+N,-V] and verb ([-N,+V]) phrases. This is correct for noun phrases, where we have expressions such as [the good (one)], but not for verb phrases. Adjectives and verbs have clearly distinct distribution. It also predicts correctly that verbs ([-N,+V ] ) cannot head adjective ([+N,+V]) phrases; however, it rules out the possibility that nouns ([+N,-V]) head an adjective ([+N,+V]) phrase, which on the whole is perfectly acceptable: (5)

rumi wasi stone house 'a stone house'

Given the limited empirical succes of the informativity loss model (in spite of its conceptual attraction in that it only requires feature infonnation where needed in the lower reaches of syntactic trees), I want to tentatively propose a model in which the syntactic structures are lexically realized by elements that may or may not have an identical categorial signature. The matching between lexicon and syntac may well be semantic rather than grammatical or morphological: rumi 'stone' can function as an adjective because of its '. descriptive qualities; likewise khipa 'back' can function as a postposition because of the metaphorical extension of its meaning; finally yalli 'exceed' can occur as a postposition in a comparative, as in other languages, because of its rather abstract meaning. Syntactic categories are lexically instantiated to different degrees in different languages. Well-known is the case of adjectives, which are absent in some languages, or limited to six or seven lexical items. Similarly, the nature and richness of the lexical category P may differ widely. Like Dutch, Quechua has PPs, but unlike Dutch, it may not have Ps.

438

Pieter Muysken

References Gijn, Rik van.

2005. A grammar of Yuracare.

Unpublished doctoral dissertation,

Radboud University N ijmegen

Haude, Katharina. 2005. A grammar Radboud University Nijmegen

of Movima. Unpublished doctoral

Lefebvre, C laire and Pieter Muysken. 1988.

Quechua.

dissertation,

Mixed Categories. Nominalizations in

Dordrecht: K luwer.

Muysken, Pieter.

1 977.

Syntactic developments in the verb phrase of Ecuadorian

Quechua. Dordrecht: Foris. Riemsdijk, Henk C. van.

1 978.

A case study in syntactic markedness. The binding

nature of prepositional phrases. Dordrecht: Foris.

1997. 35, 1 091-11 1 1 .

Zwarts, Joost.

Complex prepositions and P-stranding in Dutch.

Linguistics

Sem antic compositionality of the way­ construction Heizo Nakajima I. Introduction

l

The so-called way-construction, as instantiated by such a sentence as Jack plowed his way through the mud, has been widely considered as a sort of 'constructional idiom ', the meaning of which cannot be fully predicted on the basis of the meanings of its constituent parts. lackendoff (1990, '1992), for example, has proposed the construction-particular correspondence rule. Goldberg (1995) has linked the construction directly to 'constructional meanings'. Culicover and lackendoff (2005) have also touched on the construction, stating that "[t]here is no way to predict the meanings of [the constrictions] from the words". These constructional approaches maintain that the meaning of the way-construction cannot be determined by the Fregean principle of compositionality. This squib attempts to argue, contrary to the constructional approaches, that the meaning of the way-construction can be compositionally determined, and therefore it is not necessary to treat the construction exceptionally as a constructional idiom. 2.

Meaning of Way

The syntactic form of the way-construction can be schematically ,'represented as in ( 1 ), where PP is of path or tenninus, and will be referred to as Path-PP: NP - V - one's way - PP The core meaning of the construction, more specifically of its VP, is, according to the OED, "to effect a forward movement by the action denoted by the verb", or "to accompany one's advance by the specified action". Goldberg (1995) simplifies the meaning of the predicate in the way­ construction as CREATE-MOVE, which she attributes to the predicate '1

•

I am thankful to Karen Courtenay and Ken-ich Takami for their insightful comments and suggestions.

440 Heizo Nakajima

meanings of the two sub-constructions that are amalgamated into the way­ construction. One sub-construction is the creation one with the two arguments of a creator and a 'createe-way' , and has the meaning of the creation of a path� the other is the intransitive motion sub-construction with the two arguments of a mover and a path, and has the meaning of the movement along the path. The presence of the path-PP in ( 1 ) is essential to gain the sense of MOVE in the construction meaning. Note, in addition, that the noun way in ( 1 ) is taken as representing an 'effected entity' created as a result of the activity denoted by a verb. The presence of a path-PP, however, is not indispensable to the construction meaning of the way-construction (briefly, the way­ construction meaning). The examples in (2) do not involve a path-PP; yet, they have the way-construction meaning of CREATE-MOVE, or at least of MOVE, which is similarly observed in the way-constructions in (3): (2) a. They fought their way bravely. (KDEC) b. He made his way on foot. (KDEC) c. He had previously pedalled his way to make the first cycle crossings of the High Atlas mountains in Morocco, (BNC) d. But it quickly hushes up as another group gropes its way, begging (Kuno and Takami 2004) for assistance. ' "

(3) a.

At the station they fought their way through the crush of travellers (bnc) b. I turned and hurriedly made my way downstairs. (BNC) c. There's no shortage of activity, as you pedal your way to power in "Electrifying Leeds", (BNC) (BNC) d. We grope our way to the back door and up the stairs . . . .

_

.

.

Since the examples in (2), though lacking a path-PP, have the meaning of movement, it must be supposed that the sense of MOV� in the way­ construction meaning does not stem from the combination of a verb with a path-PP, or in Goldberg' s ( 1 995) analysis, from the sub-construction consisting of a mover and a path. The noun way, apart from that in the way-construction, expresses the sense of movement or advance. The examples in (4) are obviously not the way-construction, but the noun way in them expresses the sense of movement or advance (though its use is part of nautical vocabulary in some examples and somehow archaic in some others):

Semantic compositionality ofthe way-construction 44 1

(4) a. Our boat did not make any way because of the strong wind. (KDEC) (KDEC) b. The ship gradually gathered way. (KDEC) c. The plan has made no way. d. . . but the general revival of the old forms has made no way in those countries . . . e. The Minister of State must reassure the pressure groups. He would go some way to doing so if he told the House tonight that . . . (BNC) .

We can assume, then, that the noun way in the way-construction does not denote a physical or metaphorical path that is created by the activity represented by a verb, but the event of movement or advance. The possessive noun in the noun phrase one's way is the subject of that event, which is identical with the subject of the verb in the construction. The possessive noun in one's way is necessarily bound by the subject of the verb. 3.

Semantic Decomposition of Verbs in the Way-Construction

One class of verbs possible in the way-construction is that of creation verbs, such as build, burrow, cut, dig, hew, make, paint, plow, shovel, tunnel, write, and so on. The object of the creation verbs is a 'result object' (see Jespersen 1 9 49). In desperation, some of the men tried to dig their way through to Stourbridge to get coal for heating. (BNC) b. . . . the government hopes that Japan can build its way out of the (BNC) recessIOn. He painted his way to an international reputation. (KDEe) c.

(5) a.

. The creation verbs share the sense of MAKE, and in addition to this, represe nt the means for creation. Thus, the meanings of the creation verbs can be decomposed into the component of MAKE and that of means. We will not, in this short squib, go into the issue of how the two semantic components are to be represented in a syntactic structure. It is sufficient to note that the meaning of the creation verbs consists of the components of MAKE and of means. A result object semantically serves as the object of ' the component of MAKE:

442 Heizo Nakajima

(6) a. dig a hole = make a hole by digging b. build a house = make a house by building c. paint a picture = make a picture by painting The noun way in (5), which denotes 'movement' or 'advance', is the object of the creation verbs, and being a result object, semantically works as the object of the component MAKE in the meanings of the verbs:

(7) a. dig one's way = MAKE a movement by digging b. build one's way = MAKE a movement by building As is clear from (7), the sense of MOVE in the way-construction meaning comes from the combination of the semantic component of MAKE in the verbs with the meaning of the object way, namely 'movement', and that of the means originates from the other component of the verbs, such as 'by digging'. The sense of MOVE compositionally arises as a result of the merger of the verbs and their objects. It is not necessary, therefore, to stipulate the meaning of MOVE or GO as a part of the fixed way­ construction meaning. Moreover, the compositionality approach does not overlook the other important semantic aspect of the construction, namely, that of means. Another class of verbs that can appear in the way-construction is that of verbs of manner of movement. This class includes verbs like plod. sashay, snake, speed, stumble, thread, wind, worm, wriggle, zigzag, etc.: . .. their column of official jeeps snaked its way down into the (BNC) Rante. b. The spacecraft is speeding its way into outer space� (KDEC) c. I wormed my way along the roadside ditch. (BNC)

(8) a.

The verbs of manner of mqvement share the s€mse .�fi:MOVE, and' in addition to that� represent the meaning of the mann�r.· of the movement. Their meanings are decomposed into the component of MOVE and that of manner:

(9) a. snake = MOVE in a snakelike manner b. speed = MOVE in a speedy maIUler The merger of the intransitive verbs of manner of movement with the noun way readily forms the association of the semantic component of MOVE in

Semantic compositionality ofthe way-construction 443

the verbs with the meaning of way, 'movement', because they are cognate. The collocation of the two cognate elements, 'MOVE' and 'a movement' is, however, semantically redundant, so they fuse into the simple meaning common to them, 'move'. (lO) a. snake one's way = MOVE a movement (= move) in a snakelike manner b. speed one's way = MOVE a movement (= move) in a speedy manner Thus, the way-constructions with verbs of manner of movement as in (8) have the meaning of movement in some manner. Verbs of nondirected motion in the sense of Levin (1993) range over a variety of subgroups. What is common to the subgroups of these verbs is that the verbs, when combined with a directional or path phrase, express not just the meaning of motion, but that of movement. See Pustejovsky (1995) for details. Therefore, the way-constructions with nondirected motion verbs and a path-PP obtain the sense of MOVE in a way similar to (10). The verbs in the examples in (11) extend over various subgroups of nondirected motion verbs, including that of Run verb (11a, b), sound " emission (11 c), Push verb (11d), and Waltz verb (11e):

b. c. d. e.

Even if she fell in, Gazzer told himself, she could probably swim (BNC) her way out. He can also slowly inch his way up the stairwell at a rate of 1 foot per round. (BNC) Last year in London's Kilburn National he gargled his way through (BNC) an entire set with a bad throat infection. (BNC) The demonstrators pushed their way in and sat down . . . (BNC) We sang and danced our way through a motley repertoire.

"What is noteworthy here is that examples like (lIe-e) express neither a :means of movement nor its manner, contrary to Goldberg (1995). In ',(lIe-e), the movement is made neither by gargling! pushing (something)! 'singing and dancing, nor in a gargling! pushing! singing and dancing ':'manner. The movement is accompanied by the activities originally denoted by the verbs. Example (11 c), for instance, means that he moved forward iwhile gargling. The way-construction allows for the occurrence of Gain verbs (including verbs of deprivation), such as buy, earn, feel, fight, find. grope,

"'�I "

444 Heizo Nakajima

pay, win; cheat. con, wangle, etc., too. The verbs in this class denote accomplishment of some effect by the activities represented by the verbs. In the way-constructions in (12), the accomplished effect is advance: (12) a. Japanese bought their way to excessive influence in Washington by hiring regiments of lobbyists (BNe) b. They had fought their way to the top, often against strong (BNC) competition. (BNC) c. " .he thought he could cheat his way to victory ... The Gain verbs have two semantic aspects; one is the sense of GET, and the other is that of the means used to accomplish the effect of 'advance' (rather than 'movement' in the neutral sense, in the light of the positive meaning of GET). The semantic component of GET in the verbs is combined with the sense of the noun way, 'advance', as in (13): (13)a. buy one's way GET an advance by using money b. fight one's way GET an advance by fighting =

=

The last class of verbs to be considered here is that of verbs of bodily process (including verbs of verbal or non-verbal expression), such as belch, cough,joke, laugh, puff, sing, shout, sneeze, and so on. (14) a. They coughed their way through All Things Bright and Beautiful. (BNC) b. . ..the minister of Dunlopjoked his way around the world at various functions. (BNC) The verbs in (14) denote simple activities, and do not seem to involve such basic semantic components as MAKE, MOVE, or GET, which serve to semantically license the noun way in the way-construction..: Notice, however, that' aU' of· these verb's' are 'unergative ones, particularly those converted from nouns. Hale and Keyser (2002) have proposed that unergative verbs, especially those converted from nouns, consist of two elements, a root and an abstract verbal nucleus. The verb cough, for example, is composed of the root component cough and the abstract verbal nucleus. The verbal nucleus component of an unergative verb can be supposed to have a meaning something like MAKE, considering the fact that the unergative verb cough, for instance, is equivalent to 'make a cough'. Therefore, the unergative verbs of bodily '

Semantic compositionality of the way-construction 445

process are somehow similar to the verbs of creation in that they involve a meaning close to MAKE. These verbs are also sim ilar to some verbs of nondirected motion, such as those in ( 1 1 c-e), in that the root component expresses neither means nor manner, hut rather accompaniment: ( 1 5) a. cough one's way = v (=r MAKE) a movement while coughing b. smile one's way = v (=r MAKE) a movement while smiling The proposed analysis assumes that an unergative verb in the way­ construction always needs the verbal nucleus so as to produce, together with way, the meaning of 'make a movement'. It can be predicted then that a root element which usually does not appear as a verb by itself can occur as a verb in the way-construction, because the construction always requires the verbal nucleus which converts the root into a verb. The word amber, for example, usually does not appear as a verb, but can occur as a verb in the way-construction: ( 1 6)

As he reached the comer of Fifty-ninth Street the traffic light ambered its way to red and he stopped with the impatient crowd. (DEWGN)

An apparently troublesome problem with the verbs of bodily process is that without way, thnnese verbs cannot occur with a path-PP (Jackendoff 1990, 1 992): . ( 1 7) a. B ill belched *(his way) out of the restaurant. b. Harry moaned *(his way) down the road. c. Sam joked *(his way) into the meeting. This fact would be problematic if it were assumed that the meaning of 'move' in the way-construction was a consequence of the combination of the verbs with the path-PPs, just as the meaning of 'move' is derived as a result of the combination of a nondirected motion verb with a path-PP (e.g., . The boat floated into the bay). But, the fact illustrated by ( 1 7) is not a serious problem for the proposed analysis, in which the meaning of 'move' . . in the way-construction arises as a consequence of the semantic . . . of the abstract verbal component in the verb with the noun way, and with this already formed combination, a path-PP is to be merged ,subsequently.

446 Heizo Nakajima 4.

Conclusion

In this squib, I have argued that the semantic interpretation of the way­ construction, which is widely considered as a 'constructional idiom', is determined compositionally. It has also been shown that the verbs occurring in the construction are process verbs whose meanings are decomposable into the component of MAKE, MOVE, or GET, and the component representing means, manner, or accompaniment. The latter component is relevant to the nuances the way-construction is often claimed to have, such as difficulty, unusual manner or accompaniment, and so on. It is worthwhile to re-examine whether 'constructional idioms' are actually idiomatic. References Culicover, Peter W., and Ray Jackendoff. 2005. Simpler syntax. Oxford: Oxford University Press. Goldberg, Adele. 1995. Constructions: A construction grammar approach to argument structure. Chicago: University of Chicago Press. Hale, Ken, and Samuel Jay Keyser. 2002. Prolegomenon to a theory of argument structure. Cambridge, Mass.: MIT Press. lackendoff, Ray S. 1990. Semantic structures. Cambridge, Mass.: MIT Press. Jackendoff, Ray S. t 992. Babe Ruth homered his way into the hearts of America. In Syntax and the lexicon, ed. by Tim Stowell and Eric Wehrli, 155-178. (Syntax and Semantics 26.) San Diego: Academic Press. Jespersen, Otto. 1946. A modern English grammar on historical principles, Part 6. London: George Allen and Unwin. Kuno, Susumu, and Ken-ichi Takami. 2004. Functional constraints in grammar: On the unergative-unaccusative distinction. Amsterdam : John Benjamins. Levin, Beth. 1993. English verb classes and alternations: A preliminary investigation. Chicago: University of Chicago Press. Psutejovsky, James. 1995. The generative lexicon. Cambridge, Mass.: MIT Press.

Corpus DEWGN Dictionary of English word grammar on nouns. Tokyo: Kenkyusha. KDEC Kenkyusha dictionary of English collocation. Tokyo: Kenkyusha. BNC British national corpus =

=

=

Soft mutation at the interface Ad Neeleman 1.

Introductionl

In Welsh, lexical items may have a default form, as well as a variant with a different initial consonant. The latter are used after certain triggering elements, for instance the preposition i in ( 1 ) (all Welsh data are taken from Borsley 1999). The alternations involved in this process of 'soft mutation' are given below in (2). (1)

(2)

Aeth Megan i Fangor. (*Bangor) went Megan to Bangor P t c

�b �d (lk]) -4 g

-4 f ([v]) m 11 ([I]) �1 rh ([th]) -4 r

b � f ([v]) d � dd ([(0]) g�0

The main problem raised by soft mutation concerns the nature of the alternation. A phonological analysis is implausible, partly because not all the changes in (2) can be produced by natural phonological rules, and partly because the phenomenon is conditioned by non-phonological information. Targets include nouns and verbs, but not prepositions. Triggers include feminine, but not masculine, nouns, and the predicative marker yn, but not the homophonous locational preposition (see Harlow 1989 for more . discussion). A syntactic analysis is equally implausible, I believe. This is not because the· rule's structural description cannot be phrased in syntactic terms. On the contrary, Borsley and Tallerman ( 1 998) show that good 1 Henk van Riemsdijk was the main organiser of the conference at which

I gave my

first talk, he was a member of my dissertation committee, and he talked Peter

Ackema and me out of publishing a book with a particular finn that shall remain

!lameless. For these and many other things, I'm truly grateful. I don't

think that his

positive role in my linguistic life is a coincidence: it is an experience shared by many Dutch linguists of many generations. . guys is an understatement.

To

say that

Henk

is one of the good

448 Ad Neeleman

empirical coverage can be achieved by saying that a constituent bears soft mutation if immediately preceded by either a lexical trigger or by a phrase which c-commands it (see Borsley 1999 for a refinement of this analysis). This analysis must assume, however, that soft mutation is a feature of syntactic constituents that is phonologically expressed through their initi al consonant. The problem is that this feature is sui generis: thus far, attempts to identifY it with an independently motivated syntactic feature have failed (see, for instance, Tal1erman' s forthcoming review of Roberts 1997). The alternative I propose is based on a rejection of the idea that lexic al entries are inserted in the syntax en bloc (including their phonological form). Researchers from a wide range of backgrounds have argued that syntax is free of phonology, and that the association of syntactic terminals with phonological content is a matter to be dealt in the mapping to phonology. Of course, spell-out of syntactic terminals is not the only process that must take place at the PF interface. As is well known, there is a reasonably systematic mapping between syntactic and prosodic representations (I refer to this process as prosodic phrasing). The question therefore presents itself how spell-out and prosodic phrasing are related. Is it possible for spell-out rules to refer to prosodic structure, or are such rules only sensitive to the feature content of the terminals they realize? The standard answer, to the extent that there is one, seems to be that prosodic structure is irrelevant to spell-out. In Halle and Marantz 1993, for example, vocabulary insertion is ordered after application of morphological rules, but prior to PF. I suggest instead that spell-out may in fact sensitive to prosodic phrasing. On this assumption, soft mutation.in Welsh can be analyzed as a case of lexical allomorphy conditioned by prosodic structure. In other words, mutating words have two lexical variants, and the choice between them is determined by their prosodk context. (This analysis owes a great deal to Hannahs' (1996) account, although it differs from it in :crucial respects.) Much of the empirical coverage of TaHerman an(tBorsley�s proposal can be retained in this way, but no syntactic feature need be stipulated to deal with soft mutation. 2. The

Nature of Prosodic Representations

My analysis of soft mutation will partly be based on certain assumptions about prosodic representations. Following Selkirk (1981) and Nespor and Vogel (1986), I assume that such representations consist of a series of nested domains organized according to the hierarchy in (2) (where U stands

Soft mutation at the interface 449

for Uttterance, I for intonational phrase, cI> for prosodic phrase, Ol for prosodic word, F for foot and (f for syllable). U > I > > Ol > F > 0'

(3)

The symbols in (2) are usually taken to be labels of non-terminal nodes in a tree. In joint work with Hans van de Koot (2005), however, I have argued that there are good reasons to return to string-based prosodic representations of the type found in Chomsky and Halle 1 968 and McCawley 1968. This can be achieved by treating 0", 0), <1>, etc., as boundary symbols, rather than as category labels. Thus, cI> does not stand for a prosodic phrase, but rather for a prosodic-phrase boundary. We can then replace the standard prosodic structure in (4a) by the one in (4b). [u [I [I() [00 John's] [II) father]] [1(1 [II) suggested] [II) a two-seater]]] [I [III [II) but] [00 John 's] [II) mother]] [cJ1 [co preferred] [Cl) a fur] [00 coat]]]]. b. U John' s 0) father (f) suggested 0) a two-seater I but 0) John's 0) mother <1> preferred 0) a fur 0) coat U.

(4) a.

No infonnation is lost in the transition from hierarchical to linear structure in (4). This is because the interpretation of (4b) is guided by the condition of Proper Containment, as fonnulated below: (5)

A boundary at a particular level of the prosodic hierarchy implies all weaker boundaries .

. This principle is shared by string-based and tree-based theories. It rules out representations in which a syllable is straddled across two feet, a feet across two prosodic words, and so on. Thus, in a string like u 111 F 222 (J the material between the syllable boundaries cannot be construed as a single syllable since it contains a F-boundary. Similarly, any material not separated by a ro-boundary or a boundary of a superordinate domain, belongs to the same prosodic word, while any material not separated by a d>-houndary or a boundary of a domain superordinate to w, belongs to the :same prosodic phrase. Consequently, the representations in (4) define the same sets of prosodic domains: .. .

• ••

450 Ad Neelema n

(6) a. Prosodic words:

{John's; father; suggested; a two-seater; but; John's; mother; preferred; a fur; coat} h. Prosodic phrases: {John's father; suggested a two-seater; but John's mother; preferred a fur coat} c. Intonational phrases: {John's father suggested a two-seater; but John's mother preferred a fur coat} d. Utterance: {John's father suggested a two-seater, but John's mother preferred a fur coat}

Although the notations used in (4a) and (4b) are equivalent in this respect, there are certain tree structures that have no counterpart in the string-based notation. Crucially, the string-based theory cannot express recursion. As an example, consider a tree in which a prosodic phrase that contains an intonational phrase, which in tum dominates two prosodic phrases (as in (7a». It is simply not possible to construct an equivalent representation using only boundary symbols. (7b) seems to come close, but in fact this representation contains three intonational phrases, which themselves only contain prosodic phrases. In other words, (7b) is equivalent to (7c), and not to (7a). (7)

* . . [(1) I 11 [I [ 222 ] [111333]] 444] ... . . cI> I I I I 222 333 1444 c. b· [(I) Ill]] [ I [ 222] [
b.

.

.

••.

. .

..

It is widely accepted that recursion is absent in phonology. If prosodic

representations are trees, however, this. does not follow, as trees by their very nature allow recursion. Therefore, Selkirk (1984) and Nespor and Vogel (1986) are forced to adopt a principle that explicitly rules out representations like (7a) (notice that a tree like (7a) does not violate Proper Containment). The principle in question is known as the Strict Layer Hypothesis: (8)

A given nonterminal unit of the prosodic hierarchy, XP, is exclusively composed of one or more units of the immediately • lower category, XP•1

The fact that absence of recursion in phonology follows without stipulation in a string-based theory is one argument for representations like (4b). Other arguments, further discussion and relevant references can be found in Neeleman and Van de Koot 2005.

Soft mutation at the interface 45 1

arguments, further discussion and relevant references can be found in Neeleman and Van de Koot 2005. 3.

Mapping Rules

As explained, syntactic structures must be mapped onto prosodic representations at PF. I believe that that this process has the following stages: (i) an initial prosodic representation is created on the basis of syntactic information; (ii) syntactic terminals are associated with phonological forms; (iii) the initial prosodic representation is adjusted on the basis of properties of the inserted phonological material, taking into account weight distribution, speech rate, and so on. Models that come close to this are given in Kaisse 1985, and in particular Ghini 1993 and Monachesi 2003. The latter two also make a distinction between initial prosodic phrasing (in their terminology <1>­ domain formation) and subsequent phonological adjustment (in their terminology (I)-formation). Moreover, they use some of the same mapping rules introduced below. The initial prosodic structure is determined by alignment conditions that associate edges of syntactic categories with prosodic boundaries (see Selkirk 1986, McCarthy and Prince 1993, amongst others). In predominantly head-initial languages, for example, the right edges of syntactic XPs typically coincide with -boundaries, as expressed by the mapping rule in (9) (a reformulation of Selkirk's (1986) right alignment rule). It is this rule that licenses the -boundaries in (10) (where syntactic terminals appear in small capitals. ) Insert at [[A friend [of Mary's]] [showed [some pictures] [to John]]]. U A FRIEND ro OF MARY'S SHOWED co SOME co PICTURES TO JOHN U. Two other rules are relevant to the analysis of soft mutation. First, Chomsky and Halle (1968:372) observe that the left edge of finite CPs typically coincides with the beginning of what today would be classified as . a new intonational phrase. The relevant rule, (11), has the effect that (12a) is not mapped onto (l2b), as one might expect on the basis of the earlier for <1>- insertion, but onto (I2c). Insert 1 at

452 Ad Neeleman

( 1 2) a. [(John] [believes [cp that [Mary] [loves [Bill]]]]]. b. *U JOHN cI> BELIEVES fl> THAT co MARy LOVES ro BILL U. c. U JOHN cI> BELIEVES [ THAT 0) MARy cI> LOVES co BILL U. Second, -boundaries are weakened between prosodic phrases that correspond to a modifier and modified material. Application of the rule responsible for this is usuaJly optional, but it obligatorily affects the prosodic boundary at the right edge of pronominal modifiers. Thus, ( 1 4a) is realized as in ( l 4c), rather than ( 1 4b).

( 1 3)

Reduce to co in the context I I I 222 if: I I I and 222 spell out a modifier and the constituent it modifies, and: b. I I I and 222 do not contain {f)-boundaries.

a.

( 1 4) a. [DP A [NP [OegP VERY OLD] [NP HOUSE]]] b. * .. A VERY 0) OLD HOUSE c. .. A VERY 0) OLD co HOUSE ... .

•••

.

With these rules in place, we can return to soft mutation in Welsh. 4.

Soft Mutation

Recall Borsley and Tallerman's generalization that a constituent bears soft mutation if immediately preceded by either a lexical trigger or by a phrase which c-commands it. An example of lexically triggered mutation is given in ( I ); the more intriguing case of mutation apparently . triggered by a preceding phrase is illustrated in ( 1 5)-( 1 7). U Gwelodd 00 y dyn tv ddafad U. (*dafa4) saw the man ! . : ' sheep , , , ', ' b. U Mae 'r 0) dyn we�Lro gweld oo dafad 'Ui' " see is the man sheep after

( l S) a.

'

. '

( 1 6) a

.

b.

U Dechreuodd Emrys tv began Emrys U Mae (i) Emrys wedi OJ Emrys is after

.

ddarllen y llyfr U. (*darllen) read the book dechrau co darlleD 0) y llyfr U. begin read the book

Soft mutation at the interface 453

Mae ro is b. U Mae ro is V

( 1 7) a.

yn ro yr ardd afr U. ( * gafr) in the garden goat gafr yn Ol yr ardd U. goat in the garden

Borsley and Tallerman 's analysis takes the preceding XPs to directly trigger mutation. However, by the rule in (9), the right edge of an XP typically coincides with a -boundary (as indicated in ( 1 5)-( 1 7» . It is therefore possible to analyze soft mutation as the result of lexical allomorphy conditioned by the initial prosodic structure. Mutating words have a default form, as well as a special form which is inserted if they immediately follow a triggering element. The schemas for the relevant spell-out rules are given in ( 1 8a,b). (Note that the adjacency of the trigger T is defmed in phonological terms: the variable X that separates T from the insertion site may not be associated with a spell-out rule. 2 I briefly return to the issue of empty categories below.) The alternation in ( 1 5) can be understood if among the list of triggering elements we find the boundary symbol <1>, as stated in ( 1 8c). When preceded by an XP in syntax, A is usually realized in its mutated form laMI, because at PF it is preceded by «1> . When preceded by a (non-triggering) head in syntax, however, A assumes its default form, as it is not be preceded by <1>, but by ro. •

A +-+ /a/ A +-+ laMI in T X (where X is not associated with a spell-out rule) c. T - -{ . . . ynpRED, i, «1>, } •.•

By the elsewhere condition (formulated in ( 1 9» , ( 1 8b) will block ( 1 8a), ' because its domain of application is a subset of ( 1 8a)'s (it is a more specific -rule). Hence, soft mutation is predicted to be obligatory where a trigger is present. This is by and large the correct result, although there are some . complications involving adverbials (see footnote 3). Let R } and R2 be competing rules that have D 1 and O2 as their respective domains of application. If D } is a proper subset of D2, then R I blocks the application of R2 in D t •

2

This

fonnulation implies that elements that are optionally spelled out

. mutation

.

will

block

454 Ad Neeleman

My proposal is not that soft mutation affects words that appear initially in prosodic phrases. This would be incorrect, as prosodic phrases that appear in isolation or utterance-initially do neither allow nor require mutation at their right edge. For example, the negative marker in (20a) must appear in its mutated form, whereas the utterance-initial negative marker in (20b) must remain unmutated. This contrast falls out from the way I represent prosodic structures. A lthough dim y dyn ' not the man' in (20b) qualifies a prosodic phrase (by (5» , it is not preceded by a cI>-boundary, but by the stronger U-boundary (see section 2 for discussion). Hence, there is no trigger for mutation in this context. (20) a. b.

Dydy co Emrys cI> ddim (cI» yn glyfar U. (*dim) NEG-is Emrys NEG in clever U Dim y dyn cI> a ddaeth U. NEG the man PRT came U

The proposal so far is empirically equivalent to Borsley and Tallerman' s XP-trigger hypothesis. However, it does not require an ad hoc feature to facilitate the syntactic encoding of soft mutation. Hence, all else being equal, I believe it is more attractive. But of course, all else is never equal. Borsley discusses a number of data that are problematic for the original version of the XP-trigger hypothesis. He captures these with the help of a certain assumptions common in HPSG but not in the principles-and­ parameters framework (such as the availability of n-ary branching). Whether these data can also be captured by the rules in (18) is a question that awaits further exploration. I do believe, however, that first indications ' 3 are promising. The claim that cI> should be included in the set of triggers predicts that where a phrase is exceptionally followed by a boundary symbol other than , soft mutation will not take place. In section 3 I have mentioned two such cases. First, by the rule in ( 1 1 ), CP requires insertion of an I-boundary at its left edge. Even when it follows a phrase, CP-initial material will consequently be preceded by I rather than tI> (compare (3» . It is therefore correctly predicted that there will be no mutation in examples like (21b), even though the wh-expression pwy 'who' follows a subject pronoun. (That

:\

See also

Hannahs' ( 1 996) remarks on (lack of) mutation in coordinate structures.

Soft mutation at the interface 455

pwy is a mutating word is shown by (2 1 a), where it follows the triggering preposition 0. 4 (2 1 ) a. b.

rhoddodd Emrys y llyfr U. (*pwy) U I bwy y to whom PRT gave the book Emrys U Gwn i I pwy a ddaeth yn 61 U know I who PRT came back

Second, by the rule in ( 1 3), a -boundary between a prenominal modifier and a noun is obligatorily weakened to a ro-boundary. As a result, prenominal modifiers should not trigger soft mutation, a prediction that appears to be correct: dydd 'day' in (22) appears in its default form. 5 (22)

. . mwy ro gwyntog ro dydd <1> . . . day . . . more windy .

Other issues that Borsley ( 1 999) raises have to do with empty categories. It follows from what I have said so far that empty categories can neither trigger nor block mutation (but see footnote 1 ). However, in joint work with Peter Ackema (2004), I have argued that certain empty categories trigger the rule in (9). Those that do will indirectly trigger mutation, because the
. .j

U Pwy a PRT wh o

gwelodd tDP ddafad U. (*dafad) sheep saw

Such reanalysis is not obligatory with other types of modifiers. This predicts, correctly, that mutation should be optional with adverbials. However, adverbials also mutate if topicalized. I am hopeful that this can be explained on the basis of the intonational properties of topicalization, but it would take me too far afield to attempt to do so here (the analysis would be based on the idea that topicalized phrases are phonologically represented as prosodic phrases that remains unparsed Cit the I-level; comparable structures are discussed in Neeleman and Van de Koot '2005). ·5 This is the productive pattern with equative. comparative and superlati ve APs. Most other APs follow the nominal head. There is a handful of prenominal elements, such as unig 'only', which trigger mutation. On the present account these 'would have to be classified as lexical triggers.

456 Ad Neeleman

The empirical impact of this suggestion depends on what class of empty categories trigger the rule in (9). The simplest answer would be: all phrasal ones (that would require no adjustment of the theory as it stands). Whether a claim along these lines can be upheld depends on one's view of empty categories. The mutation data would fall into place if pro and the trace of A'-movement exist, but PRO and NP-trace do not. Alternatively, a complication of the theory is necessary: non-ease-marked empty categorie s must be assumed not to trigger (f)-insertion. Why this should be so, I will have to leave for future research (but see Berendsen and others for supporting evidence). The proposals just sketched are reminiscent of Hannahs' ( 1 996) suggestion that soft mutation takes place at the juncture between two phonological phrases (hence in the context [II> ] [II> D. I believe my proposal is simpler in certain respects. This is partly because the alignment rules adopted in section 3 differ from the mapping rules Hannahs works with. Certain other differences have their source in the fact that Hannahs assumes that prosodic representations are trees, while I assume they are strings. To begin with, a tree-based theory requires a complication of the above structural description in order to deal with cases like (2 1 ). But more importantly, Hannahs' analysis must treat soft mutation triggered by preceding phrases as fundamentally different from soft mutation triggered by preceding lexical items. After all, the contexts of the two cases must be rather different. The proposed reintroduction of boundary symbols allows a single structural description: soft mutation is always triggered by an immediately preceding triggering element.6 •••

_

•••

References Ackema, Peter, and Ad Neeleman. 2004. Beyond Morphology. Oxford: Oxford University Press. Berendsen, Egon. 1 985. Tracing ' Case in Phonology. Natural Lqnguage and Linguistic Theory 3 : 95- 107.

.

,

"

Borsley, Robert. 1999. Mutation and Constituent structure in Welsh.

.

Lingua 109:

267-300.

Borsley, Robert, and Margaret Tallerman. 1 998. Phrases and Soft Mutation in Welsh. Journal ofCeltic Linguistics 5: 1-33. Chomsky, Noam, and Morris Halle. 1968. The Sound Pattern of English. Cambridge: MIT Press. (, Notice that both boundary symbols and specific lexical items are legitimate objects at the PF interface, assuming that this is the locus of vocabulary insertion.

Soft mutation at the interface 457 Ghini, Mirco. 1 993. -Fonnation in Italian: A New Proposal. Toronto Working Papers in Linguistics 12 : 4 1 -78. Halle, Morris, and Alec Marantz. 1993. D istributed Morphol ogy and the Pieces of I nflection . The View from Building 20, ed. by Kenneth Hale and Samuel 1. Key ser, 111- 176. Cambridge: MI T Pre ss . Hannahs, SJ. 1 996. Phonological Structure and Soft Mutation in Welsh. In Interfaces in Phonology, ed. by Ursula Kleinhenz, 46-59. Berlin: Akademie Verlag. Harl ow, Steve. 1 989. The syntax of Welsh Mutation. Natural Language and Linguistic Theory 7: 289-3 1 6. Kaisse, Ellen. 1 985. Connected Speech. Orlando: Academic Press . McCawley, James. 1968. The Phonological Component of a Grammar of Japanese. The Hague: Mouton. McCarthy, John, and Alan Prince. 1 993 . Generalized Alignment. In Yearbook of Morphology 1993, ed. by Geert Booij and Jaap van Marie, 79- 153 . Dordrecht: Kluwer.

Monachesi, Paola. 2003. The Verbal Complex in Romance. Oxford: Oxford University Press. Nee leman, Ad, and Hans van de Koot. (2005). On Syntctic and Phonological Representations. Ms. UeL. To appear in Lingua. Nespor, Marina, and IreneVogel. 1986 . Prosodic Phonology. Dordrecht: Foris. Roberts, Ian. 1997. The Syntax of Direct Object Mutation in Wel sh. Canadian Journal o/Linguistics 42: 141-168. Selkirk, Elizabeth. 1 98 1 . On Prosodic Structure and its Relation to Syntactic Structure. In: Nordic Prosody 11, ed. by Thorstein Fretheim, 111 -140. Trondheim: TAPIR. Selkirk, Elizabeth. 1984. Phonology and Syntax. Cambridge: MIT Press. Selkirk, Elizabeth. 1986. On Derived Domains in Sentence Phonology. Phonology Yearbook 3 : 37 1-405. Tal l ennan, Margare t. In press. The Sy ntax of Welsh 'Direct Object Mutation' . Revisited. L ingua.

Abracadabra, the relation between stress and rhythm Anneke Neijt Introduction 1

1.

Generative linguistic theory aims to characterize what human beings know about their language, the structure they assign to utterances. Research over the past thirty years has shown that the structural complexities of language are understood best in a modular approach, with components or levels of description with autonomous internal consistency, and a mapping relation between these. For instance: separate constituent structures in prosodic phonology and syntax, separate tiers in segmental phonology for the description of manner and place of articulation. This paper addresses stress and rhythm, two levels within prosodic phonology. The fIrst section shows the necessity to distinguish between both levels. The sections thereafter show that the position of main stress is dependent on rhythm and that rhythm is dependent on the position of main stress, in a way that seems to require global rules or constraints. Stress and rhythm thus illustrate the challenge of future linguistics: to explain how different levels of language structure are related to one another. 2.

The distinction between stress and rhythm

Stress and rhythm are not the same. Stress is assigned to words and phrases in· order to establish heads and dependents. Rhythm forms chunks of infonnation with an iterative pattern. Consider for instance single digit pronunciation of telephone numbers, with groups of two or three syllables, according to rhythmic constraints that hold for poetry as well. The number 02436 1 2048 will be split in 024-3 6- 1 20-48, or 02-43-6 1 -20-48, or any other pattern of two or three syllables, as illustrated in ( 1 ). ( 1)

I My

Rhythmic structure of strings of numbers 024-36-1 20-48 uu - / u - / uu - / u 02-43-6 1 -20-48 u - / u - / u - / u - / u thanks go to Mirjam Emestus, Paula Fikkert and Judith Hanssen for their disclaimers apply.

comments. All

Abracadabra

459

Such constituents of two or three elements are rhythmic feet, for which a notation with macrons (-) for beat positions and breves (u) for lapses between the beats is current in poetic analyses. Strings of feet form perfect rhythms when (a) a pattern of at least two beats is created, (b) beat positions are not adjacent, and (c) no more than two lapses occur between the beats. One of the essential differences between rhythm and stress is structural ' depth. For rhythm a constituent structure with most often only two layers of infonnation is needed: the items relevant for rhythm (in our examples: syllables) and the prominence relations between these. In exceptional cases, rhythmic patterns are embedded in larger rhythmic patterns, as in the phrasing 0243-6 1 -2048, with phrases of four syllables that consist of the smaller phrases 02-4 3 and 20-48. Stress structures on the other hand contain constituents that in principle are as deeply embedded as syntactic constituents, although of course they are subdivided into flatter structures in phonology. Another difference between stress and rhythm manifests itself in focus assignment. Stress, but not rhythm, detennines the scope of accents. For instance, contrastive accent on the modifier of a compound in Dutch will allow for both wide and narrow scope, whereas contrastive accent on the head of a compound signals narrow scope only, cf. (2). In these examples, the syllable with main stress is underlined, and capitalization indicates the syllable with contrastive accent. TIMmerman ofkapiTEIN 'carpenter or captain' (wide scope: CARPENTER contrasts with CAPTAIN) b. TIMmerman ofKL!lSjesman 'carpenter or handyman' (wide or narrow scope: TIMMER 'to hammer' contrasts with KLUSJES 'odd jobs') c. TimmerMAN oftimmerVROUW 'carpenter or carpentry-woman' (narrow scope only: MAN 'man' contrasts with VROUW 'woman')

(2) a.

The relation between contrastive accent and stress is systematic and functional: wide scope interpretations are available only for domains in which main stress and accent coincide. When main stress is overruled by .constrastive accent, only the narrow scope interpretation is available. This :is illustrated by (2c), in which the accented syllable is more prominent than ,the syllable with main stress in the compound.

·.� 460

Anneke Neijt

Mismatches between accent and rhythm, on the other hand, are irrelevant. Syllables with contrastive accent often are rhythmically prominent as well, but they need not be so, cf. (3). (3) a. Op een NIET passend moment willen komen. b. At a N OT suitable moment, they want to arrive.

In (3), syllables with main word stress are underlined, syllables indicating contrastive accents are capitalized and rhythm is indicated with macrons and breves. The examples in (3) show that scope of focus is independent of rhythm, those in (2) show that scope of focus depends on stress. Mismatches of stress and accent are infonnative, but m ismatches of rhythm and accent are not. Another indication of the difference between stress and rhythm is variation. Within a focus domain, the position of main stress is fixed, but rhythm is variable. Consider for instance the names [thaka and Polyf?ides. When pronounced in isolation, full rhythm is required (lthiikii and Po/f1ewes), a pattern with at least two beat positions, ( u and u-

-

-

u).

When embedded in an utterence, a reduced rhythmic form, with only one beat, is available. Compare (4a) and (4b), from lines of hexameters, a context that requires iterations of six feet, u u or u (Homerus Odyssee, ed. De Roy van Zuydewijn, 1 992). -

-

(4) a.

Full rhythm over te halen zijn Jthaka te vergeten (Book i:58) 'to persuade to forget his Ithaka? . . . maakte Apollo/ PolyVfeldes tot de uitnemendste ziener der mensen (Book xv:258) 'Apollo made Polyfeides the best visionary of mankind' h. Reduced rhythm Zelf zal ik nn nam- Ithaki gain om zijn zoon te bezielen . (Book i:90) 'I myself wil go now to Ithaka to inspire his son' zo ook Mantios, die Polyfeldes verwekte en KleItos (Book xv:252) 'so too Mantios, who begot Polyfeides and Kleitos' _

Finally, rhythmic variation is not only dependent on context, but also on tempo, as illustrated by the magic word iibriiciidgbra, in slow speech pronounced as iibraciidgbra. The rhythmic realization is variable, though the position of main stress remains unchanged. In sum, structural depth,

A bracadabra 461

focus and v ariation show that stress and rhythm need different levels of description. 3.

Rhythm's influence on the position of main stress

Ev en though there are good reasons to assume that stress and rhy thm require different levels of descripti on, there is a ti ght conn ecti on between the two, as is ev ident for instanc e in studies on the typology of stress sy stems and in studies on stress clashes. I nterestingly , we also found influences of stress c lash on the position of main stress i n simplex words (Em estus a nd Neijt to appear). The c onclusions of our paper are that the location of main stress in Dutch, G erman, a nd English monomorphemic words is affec ted by word length and that there is im portant variation. In the c ontex t of the present paper it wi ll be relevant to consider in m ore detail how the number of syllables in the word (three or four) may affect the posi tion of m ain stress, bec ause it wi ll be shown that rhythm codetermines the position of main stress. I n our study, we c ollected j udgements by na tive speakers on the position of main stress in pseudowords. The words v ari ed in their number of syllables ( three or four), but the final three syllables were identicaL The weight of these syllab les allows the position of main stress to vary, see (5). (5)

Phonological restrictions of the words included in the experim ents a. fi nal sy llables are (i) closed (VC) or open (VV) i n D utch and G erman, (ii) and heavy (VVC or VCC) or light (Ve) i n English b. prefi nal sy llables are open (VV) in all three languages

.. Several lists of pseudowords were created, such that participants saw either the quadri sy llabic or the trisy llabic version of a word. F or instance, one list contained potabovoo and bokatass, another list c ontai ned tabovoo and dibokatass. This way, the factor length form ed the only variable factor, under the assumption that stress is assigned from right to left in Dutch, Gennan, and E nglish. Chart (6a) presents the distribution of penultimate stress assignm ents. I nterestingly , chart (6b) shows that the variation in . preferences for m ain stress by nativ e speakers ob served in the pseudoword . experiment is also present grosso m odo in the CELEX-c orpora, which contain 3 1 1 Dutch, 55 G erman and 11 6 E nglish simplex words of three and four syllab les wi th the c haracteristic s of (5).

462 Anneke Neijt

(6) a. Pe nultimate main stre ss in pseudo wo rd experime nts (% , n=2343) 60 �------� 50 40 30 20 10 o

h.

Dutch

German

English

Pe nultimate main stress in exi sting wo rds (%, n= 146) 60 �------� 50 40 30 20 10 o

Dutch

German

English

The differences betwee n the l anguages, whic h may be due to othe r facto rs such as differences in syllable weight, are o f no concern to the issue at hand. For the purpo se s of our inve stigatio n, the simil aritie s between the l anguage s are important. I n eac h l anguage, penul timate stre ss is assigned more o ften in words of fo ur syllable s than in wo rds of three syllables. S imilar, but mirro r imaged charts represent the distri bution of antepenultimate and final stre ss in words of three o r fo ur -syll able s. The p refere nce s found are: ( 7) a. A ntepe nultimate stre ss: mo re o ften assigned to words of three syll ables. b. Pe nul timate stre ss: mo re ofte n assigne d to words of four syll ables. c. Final stre ss: more often assigned to words of three sylJ able s.

Abracadabra 463 We explain this effect of length as being driven by a preference for initial stressed syllables and alternating patterns of stressed and unstressed syllables. The following tableau illustrates the relation between stress and full rhythm (notice that the pseudoword experiment evoked full rhythm, because the words were presented in isolation , as items of a list).

(8) 3 syllables

ANTEPENULTIMATE gria -

4 syllables

u

-

Penelope

PENULTIMATE Maria u-u

rododendron

u-u-

-

u -v

FINAL STRESS

chocola -

u

-

jilatelie

u - u - or

-

u u-

On the basis of general constraints on perfect rhythm (at least two beats, no clashes or lapses), and the assumption that rhythm signals initial word edges in Germanic languages, the explanation of the patterns found are as follows (comparing rhythm in the columns of this tableau, according to the design of our experiment):

(9)

antepenultimate stress: U is preferred over u - u penultimate stress: u u is preferred over u - u final stress: U is preferred over u u - (but equals uu -

-

-

-

-

-

-

-

-

-

)

.

Alternatively, one might compare the set of rhythms available for words of three or four syllables (the rows in the tableau): (1 0)

3 syllables: . 4 syllables:

-

-

u

is preferred over u - u, and u - u and - uu are preferred over u - u -

-

,

Both ways, rhythm straightforwardly explains the length pattern found in the three languages. 4.

Stress's influence on rhythm

The variation in rhythmic phrasing is influenced by context. Reconsider the poetic lines in (4), which show that preceding and following context detennine rhythmic phrasing of variable words such as Ithaka and

464 Anneke Neijt Polyfeides. The following pilot experiment shows that this context effect can be observed within words as well: the position of main stress at the end of the word codetermines rhythmic phrasing of the initial part of the word. Thirty six participants, speakers of Dutch, were asked to determine phrasing within a word, as they would do, for instance, in explicit diction of the word ondoorgr.ondelijk ' incomprehensible': onldoorgron/de/ijk, or ondoorlgrondellijk, or any other partition. The data presented (mostly unknown words or pseudowords with an odd number of syllables before the position of main stress; this position underlined and bold) and the results are summarized in ( 1 1 ) Sometimes participants divided a word into monosyllabic groups only (cf. onJdoorlgr.on/del/ijk). Such answers are uninformative and therefore not included in ( 1 1 ). The word abracadgbra was incorporated with its three reduction variants (schwa written in this experiment as e, the usual orthographic form of schwa in Dutch). .

(1 1)

Rhythmic phrasing at word onset After each word, the number of answers with a first constituent of respectively one, two, three, four or even five syllables is presented, along with the distribution of initial constituents with one, two and three syllables in percentages. a. Antepenultimate stress: amorosissimo 1 0- 1 2- 1 2- 1 -0 (29-35-3 5%) mimimimimim!mimi 3- 1 3- 1 4-2-2 ( 1 0-43-47%) b. Penultimate stress: abracad!!bra 7-29-0-0-0; abraced!!bra 1 1 -22-3-0-0; abrecad!!bra 4-29-3-0-0; abreced!!bra 7- 1 8- 1 1 -0-0 (20-68-1.2%) amerikanisme 1 7-8-3-8-0 (6 1 -29- 1 1 %) latitudinarisme 8-2 1 -5- 1 -0 (24-62- 1 5% dududududud!!du 3 - 1 8- 1 0- 1 -2 ( 1 1-57-32%) c_ Final stress: fonolo� 5-27-3-0-0 (9-8 1 -9%) tatatatatat!! 0- 1 8- 1 2- 1 -2 (0-59-4 1 %)

The data collected for amerikanisme 'americanism' suggest that not only length of the word determines the choices, but also morphology. The fact that the base of this complex word, amerika, has four syllables undoubtedly influenced the high number (8 choices) of first constituents with four syllables in this word. The issue of the influence of morphology on rhythmic phrasing needs further study.

Abracadabra 465

Several trends can be observed. First, in longer words, participants prefer larger phrases. Observe that larger constituents are also prefered in the reduced forms of abracadabra, illustrating the relation between syllable weight, speech tempo, and seize of rhythmic constituents. Second, in words with final and penultimate stress, initial constituents of two syllables are most frequent (with the exception of amerikanisme, cf. above). This suggests that rhythmic phrasing is applied from left to right (which forms further evidence for the suggestion by Ernestus and Neijt, and others, that prosody indicates word onsets in Germanic languages). Third, and most important for the issue of this paper, the position of main stress influences rhythm. The highest mean percentage of trisyllabic first constituents is found in words with antepenultimate stress. Rhythmic phrasing at the end of the word, induced by the position of main stress, influences rhythmic phrasing of word onset. 5. Conclusion

The miraculous abracadabra of human language, its robustness and flexibility, is due to its multitude of levels, and the intricate ways these levels relate. Above data illustrate the issue of matching and mismatching levels in a nutshell. Using familiar terminology of generative grammar, the influence of rhythm on the position of main stress can be characterized as a global mechanism in which different derivations are evaluated. Similarly, the influence of the position of main stress on rhythmic phrasing can be characterized as a mechanism in which several options of phrasing the string are evaluated, and most often the best option is chosen. It turns out that the individual rules, constraints, or patterns within a level are strictly local, but preferences of use and interpretation are governed by comparison of different levels of the same derivation or of different derivations. Even though derivations are no longer part and parcel of current linguistic approaches, the concepts introduced will be useful in new frameworks, and it still makes sense to aim at local descriptions, to distinguish types of globality, and to further investigate the relation between levels. References Emestus and Neijt. To appear. Word length and the location of primary word stress in Dutch, German, and English. LinguistiCS.

What do we learn when we acquire a language? Marina Nespor, Judit Gervain and Jacques Mehler 1. The

problem 1

One of the aims of the theory of generative grammar is to account for all and only the possible sentences of a language in a way that explains not only the l inguistic competence of adults, but also the acquisition of this competence by children. Parts of this knowledge concern properties shared by the over six thousand languages spoken on our planet. Humans have a genetically specified disposition to exhibit these properties during the normal course of their development given regular or sometimes even impoverished input (Chomsky 1 959, 2004). Other properties show systematic variation among the world's language, and although the set of possibilities are also genetically given, infants have to be able to select the structures characterizing their native language. Two basic questions must be answered in order to understand this procedure: ( 1 ) a. b.

Which properties of language must be learned? By what mechanisms are they learned?

Among the non universal properties, i.e . . those that must be learned, there are not only arbitrary properties such as the lexicon, including derivational and inflectional morphemes, but also many of the regular properties of the different grammatical components including the value of grammatical parameters. Classical examples of systematic variation among languages include word order (Object-Verb vs. Verb-Object order, . the Head­ 2 Complement parameter , e.g. Dryer 1 992), morphological type (the extent I The following are some thoughts we have started discussing with Henk and we hope to continue discussing in the future. 2 In generative grammar, sentences are built up of phrasal constituents. A phrase has a defined inner structure consisting of three sub-constituents: the Head, the Specifier and the Complement. The syntactic behavior of the phrase is defined by its Head, which is a lexical entry (bound or free morpheme) level category. The Head subcategorizes for the Complements it requires in its phrase. Complements

What do we learn when we acquire a language? 467

of aggl utination, e . g. Jul ien 2002) or the optionality of pronominal subje cts (pro- drop, Rizzi 1 992). Given the differe nt nature of the properties that mus t be learne d, it is l ikely that dive rse me cha nisms are involve d in this enterprise. It is quite clear that one does not learn the lexicon in the same way as one le arns syllabic structure or word order. S yl labic structure, in tum, is not learned li ke word orde r or the s cope of quantifie rs. O ne main reason for believing that differe nt mechanisms are involve d is that the different properties are present or s ignaled in the s pee ch input to different extents. Some aspects of language s tructure, e. g. the syllabic repertoire or the pro- drop property, are more ore less dire ctl y observable in the spee ch si gnal, where as others, e. g. the abstract Head-C ompleme nt parameter, are not. This latter fact cons titutes the empirical and l ogical b asis of Ch oms ky'S poverty of s timul us argume nt (C homs ky 1 959), which he used in order to argue for the necessity of innatel y s pe cified linguis tic ' knowledge '. Although parameters have been proposed to account for l inguistic variation in all grammatical components, the me chanis ms throu gh which the properties they des cribe are learne d might differ. I n addition, more than one mechanism migh t be i nvolve d in the learning of one si ngle property . Research on language a cquisition in the generative traditi on has al most exclusivel y focused on the setting of highl y complex stru ctural parameters tha t childre n te nd to acquire rather late, i.e. when they al re ady produce l anguage (Wexler and Gibs on 1 994, and for a revie w of the fiel d Guasti 2002). Strongl y domain-specifi c mechanisms, such as trigge red parame ter setting, h ave been proposed to account for the acquisition of these properties . Howeve r, certain as pects of acquisition start mu ch earlier, already in the pre-linguistic phase, possibl y s oon after birth, as s uggested indire ctl y by some neurophysiol ogi cal fi ndings (Pefia et al. 2003). U nl ike in l inguistics, this devel opmental stage has received some attention in devel op�ental ps ychol ogy and psychol inguistics (Wanner and Gleitman 1 982, Morgan and De muth 1 996, Saffran e t al. 1 996 among others), bringing to the fore the potential contribution not onl y of language-specifi c, but als o of domain- ge ne ral learni ng me chanisms. are typically phrase level categories themselves. The Head can also take certain modifying elements in the form of Specifiers, which are usually also phrasal in nature. The world 's languages differ systematically as to what the relative orders of the Head and Complement, and the Head and Specifier are within a ph rase, which

have been fonnalized as the Head-Complement and the Head-Specifier parameter, respectively.

468

Marina Nespor, Judit Gervain and Jacques Mehler

The main goal of the present paper is to introduce both the linguists' and the psychologists' approach, and argue for the necessity of a synthesis. 2.

The acquisition of the lexicon

One part of language acquisition concerns the learning of the arbitrary relation between sounds (or signs in sign languages) and meanings, i.e. the lexicon. Generative grammar, which is primarily a theory of syntax, has paid relatively little attention to this domain until recently. With the advent (Chomsky 1995) of the Minimalist Program (MP), the acquisition of the lexicon, in addition to the inherent importance of understanding how sound/sign. meaning associations are learned, has gained significance for the theory of syntax itself. In the framework of the MP, syntax is conceived of as a (nearly) optimal computational system mediating between the acoustic/auditory and the conceptual systems, themselves external to language. The perfection of grammar is understood as syntax being m in imal, introducing as little material into a syntactic derivation as possible (preferably none). Such a minimal system cannot accommodate linguistic variation, which is thus all delegated to the lexicon (Chomsky 1995, 2004). Fonnulated in its strongest form, the claim is that all arbitrariness in the linguistic system, including cross-linguistic variation, is ultimately reducible to differences in the properties/features of lexical items or classes thereof. In this perspective, typological variation such as word order or morphological type would be driven by the features of inflectional morphemes. Thus, what the child needs to learn when acquiring the native language reduces to arbitrary lexical features, which in tum account for the emergence of cross-linguistic differences . It remains to be tested whether this strong thesis is tenable. To date, no experimental work exists investigating its plausibility. Therefore, we leave this proposal unevaluated here, and only note that it clearly marks a renewed tendency in generative grammar to take leamability issues into consideration, and to integrate the results of more psychologically-based approaches, which, unlike generative grammar, have paid considerable attention to mechanisms of word learning. The first steps in word learning have to be the recognition that the speech stream is actually made up of discrete units, i.e. words, and the identification of their boundaries. Saffran et a1. ( 1996) have shown that adults and infants can exploit the statistical properties of the speech stream to discover words. In particular, they use the dips in transition probabilities

What do we learn when we acquire a language? 469

i

(TPs between con secutive syllables to posit word boundaries. Such a procedure clearly con stitu tes an example of a domain- gen eral learn ing mechan ism, sin ce statistical learn in g applies to visual as well as lin guistic in put (F iser an d A slin 2002), an d has been observed in non- lin guistic animals (H auser et al. 200 1 ). W e believe that, b esides the general learnin g mechani sms that are central for the lexicon , we have to in vestigate the grammatical representation s that con strain the material on which statistical computation s may be carried out. In this line of research, we have proposed that given the differen t prevalent fun ction s of con sonants an d vowels in lan guage - the latt er specialized as indicators of syntax, the form er of lexical distinction s - on ly the fonn er should be fun ctional to segmen tation an d the sub sequent acquisition of the lexicon (N espor et a1 . 2003). B oth V s an d Cs have been proposed to be adj acent on their respec tive autosegmental tier, even when they are n ot so in the phon etic sequen ce (G oldsmith 1976). Thus, in prin ciple, b oth categcries are equally available for statistical computation . I f TPs are calculated .on one tier, but nc t on the other, this would be an in dication that a gen eral learning mechanism is con strain ed by represen taticn s that are specifi c to language. Indeed, experimental results have shown that TPs are successfully computed .on conson an ts, when the interv en in g vowels vary , but nc t on vowels, when the interv enin g con son an ts vary (B onatti et al. 2005). 3.

Parameter setting

W hat is exactly param eter sett in g? What are the mechan isms respon si ble for it? A ccordin g to the original form ulation of the theory (Chom sky 1981) , lin gui stic competence can be described by prin ciples accoun ting for the un iversal properties of all natural languages, while cross- lin guistic variation boils down tc the differen t settin gs of a possibly small number of abstract binary parameters. More specifically, the set of parameters and their possib le settin gs/values are also un iversal, that is, the possible cross­ lin gui stic differences are con strain ed. This is empirically supported by the fact that ty pological variation is not random: languages come in certain varieties, but n ot in (logically perfectly possible) others. For instan ce, out .of the six possible combin ation s of basic word order (SOV, SVO, VSO, VOS, OVS, OSV), the first two accoun ts for about 80-90% of all the 3

The (forward) transition probability is the conditional probability of the Occurrence of a syllable given the presence of the preceding syllable.

470 Marina Nespor, Judit Gervain and Jacques Mehler

world's languages, and the last two are basically not attested at all. Crucial to the Principles-and-Parameters (P&P) theory is the idea that once a parameter is set to a certain value, many properties apparently unrelated on the surface follow. To stay with the previous example, if word order is indeed encoded by one or two parameters (e.g. Head-Complement) as frequently proposed in the literature (Baker 200 1 ), the correct setting of such a parameter guarantees the well-fonnedness of word order in all the different types of phrases, e.g. prepositional phrases, noun phrases etc. Under this view, the learning of syllabic structure is not the result of parameter setting, because it is directly observable in the input and requires no deep structural generalizations. We may assume that by hearing the syllables occurring in their language of exposure, infants will learn the different possibilities allowed in their language of exposure, given the universal syllabic structure. In this case, there are no unrelated properties that follow. One theoretically interesting feature of parameters then is that they are highly abstract and general, accounting for a large number of disparate phenomena. This, however, raises a logical problem, first pointed out by Pinker (1 984). How are these abstract mental entities to be learned from the concrete, physical input? One possibility to get around this linking problem is if the input contains indices that guide the infant. Depending on the nature of the cues, we can talk about semantic (Pinker 1 984), syntactic (Gleitman 1 9906) or phonological bootstrapping (Demuth and Morgan 1 996). While these theories provide general frameworks to think about parameter setting and language acquisition in general, we still need to understand the exact underlying mechanisms. How are parameters set? Some are likely to be s�t before knowledge of the lexicon, if we want to account for the pace of language acquisition. Possibly the parameter that must be set first is that concerning the general word order, i.e. the Head-Complement parameter. Through the example of a possible scenario that we are presently testing experimentally and that we simplify here for the sake of brevity, we would like to illustrate how the integration of language-specific and general-purpose learning mechanisms leads to a better understanding of language acquisition than either approach separately, giving us a hypothesis that is plausible both formally/linguistically and psychologically. We know that main prom inence coincides with new information and that new infonnation is carried by the Complement rather than by the Head, at least in the

What do we learn when we acquire a language? 47 1

4 unmarked case and with unmarked intonation. The hypothesis has recently been formulated that in addition to the different location of phonological phrase prominence in Complement-Head and Head-Complement languages - leftmost in the first and rightmost in the second (Nespor and Vogel 1 986), the physical realization of prominence might also be different: it is claimed to be realized through pitch in the first case, and through lengthening in the second case (Nespor et al. in preparation). In addition, at the right edge of the phonological phrase, lengthening due to prominence is reinforced by lengthening due to constituent final position. Thus the two word orders physically differ because stress realization and final lengthening coincide in Head-Complement, but not in Complement-Head languages. It is also known that through a general perceptual mechanism, humans tend to group both linguistic and nonlinguistic sounds that differ only in pitch trochaically and sounds that differ only in duration iambically (Cooper 1 977 Hayes 1 995). Thus the specific instantiation of prominence at the level of the phonological phrase may guide the infant in the identification of the value of the Head-Complement parameter in its language of exposure. If experimental evidence will show that our hypothesized scenario has psychological reality, then we will have shown that for the setting of a property specific to language, in this case the Head-Complement parameter, general properties of the perceptual system are involved. In addition, we will have shown that infants might very well set this parameter at the prelexical stage. 4. Conclusions

In this paper, we have argued that in order to achieve one of the main goals of generative grammar, i.e. the comprehension of the steps children make towards the acquisition of the language they are exposed to, more attention should .be paid to the mechanisms responsible for it. While general · mechanisms have been proposed for the acquisition of the lexicon and mechanisms specific to language for the acquisition of grammar, in particular of syntax, we have shown that this division is not clearcut . On the one hand, the general mechanism of statistical, TP-like computation, shown to be utilized to segment a speech stream and identify the lexical items it 4 Of

course, it is not always the case that a phonological phrase contains both a head and its complement. There are, however, enough such phrases to ensure that the infant is correctly cued.

472 Marina Nespor, Judit Gervain and Jacques Mehler

contains, is constrained by representations specific to language. In addition to being computed on syllables, transition probabilities can be computed exclusively on consonants, but not exclusively on vowels. On the other hand, the language-specific acquisition mechanism of parameter settin g might rely, for the setting of at least certain parameters, on general perception mechanisms. It is our belief that in order to make progress in the field of language acquisition, linguists and psychologists must cooperate in the analysis of the complex interaction between domain general mechanisms and specifically linguistic ones.

References Baker, Mark. 2 00 1 . The Atoms ofLanguage. New York: Basic Books. Bonatti, Luca, Marcela Peila, Marina Nespor and Jacques Mehler. (2005). Linguistic constraints on statistical computations: The role of consonants and vowels in continuous speech processing. Psychological Science, 16:6. Chomsky, Noam. 1 959. A review of B .F. Skinner's Verbal Behavior. Language, 35(1 ):26-58. Chomsky. Noam. 198 1 . Lectures on Government and Binding. Dordrecht: Foris. Chomsky, Noam. 1 995. The Minimalist Program. Cambridge, Mass.: The MIT Press. Chomsky, Noam. 2004. Language and Mind: Current Thoughts on Ancient Problems. In: Lyle Jenkins (ed.) : Variation and Universals in Biolinguistics. Elsevier. Cooper, William E. 1976. Syntactic Control of Timing in Speech Production. PhD dissertation. MIT. Dryer, Matthew S. 1 992. The Greenbergian Word Order Correlations. Language, 68: 8 1 1 38 . Fiser, Jozsef, and Richard N Aslin. 2 002 . Statistical learning of new visual feature combinations by infants. Proceedings of the National A cademy of Sciences. 99, 1 5822-1 5826. Gleitman, L il a 1 990. The structural source of verb meaning. Language A cquisition 1 :3-55. Goldsmith, lohn. 1 976. A utosegmetttal Phonology. Indiana University Linguistics Club. Guasti, Maria Teresa. 2 002 Language Acquisition. Cambridge, Mass. : The MIT Press. Hauser, Mark D., Elissa L. Newport and Richard N Aslin. 200 1 . Segmentation of the speech stream in a nonhuman primate: Statistical learning in cotton-top tamarins. Cognition 78: B53-B 64 Hayes, Bruce. 1 995. Metrical stress theory: Principles and case studies. Chicago: University of Chicago Press. -

.

.

.

What do we learn when we acquire a language? 473 Julien, Marit. 2002. Syntactic heads and word fonnation: A study of verbal inflection. New York: Oxford University Press. Morgan, James L. and Katherine Demuth. 1 996. Signal to syntax: Bootstrapping from speech to grammar in early acquisition. Mahwah, N.J.: Lawrence Erlbaum. Nespor, Marina and Irene Vogel. 1 986. Prosodic phonology. Foris Dordrecht. Nespor, Marina, Jacques Mehler and Marina Pena. 2003. On the different roles of vowels and consonants in speech processing and language acquisition. Lingue e Linguaggio. 22 1-247. Nespor, Marina, Cinzia Avesani, Caterina Donati and Mohinish Shukla. In preparation. Different prominence realizations in OV and VO languages? Pena, Marcela, Atsushi Maki, Damir Kovacic, Gislaine Dehaene-Lambertz, Hideaki Koizumi, Furio Bouquet and Jacques Mehler. 2003. Sounds and silence: an optical topography study of language recognition at birth. Proc Natl A cad Sci U S A, 1 00(20), 1 1 702-05. Pinker, Steven. 1984. Language Learnability and Language Development. Cambridge� MA : Harvard University Press. Rizzi, Luigi. 1 992. Early Null Subjects and Root Null Subjects. In: T. Hoekstra and B. Schwartz (eds.): Language Acquisition Studies in Generative Grammar. Amsterdam/Philadelphia: John Benjamins. Saffran, Jenny R., Richard N. Aslin and Elissa L.Newport. 1 996. Statistical learning by 8-month-old infants. Science, 274(5294): 1 926-28. Wanner, Eric and Lila Gleitman. 1 982. Language acquisition: The state o/the art. Cambridge University Press. Wexler, Kenneth and Edward Gibson. 1 994. Triggers. Linguistic Inquiry 25:3, 407454.

A prosodic contrast between Northern and Southern Dutch : a result of a Flemish-French sprachbund Roland Noske 1. Introduction l In descriptive studies of varieties of Dutch, up till now little or no attention has been given to differences in prosodic structure between the Northern and Southern variants of the language, spoken in the Netherlands and Belgium respectively. In the literature one finds only cursory remarks, like the one by Goossens ( 1 973), who, after having described a number of segmental phonological differences between the Northern and Southern variants of Standard Dutch (like the use of the labiodental approximant [u] in the North versus the labiovelar [w] in the South), writes: Apart from these characteristics that can be easily described, there are a number of others, which concern the variational space of the separate phonemes and allophones, their connections and intonations. (Goossens 1 973 :23 8, my translation). Despite the lack of descriptions it is immediately clear to anyone confronted with Northern and Southern Dutch that important prosodic differences exist between the two variants. This paper will focus on one ' aspect of the prosodic organization, viz. the process of syllabification and the placement of syllable boundaries in morphologically complex words. It will also treat a contrast in vowel deletion between the North and the South.

Glottal stop hisertioD and syllable boundaries

2.

In both Northern and Southern Dutch, a glottal stop is inserted in hiatus position if the second vowel is stressed: (1)

I

I

beamen [ba'?a:man] 'acknowledge'

(lba+/, verbal prefix, /+:m/, infinitival ending) would like to thank Hans Van de Velde, Marleen Van Peteghem and Julien

Venneulen for help on the Southern Dutch data.

A prosodic contrast between Northern and Southern Dutch 475 Since a phonetic glottal stop2 can only occur in Dutch in the onset of a syllable, without other elements in this position, its occurrence can serve as an indicator for the location of a syllable boundary. Let us now look at the Northern Standard Dutch forms in (2), where we see that a glottal stop is inserted if a consonant final morpheme is combined with a vowel initial one: (2) Northern Standard Dutch a. uit[?]eindelijk 'finaI(ly)' (uit, 'out'; einde 'end'; -elijk, adjectival suffix) b. ver[?]armen 'empoverish ' (ver-, verbal prefix; arm 'poor' ; -en infinitival suffix) c. on[?]eens (adj.) 'in disagreement' (on-, 'un-'; eens 'in agreement' (adj .)) d. berg[?]achtig 'mountainous' (berg 'mountain'; -achtig, adjectival suffix) We can conclude that in (2), because glottal stop can only occur as the sole element of the onset, the final consonants of the initial morphemes are maintained within the coda of the first syllable, while the glottal stop constitutes the onset of the second syllable.3 In this respect, Northern Dutch behaves like Standard German, where a glottal stop is also inserted in cases parallel to the ones in (2), e.g. in the Standard German equivalent of (2a), letztendlich [lEtst' 1EntlI�] 'final(ly)'. If we compare the Northern Dutch data with the corresponding forms in Southern Dutch, we find a marked contrast. My informants report that they find the pronunciation with a glottal stop in these cases unnatural, very official, affected and exaggerated (,resembling Northern Dutch'). Hence, the forms corresponding to the ones (2a-d) are not pronounced with a glottal stop in Southern Dutch, not even in fairly slow speech. Because, as we have seen, glottal stop insertion does occur into an empty onset in both Northern and Southern Dutch, one is forced to conclude that in the Southern Dutch forms, the final consonant of the initial morpheme is

2 Glottal stop has no phonemic value in Dutch .

3 In Northern Dutch, the glottal stop in ( 1 a-d) is sometimes omitted and the intervocalic vowel fmds itself in the onset of the second syllable. However, it should be emphasized that this phenomenon is strictly limited to allegro speech.

476 Roland Noske

syllab ified into the onset of the second syllable.4 To summarize, we find the fol lowing contrast (' . 'indicates a syllable boundary): underlying fonn

(3) a. b. c. d.

uiteindelijk verannen oneens bergachtig

Ireyt+Einda+lakl Ivar+ann+anl I:m+e:nsl

Ibery+ax.taxl

Northern Dutch [reyt. ' ?£in. da. l ak]

[var.'?ar.man]

[:m. ' ?e:ns] [' berx., ?ox. tax]

Southern Dutch [ rey. 'tein.da.h}k] [va . 'rar.man] [:l. ' ne:ns] [ 'ber .,yox.t ax ]

In (3d) we see an additional indication for the contrast in syllabification: in Southern Dutch, the final voiced obstruent Iy/ of the initial morpheme /beryl has not undergone syllable final devoicing, while in Northern Dutch it comes out as voiceless [x]. This independently confirms that this segment is in onset position of Southern Dutch. 3.

Cliticization and vowel deletion

Another, partially related, contrast between Northern and Southern Dutch concerns the deletion of vowels in pronouns due to cliticization. The neuter pronoun het [at] s, e.g., can be cliticized and can loose its schwa more easily and in more positions in Southern Dutch than in Northern Dutch. In both Northern and Southern Dutch, schwa can be deleted if it is followed by a vowel initial inflected verb, if the main sentential stress is not located on one of the words. (4)

het is it is

lat IS I

[tIs]

(Northern and Southern Dutch)

B ut in Northern Dutch this deletion is optional, depending on the speech rate. In many varieties of Southern Dutch, however, it seems to be 4

Alternatively, the intervocalic consonants in the Southern Standard Dutch case could be described as ambisyl1abic. I do not go into this matter, as it is immaterial to the issue discussed here. S In stressed position, this pronoun is pronounced [het] in Northern Dutch, [hat] in Southern Dutch (but not West-Flemish, in which the /hi is realized as zero). The exact nature of the alternation [hEt]/[hat] - [at] is unclear (i.e. whether it is a result of polymorphy or of rather isolated phonological processes). The existence of [hat] in Southern Dutch and the impossibility of * [het] in that variety are problematic for Van Oostendorp' s ( 1 995 : 197; 2000) assumption and prediction that schwa and [h] cannot be tautosyllabic.

A prosodic contrast between Northern and Southern Dutch 477

obligatory. A more marked contrast between the North and the South (in this case West-Flemish6) can be found if het is in a position following a tensed verb, e.g. in: (5) a. was het luas �tJ was it b. was het Iwas at! was it

[uasat]-[uOZ)t] * [uastJ ' [wast]

(Northern Dutch) (West-Flemish)

It thus appears that in West-Flemish, also in normal to slow speech, lat! usually looses its vowel and it is cliticized to tensed verbs both to its left and to its right. Also other unstressed pronouns, like ik 'I', je 'you', we 'we', can loose their vowels much more easily in West-Flemish and other Southern variants than in Standard Northern Dutch, where again allegro speech is required if vowel deletion is to take place at all. Examples of deletion of the vowel in ik Ilk! are given (6): (6) a. dat ik that I b. ik hoor I hear'

Ida Ik/

[dak]

(West-Flemish)

Ilk or/8

[ko:r]

(West-Flemish)

4. A difference in constraint ranking The contrast between Northern and Southern Dutch can be explained elegantly in the framework of Optimality Theory. As shown by Van Oostendorp ( 1 995 : 1 83, 2000), this type of contrast can be analyzed as a different constraint ranking of the constraints ALIGN and ONSET. ALIGN says that word boundaries and derivational morpheme boundaries must 6

My data are based on recordings and observations of West-Flemish (the dialect of

the province of West-Flanders), in which the phenomenon of vowel deletion in pronouns appears to be the strongest. However, the obligatory vowel loss in unstressed pronoums is not limited to West-Flemish but seems to be wide-spread in Southern dialects. Two recordings of West-Flemish can be found on: http://neon.niederlandistik.fu-berlin.de/langvar/westvlaams/. [uast] (without a schwa) is possible in allegro speech in Northern Dutch, but not

7

in a normal speech rate, like [wast] in the South. The underlying fonn does not contain /hi because West-Flemish has lost this

S

segment. In tum,

[V]

in other dialects, shows up as [h] in West-Flemish. Thus his­

torically the following evolution took place in West-Flemish: /hi ::::>

0, Iyl ::::> /hi.

478 Roland Noske

coincide with syllable boundaries, while ONSET says that syllables must have onsets (or in another version, that their onsets must be filled). It can now be assumed that in Southern Dutch, ONSET is ranked above ALIGN � (ONSET » ALIGN) while in Northern Dutch the order is the reverse. We can now account for the contrast in (3c) in the following way: Tableau 1: Southern Dutch

r:ir

Candidates (. on.)(.e:ns.) (.�.)(.ne:ns.)

ALIGN : . :.

ONSET **

. ".

*

' '

'

. . :*.

Tableau 2: Northern Dutch qr

Candidates (.on.)(.e:ns.) (.0.)( .ne:ns.)

ALIGN

ONSET

*

:*

: *'* . -.. . . . .

At a later stage, a phonetic glottal stop is inserted before eens in Northern Dutch. The deletion of vowel in pronouns in Southern Dutch can also be analyzed as being the result of the ranking of ONSET above ALIGN. Let us consider the case of (6b), ik hoor [ko:r], '1 hear' : Tableau 3 : West-Flemish

r:r

Candidates (.Ik.)(.o:r.) (.ko:r.)

ONSET **

ALIGN . .. . ..

.: . •

;: ,,; .:�*

As we can see, because the first candidate violates ONSET (even twice), the second candidate wins, although it twice violates the lower ranked con­ straint ALIGN. 5. A Flemish-French sprachbund

The question comes to mind what are the origins of the contrast in sylla­ bification and vowel deletion between Northern Dutch and German on the one hand and Southern Dutch on the other. I tentatively advance here the hypothesis that the observed behavior of Southern Dutch with respect to

A prosodic contrast between Northern and Southern Dutch 479

syllabi fication and vowel del eti on in p ronouns is du e to the influence of Rom ance dial ects. This assump ti on is not as far- fetched as it may s eem at first sight. F irst of all , Sou th ern Dutch and French have cohabitated for centu ries, especially in South-Western Flanders. The town of Kortrijk (Cou rtrai ), e.g., was under F rench domi nati on s everal tim es i n its history and part of its p opul ation was French-sp eaking. Second, i n the l iteratu re, a number of phonol ogi cal p henom ena that have cross ed the Germ ani c-Romance linguistic border i n p res ent-day Belgium N orthern F rance have been attested. There have also been common devel opments in neighboring Rom ance and Germanic dial ects. D e Schutter (1 999) m entions five of these phenom ena: i.

fi nal devoi ci ng (a steady feature of Dutch and G erm an), showing up in F rench and Pi cardian (herbe ' grass' is pronounced with a fi nal [p] i n Romance dialects of N orthern France as well as of N orth-Eastern F rance and Wall onia) , which consti tutes an infl uence of Germ anic onto the Rom ance dialects; II. the breaki ng of vowels (like in F r. jievre, piece) (in other words the devel opment of risi ng diphthongs, a histori c process that has taken pl ace l arge p arts of the Rom ance linguis ti c area), whi ch has gi ven rise to the breaking Ia! before Irl + d ental plosive in Southern Dutch di al ects: paard => [p(j )E!(r) t] ' hors e' ; iii. the occurrence of /hI as a phoneme in Wall oon dial ects, due to Germanic i nfluences ; IV. p al atization (fronting) of vowels (like in Fr. mur [m yr] and Du . muur (i n both cas es: < Lat. murus), South Wes tern Dutch veugel [veyal] 'bi rd' < vogel [voyalD (s ee also Ryckeboer 1991 , 1997: 1 4 7-169); v.

l eniti on of dental cons onants (like F r. feuille [frej ] ' l eaf , < Lat. folia) , South Wes tern Dutch diminu tive l+tSal « I+kal).

To this list we may add a sixth phenom enon, m entioned by Ry ckeboer (2004:44) for the variety of West-Fl emish sp oken i n the p art of Flanders that is presently l ocated in Fran ce, but which pertai ns to West-Fl em is h in general: vi. The m onophthongis ati on from [au] to [u] before dental or al veolar obs truents. This devel opm ent is also found in the sam e period in the nei ghboring Rom ance dialect of Picardi an.

480 Roland Noske

Given all these mutual phonological influences9, it seems legitimate to search for an explanation of the discussed contrasts between Northern and Southern Dutch in the language contact with Romance. Since the difference between the North and the South is a difference in syllabification (as shown by the lack of glottal stop insertion in the South for the words listed in (3), or are closely related with syllabification (like the vowel drop in pronoun s Southern dialects), we should take a close look at French syllabification. As is well-known, French syllabifies right through morpheme boun­ daries stopping only at boundaries above the word level, perhaps phrase boundaries. Hence, il arrive 'he comes' is pronounced [iJariv] and not * [il.ariv]. This phenomenon is called enchainement in descriptions of French pronunciation like Grammont ( 1 922). We have just seen that in the Southern Dutch data in (3), in words like uiteindelijk ([rey.'tein.d�.l�k] ), the morpheme boundary is ignored by syllabification, like in French, but unlike Northern Dutch. This points already to the possibility of an influence of French on Southern Dutch. The likelihood of a change through language contact becomes greater of one looks at the phenomena of vowel deletion in pronouns and other words in French. In French, schwa is deleted in many instances, among other cases before as well as after full vowels. This constitutes a wider variety of cases that schwa deletion in Standard Dutch. As we have seen, in (Sa), was he! Iuas �tI [uas�t] - [uaz�t] the schwa cannot be deleted in Northern Standard Dutch (except in allegro speech), but it is obligatorily deleted in West-Flemish and other Southern Dutch dialects. So, here again, it seems that Southern Dutch behaves more like French than like Northern Dutch. With these two parallel phenomena in French and West-Flemish, syllabification through morpheme boundaries and . vowel deletion in postvocalic position, one can at least tentatively assume that· there has been a French influence on Southern Dutch and on West:-Flemlsh in particular. · As I have shown, the difference between Northern and.'Southern Dutch can be analyzed as the result· of i a difference in' constraint · order: Van Oostendorp ( 1 995: 1 83, 2000) explains this contrast between (Northern) Dutch and French by exactly the same difference in constraint ranking which I have used here for explaining the contrast in syllabification and vowel deletion between Northern and Southern Dutch, viz. the contrast 9 Apart from the phonological influences of Romance on Southern Dutch there are

a number of syntactic influences, especially on West-Flemish dialects Haegeman, personal communication).

(Liliane

A prosodic contrast between Northern and Southern Dutch 48 1

between the orders: ALIGN » ONSET and ONSET » ALIGN respectively. It therefore seems natural to assume that. the constraint order: ONSET » ALIGN was adopted by Southern Dutch due to the influence of French and/or Picardian. 6.

Remaining issues

It should be pointed out that West-Flemish behaves only partially like French with regard to vowel deletion. For instance, flectional schwa's are not deleted postvocalically, just l ike in Standard Dutch. In Van Oostendorp's analysis this is due to a constraint, MaRFA, which forces every morpheme to get at least a partial realization in prosodic structure. A lso, as we have seen in (6), another vowel then schwa can be also be deleted, unlike what we see in French. 1O According to many analyses (Charette 1 99 1 , Noske 1 993, Van Oostendorp 1 995, 2000), schwa is devoid of featural content and it is this quality that allows for this vowel to be deleted in certain contexts. However, as we have seen, in (6) there is no underlying schwa. The explanation for the fact that there is vowel deletion nevertheless, can be found in the fact that in Dutch, unlike French, unstressed full vowels can reduce to schwa. I I This means in (6) the tIl of ik is first reduced to schwa (i.e., it looses its featural content) and then, as a result of this feature loss, becomes a candidate for deletion. 7.

Conclusion

In this paper, I have treated the hitherto unaddressed issue of differences in behavior between Northern and Southern Dutch with respect to syllabification and vowel deletion. I have shown that the contrast in both . syl labification and vowel deletion between Northern and Southern Dutch boils down to a single difference in constraint ranking. This difference in ranking is exactly the same as the one that was found by Van Oostendorp between (Northern) Dutch and French. This resemblance in constraint ranking between Southern Dutch and French is striking. Therefore, I have advanced the hypothesis that Southern Dutch has undergone influences of 10

In French deletion of other vowels than schwa is possible only in two words, viz.

tu 'you' and the feminine definite article II

lao

The schwas in French are the result of a historical process of reduction of unstressed vowels to schwa which arose under the influence of the Franconian superstrat. This process, however, has long ceased to be productive.

482 Roland Noske nearby Romance dialects like French and Picardian. We have seen that the language contact between Romance dialects in Northern France and Southern Dutch have given rise to a number of other changes in both Romance and Southern Dutch. It therefore does not seem strange to assume that the French constraint order: ONSET » ALIGN is a feature that has crossed the border between French and Dutch.

References Charette, Monik. 1 99 1 . Conditions on Phonological Government. Cambridge: Cambridge University Press. Goossens, Jan. 1 973. De Belgische uitspraak van het Nederlands, De Nieuwe

Taalgids 66/3, 230-240. Grammont, Maurice. 1922. Traite pratique de prononciation [ranfaise. Paris: Lagrave. Noske, Roland. 1 993 . A Theory ofSyllabification and Segmental Alternation. With

Studies on the Phonology of French, German, Tonkawa and Yawelmani. Tiibingen: Niemeyer. Oostendorp, Marc van. 1 995. Vowel Quality and Syllable Projection. Doctoral dissertation, University of Tilburg. Oostendorp, Marc van. 2000. Phonological Projection. A Theory of Feature Content and Prosodic Structure. BerlinlNew York: Mouton de Gruyter. Ryckeboer, Hugo. 1 99 1 . De spontane palatalisatie. Een Nederlands-Picardische parallel? In Handelingen XL V der Koninklijke Nederlandse Maatschappij voor

Taal- en Letterkunde en Geschiedenis, 1 1 3- 1 35. Ryckeboer, Hugo. 1 997. Het Nederlands in Noord-Frankrijk: sociolingulstische, dialectologische en contactlingurstische aspecten. Doctoral dissertation, University of N ijmegen. Ryckeboer, Hugo. 2004. Frans V/aams. Tielt: Lannoo. Schutter, Georges de. 1 999. Fonologische ParaIleIlen aan Weerszijden van de Germaans-Romaanse Taa l grens . Taal en Tongval 5 l , 1 1 1 - 1 30.

The obj ect of verbs like help and an apparent violation of UTAH Christer Platzack 1.

Introduction·

In all the Germanic languages with morphological case, some verbs take their object in dative case, although accusative case is prototypical for objects. In this squib I will discuss one type of such verbs, represented here by English help, a c1ass that also includes verbs like rescue, nurse, serve etc. In the Principles-and-Parameters tradition, these verbs are usually said to take lexical (or semantic) case, whereas verbs with an accusative object, like buy, do not specify any case, the assumption being that accusative is structurally assigned. The difference between structural and lexical case is shown by the well-known fact that lexical case is preserved in passive, whereas structural accusative changes into nominative. See the Icelandic examples in ( 1 ): Honum var hjalpa3. was helped hedat h. B6kin var keypt. booklef.nom was bought

( 1 ) a.

a'. Active: E.g I b'. Active: E.g I

hjeilpaoi honum. hedat helped keypti b6kina. bought bookdef.acc

Some of the Germanic languages, like Dutch, English and mainland Scandinavian, lost morphological case during the Middle Ages. It is usually assumed that the distinction between lexical and structural case is not lost in this process. In this squib I will discuss whether the loss of morphological case has any influence on the structure of the VP headed by verbs of the help-type, mainly using Scandinavian data. My d iscussion will be framed within the Minimalist program, more specifically in a particular implementation of the feature driven account of Chomsky (200 1 ) and Pesetsky and Torrego (200 1 ). The basic theoretical machinery is outlined in the rest of this introduction. Section 2 will present the analysis of VPs headed by verbs like help, comparing the VP structure of languages with .and without morphological �ase. In section 3 I will give some syntactic arguments indicating that VPs headed by verbs like help are structured t

Thanks to Cecilia Falk and Halld6r Sigurosson for valuable comments.

484 Chr;";ter Platzack

differently in these two groups of languages, in particular that the object is in different positions. Unless there is a thematic difference as well, this finding goes against the Uniformity of Theta Assignment Hypothesis (UTAH), as described in Baker (1 997). In section 4, I will highlight verbs like throw that also take dative objects in Icelandic, which like the help­ verbs seem to have different structured VPs in languages with and without morphological case. For these verbs, however, a difference with regard to thematic roles is much easier to envisage. 1. 1. Theoretical prerequisites

1 . 1 . 1 . Features and the operation Agree Syntax is a computational system (an algebra), mediating between form and meaning. Syntactic features have a central role to play in the computation. A syntactic structure is the result of merging elements taken from the lexicon, consisting of features which have either a semantic or a phonetic value. These features may come in two guises, interpretable and uninterpretable. Only interpretable features are allowed at the interfaces, which means that the syntactic computation must delete the un interpretable instances for the derivation to converge. I will use only two features in my presentation tense (t), interpretable in TO and vo, and person, gender and number (�-features), interpretable in DP. Uninterpretable features are deleted with the help of an operation called Agree. This operation first selects a probe, i.e. a head with at least one un interpretable feature -F. The relevant probes are VO and TO. I will assume that V o has the same inherent feature set up for all types of verbs, contra Chomsky but in line with Legate (2003). This probe will search its c-command domain for the closest goal with the same feature but with reversed value for interpretability, +F. Finally, the uninterpretable feature is given the value of the interpretable feature. Sometimes, the valuing of an uninterpretable feature is accompanied by movement of the goal to the position of the probe. Such movement is forced by the presence of EPP on some un interpretable feature, see Pesetsky and Torrego (200 1 ). 1 . 1 .2. UTAH In Baker ( 1 997), UTAH is seen as a universal one-to-one correspondence between thematic role (in a particular broad sense) and first merge: a DP with the role of Agent (Cause, Instrument) is first merged in Spec-vP, a Benefactive, Goal, Recipient or Experiencer is first merged in Spec-VP,

An apparent violation of UTAH 485 and a Patient or Theme is first merged in the complement of V . This is illustrated in (2) with a ditransitive verb:

(2)

vP

DP

v'

I

VP

gave

�

DP

I

MarYgoal Vo

V'

�

gave

DP

� the bookrhcmc

For languages with morphological case, (2) implies a possible correlation between case and position in addition to UTAH: Spec-vP is Nominative, Spec-VP is dative, and the complement of V is accusative. 2

1 . 1 .3 . The derivation of a transitive clause: John ate an apple

·

·

The derivation of a simple transitive clause is outlined in (3). The agree­ relations at hand are indicated by arrows. Note that there are two probes and two goals. John ends up in Spec-TP due to the EPP feature associated with the un interpretable «/l-feature in T (_$EPP) .

2

·

This is a simplification, as can be seen e.g. by the fact that the two objects in lcelandic double object constructions have the following case possibilities: DAT­ ACC (the give-class), DAT-DAT, DAT-GEN, ACC-GEN, ACC-DAT (the retumClass). See Jonsson (1 996: 1 36).

486 Christer Platzack (3 )

CP

�

TP

�

ate [+'t -41] Vo

I

2

an apple

Verbs with dative objects: the case of help

In Platzack (2005 ) I have argued that dative case has the property of making DP-features invisible for probes, an effect that may be seen long after the loss of case morphology. In this way, I am able to account for the fact that both in English and in Mainland Scandinavian morphological dative was lost during the Middle Ages (Allen 1 995, Falk 1 997, Skrzypek in press), but promotion of the recipient (the former dative argument) was not an immediate consequence; as a matter of fact, both in English and in Swedish normative grammarians as late as in the 20th century (Sweet 1 900, Wellander 1 95 9) advice against the promotion of the indirect object to . subject in passives, as in (4): (4)

Mary was given the book.

With features in the indirect object invisible for the probe vo, the direct object must be responsible for the elimination of the uninterpretable $-feature in vo. Verbs taking a dative object as its sole argument are potential problems to such an analysis: with the features of the dative OP invisible for the

A n apparent violation o/ UTAH 487

probe vO� it is not obvious how VO gets rid of its uninterpretable $-feature. Extending an idea from Hale and Keyser ( 1 993), Platzack (2005) claims that such verbs take an invisible cognate object which is incorporated in the verb-an analysis which has a long history in l inguistics, see e.g. March ( 1 87 1 ) and Wessen ( 1 967). According to this idea, dative-taking verbs like German hilfen and Icelandic hjalpa 'to help' are ditransitives underlyingly with the meaning 'provide X with help', where help is the cognate object. Consider the analysis of (5a) in (Sb): (5) a. Hann hjalpar Onnu. he help Ann8dat b. TP

�T ' I �

DP

hann

hialpar[+t _$]

V'

DP

I

Onnu

�DP

VO

I

hjalpar

L

I

he!pr-t +41]

As seen in (5b), this analysis removes the problem with the dative blocking the features of Onnu. 3.

The Structure of Help-verbs in Modern Swedish

3. 1.

Possible analyses

We will now consider the analysis of the corresponding modem Swedish example, where the object is not inflected for case:

488 Christer Platzack

(6)

Han hjalper Anna. he helps . her

Since there is no dative object in (6), we may ask whether the structure underlying (6) is identical to the structure outlined in (5b)-without dative case� the features of the object should be visible to vo, and there does not seem to be any reason to assume an invisible cognate object in the complement of V . There are at least two analyses of (6) available within the framework used here: either the structure in (7), equal to (5b) with the exception that the visible object and not the invisible cognate has the features [-or +<1>], or (8), where the object is in the complement of the verb. (7)

TP

DP� I

han

T'

�vP

T(+-r _�EPP]

� DP� V' I

-RaR[ + _t 41]

�

VP

DP � \ I VO�DP

hjalper[+t -9J

Anna[_t +t\l]

I

fljiilpef

V'

L

I

help

An apparent violation of UTAH 489 (8)

TP

tj8lPcr --"

hjalper[+t .•1

VO

DP

In both (7) and (8), VO is agreeing with the visible DP, which also gets its tense-feature valued by vo. In this respect, there is a clear difference between (7)-(8) on the one hand and (5b) on the other, since VO agrees with the cognate object in (5b). The important difference between (7) and (8) is the position of the visible object: in (7), as in (5b), this DP is in the specifier of VP, whereas it is in the complement of V in (8). Thus, given UTAH, we would expect a change in thematic roles if (8) is correct, whereas no such change is implied by (7). Furthermore, if (7) is the right structure, we expect to be able to find some trace of the invisible cognate object, whereas such a trace should not be visible, given (8). In the remainder of this section, I will provide evidence that indicate that (8) is the structure of present day Swedish (6).

3.2. Syntactic arguments for the structure in (8) There are some syntactic arguments that support the analysis in (8). Consider first examples like (9), and compare with ( 1 0):

490 Christer Platzack

( 9) a.

den dodade mannen the killed man b. det appnade f6nstret the opened window c. det erbjudna jobbet the offered job

( 1 0)

*den erbjudne mannen the offered man intended meaning: 'the man to whom something was offered'

As these examples show, the direct object, but not the indirect one ( 1 0), may correspond to the head of a noun phrase, taking the past participle of the verb as its modifier. In the active VP, this head corresponds to the object in the complement of V (the Theme argument). Looking now at Swedish hjalpa 'help', we notice that its past participle behaves as the past participle of ordinary transitive verbs, as shown in (I la); this is not possible in languages where the cognate of help takes a dative object, as ( 1 1 b,c) show: ( l l ) a.

den hjalpta mannen the helped man b. *hjalpa3i ma3urinn c. *der geholfene Mann

These facts indicate that the structure underlying the examples in (6) presumably is (8), not (7). Another indication that the object of help is in the complement of V in modem Swedish is the following datum. In Swedish, existential passives are usually well formed, both with verbs taking ordinary objects in the active, and with verbs like help. ( 1 2) a.

Det koptes there boughtpassive b. Det hjiilptes there helpedpassive

manga backer. many books en man over gatan. a man across the street

The associate, i.e. miinga backer 'many books' in ( l 2a) and en man ' a man' i n ( l 2b), must be indefinite (the well-known definiteness restriction on existentials). Existential passives are possible with ditransitive verbs as

II

I

An apparent violation of UTAH 49 1

well, but only with the direct object of the active clause as the associate, as shown by the definiteness restriction on this argument, but not on the indirect object: (13)

Det gavs henne presenter/*presentema. there gavepassive her gifts/the gifts

The example in ( 1 3) implies that the associate of a Swedish existential must be placed in the complement of V. Applied to ( 12), this seems to be an argument in favour of the analysis in (8) for verbs of the help-type. This analysis is further strengthened by the fact that Icelandic, which accepts existential passives like ( 1 2a), does not accept the correspondence to ( 1 2b): (14) a. l>a5 vorn keyptar margar brekur. there were bought many books * b. 1 1>a5 var hjalpa5 gomlum manni yfir gotuna. 3 there was helped old man across the street (Maling 1 988, p. 1 80)

Hence, there are indications that the object of verbs like help occupies different positions in languages with morphological case, like Icelandic, and in languages without morphological case, like Swedish. Given UTAH, this difference implies that the object has different thematic roles in the two types of languages. However, at the moment I am not able to see how this difference is manifested, hence the proposed analysis does not seem to be comptatible with UTAH in the form suggested by Baker ( 1 997). In the next section we will consider another type of verbs taking dative objects in . Icelandic and Old Scandinavian, where the change of position of the object

3

[f the associate is placed in front of the participle, the example is well-formed: l>a6 var gomlum manni hjalpa6 yfir gotuna. (Mating 1 988, p. 1 80)

(i)

Notice that Swedish, which also has a passive with a past participle, accepts both the word order with the associate in front of the participle, and the word order with the associate following the participle: Det blev hjalpt en man over gatan there was helped a man across the street b. Det blev en man hjalpt over gatan. there was a man helpt across the street

(ii) a.

492 Christer PJatzack

from Spec-VP to complement of V seems to result In a change of perspective, indicating different thematic roles.

4. The case of kasta 'throw' The verb kasta 'throw' takes its object in dative both in Old Swedish ( l Sa) and in modem Icelandic ( I Sb), whereas in modem Swedish there is no indication of case ( l Sc): ( l S) a. kasta stene (Old Swedish 1 300; Skrzypek in press) throw stonedat b. Kristjan kastaoi sleggjunni. Kristjan threw the-hammerdn, c . Johan kastade sUiggan. Johan threw the-hammer Kasta 'throw' belongs to a group of verbs accompanied by a dative object denoting the tooV object being set in motion, see Jonsson ( 1 996) and Maling (2002). Since the object is in the dative, the first hypothesis is that it occupies Spec- VP; if so, it must express a Recipient of some kind, e.g. the object is assigned momentum by the Agent/the subject. For modern Swedish, there is no indication that the object is anywhere else than in the complement of V, carrying the Theme role, i.e. plainly indicating the moved entity. Syntactic arguments are provided in ( 1 6) and ( 1 7) below. Hence, contrary to the situation with verbs like hjalpa 'help', it seems to be possible to detect a slight change in perspective connected with the loss of dative. In the final part of this section, I will give two syntactic arguments for assigning Swedish verbs of the /casta-type the structure in (8), whereas the corresponding Icelandic verbs display the structure in (Sa). Firstly, parallel to (9)-( 1 1 ) above, only Swedish can use the participle of these verbs as an attribute to a noun corresponding to the dative object:

( 1 6) a. den kastade bollen the thrown ball b. *kastaoi boltinn thrown ball Secondly, assuming that the VP-structure of the throw-verbs differs between Swedish and Icelandic in the way mentioned, we correctly predict

An apparent violation of UTAH

493

that a free 'dative' should be possible with these verbs in Swedish but not in Icelandic, assuming that free 'datives' occupy Spec-VP. Hence in Icelandic, but not in Swedish, the free 'dative' compete for the same position as the object: ( 7) a. Jag kastade henne bollen. I threw her the ball E b. * g kastaoi henni boltanum. threw her the-balldat I 5.

(Swedish) (Icelandic)

Conclusion

In this squib I have argued that VPs of verbs of the help-type and the throw-type have undergone a structural change in connection with the loss of morphological dative, to the effect that the DP expressed as a dative object in Old Swedish, which had its place in Spec-VP, in present day Swedish is placed in the complement of V. Given UTAH, this implies a shift in perspective: whereas the speakers of Old Swedish are supposed to view this object as a kind of Benefactor or Recipient, in present day Swedish it is merely taken as a plain Theme. Such a change may be envisaged with the throw-verbs but is harder to detect with the help-verbs. Given the fact that English has developed in the same way as Swedish with respect to these two classes of verbs, we might expect the same change in underlying structure, thus also for English UTAH predicts a change in perspective with respect to the object of verbs like help and throw. Whereas such a change is visible with throw-verbs, it is not with help-verbs, hence for these verbs UTAH seems to provide the wrong prediction.

References Allen, Cynthia L. 1995.

Case Marking and Reanalysis. Grammatical relations from Old to Early Modern English. Oxford: Clarendon Press. Baker, M ark . 1 997. Thematic roles and syntactic structure. In Elements of Grammar. Handbook in Generative Syntax, ed. by Liliane Haegeman, 73- 1 37.

Dordrecht: Kluwer. Chomsky, Noam. 1 982. Some Concepts and Consequences of the Theory of Government and Binding. Cambridge, MA and London, England: The MIT Press Chomsky, Noam. 200 1 . Derivation by Phase. In Ken Hale. A Life in Linguistics, ed. by Michael Kenstowiccz, 1 -52. Cambridge, The MIT Press. Falk, Cecilia. 1 997. Fornsvenska upplevarverb. Lund: Lund University Press.

494 Christer Platzack

Hale, Ken and Jay Keyser. 1993. On argument structure and the lexical expression of syntactic relations. In The View from Building 20: A Festschrift for Sylvain Bromberger, 53- 108. Cambridge, Mass.: The MIT Press. J6nsson, Johannes Gisli. 1 996. Clausal Architecture and Case in Icelandic. PhD diss., Uuniversity of Massachusetts. Legate, Julie Anne. 2003. Some interface properties of the phase. Linguistic Inquiry 34: 5 06-5 1 6. Mating. Joan. 1 988. Variation on a Theme: Existential Sentences in Swedish and Icelandic. In McGill Working Papers in Linguistics, Special Issue on Comparative Germanic Syntax, 1 68- 1 9 1 . Mating, Joan. 2002. Icelandic verbs with dative objects. Working Papers in Scandinavian Syntax 70: 1 -59. March, F. A. 1 87 1 . A Comparative Grammar of the Anglo-Saxon Language. New York Pesetsky, David and Esther Torrego. 200 1 . Tense-to-C movement. Causes and consequences. In Ken Hale. A Life in Linguistics, ed. by Michael Kenstowiccz, 355-426. Cambridge, The MIT Press. Pesetsky, David and Esther Torrego. 2004. Tense, case, and the nature of syntactic categories. In The Syntax of Time, ed. by Jacqueline Gueron and Jacqueline Lecarme, 495-538. Cambridge, MA: The MIT Press. Platzack, Christer. 2005. Cross-Gennanic promotion to subject in ditransitive passives - A feature-driven account. In Grammar and Beyond Essays in honour of Lars Hellan, ed. by Mila Vulchanova and Tor A. AfarIi, 1 35-1 6 1 . Oslo, Novus Press. Skrzypek, Dominica. in press. The Decline of the Nominal Flection in Swedish: The Loss of the Dative. Lund Sweet, Henry. 1 900. A new English grammar: logical and historical. Oxford: Clarendon Press Wellander, Erik. 1 959. Riktig svenska. En handledning i svenska sprakets vard 3rd edition. Stockholm: Svenska Boktorlaget Norstedts. Wessen, Elias. 1967. Svensk sprakhistoria 1/1. Grundlinjer till en historisk syntax, 2nd ed. Stockholm, Goteborg, Uppsala: Almquist and Wiksell. .

.

J

i

A note on relative pronouns in Standard German Martin Prinzhorn and Viola Schmitt 1.

Introduction

Standard German (SG) allows for two patterns ofN-headed relative clauses (RC), which seem to show no distributional differences. On closer inspection, however, differences in various contexts can be observed. What we will thus try to do, is sketch those differences and outline a hypothesis, according to which it is the nature of the head that is responsible for them. By addressing the concept of semi-lexical heads in the analysis of relative pronouns, we combine two of the many topics for which Henk van Riemsdijk provided pioneering analyses.

2. Patterns of welch- and dN-headed RCs in SG may be introduced by two different relative pronouns, d- and welch- ( 1 ), where d- is morphologically equivalent to the definite determiner, while welch- is morphologically equivalent to the partitive wh­ item (see Heim 1 987). In both cases the RPs inflect for case, which is assigned to the base position inside the RC, as well as for gender and number, depending on the matrix OP. (1)

Der Schwan, der/welcher im Teich schwimmt, ist schwarz. the swan RPnom.masc.sg in-the pond swims IS black

There is no obvious difference in the interpretation of these two examples. Furthe�ore, �s opposed to many of its dialects, where different relative pronouns mark the restrictive/appositive distinction (Brugger and Prinzhom 1 996, Schmitt 2005), Standard German allows for both pronouns in both cases. (2) a.

Jeder Mann, der/welcher schon einmal Genever getrunken every man already once Genever drunk RPmasc.sg (restrictive) hat, ist begeistert. . has is taken-by 'Every m an who has ever drunk Genever is taken by it.'

496 Martin Prinzhorn and Viola Schmitt

b.

Hans, der/welcher ja immer Genever trinkt, sol lte Hans RPmasc.sg PRT always Genever drinks should (appositive) nach Holland ziehen. to Holland move 'Hans, who is always drinking Genever, should move to Hol land.'

In addition, differences with respect to extraposition, stacking and case mismatching, as can be found in dialects (Bayer 1 984), cannot be observed in SG. 3.

Problems

As opposed to the picture given above, there are in fact divergences in the distribution of d- and welch-, which we wi ll review in the following paragraph, testing different hypotheses as to what is responsible for this distinction as we go along. The first difference can be observed in a certain kind of light-headed RC . If there is no distinct contextually salient (or defined) property or individual, l ight-headed RC can only show up with a third kind of RP, which so far has not been discussed, namely was (what) (3a). ' If, however, there is such a contextually given property (like 'sandwich ' in (3 b)), d- but not welch- has to occur (3 b). In cases where there is a contextually salient individual, and where D of the matrix can be thought to be a demonstrative, both RPs are licit (3c). Note, that only in case of (3b) the D of the matrix may be reduced. (3 ) a. Das, was Paul da isst, sieht giftig aus. Paul there eats . looks poisonous PRT the! that RP 'What Paul is eating there looks poisonous.' b. A: I think the sandwiches have gone bad. B: Das! dQs, das/*welches ich gegessen habe, war gut. RP I eaten have was good the 'The one I have eaten, was good. ' c. DAS hier, das/welches ich ja gestern probiert habe, that here RP I PRT yesterday tried have kann ich empfehlen. can I recommend 'This one, which I already tried yesterday, I can recommend.' I German light-headed RC with was -RP certainly need further treatment, since Citko's analysis of light-headed RC (Citko 1 999) does not seem compatible with them.

A note on relative pronouns in Standard German 497 Comparing the (3b) with the data in ( 1 ), it seems that the crucial difference here lies in the presence or absence of an overt predicate that could be modified by the RC. This assumption is strengthened if we look at the quantifiers in (4): if there is a predicate that can be modified by the RC (4a) both RPs are possible, if this is not the case (4b), only d-RP is. (4) a. Jeder Jugendliche, cler/welcher sich flir schlau halt, sollte every teenager RP REFL for smart holds should sich bewerben. REFL apply 'Every teenager who believes himself to be smart should apply. ' der/*welcher sich flir schlau halt, sollte sich b . Jeder, everyone RP REFL for smart holds should REFL bewerben. apply , Everyone who believes himself to be smart should apply.' So we find the same pattern in both cases. Furthermore, distributional differences have so far only been observed in restrictive RC, so that we can put forward a first hypothesis (5):

(5)

In restrictive RC, welch-RP may occur only if a predicate is present that can be modified by RC.

Now we will not only have to find evidence for (5) but we will also have two clarify two things: first of all, whether the presence of any predicate allows for welch-RP and secondly we will have to make precise what is meant by 'being present' . Let us first consider the first question, by looking at some other quantifiers, namely niemand (nobody) and jemand (somebody). Unlike the quantifiers in (4) they never allow restriction by means of N (6a). Since at first sight no predicate is present that could be modified · by the RC, welch-RP should be out, and indeed it is (6b). (6) a. Niemand/ Jemand (*Mann) kommt. nobody/somebody (man) comes b. Niemand/Jemand der/* welcher eingeIaden wurde hat nobody/somebody RP invited has was abgesagt. declined 'Nobody/somebody who was invited has declined. '

498

Martin Prinzhorn and Viola Schmitt

However, while niemand and jemand do not allow for N-restriction, they do allow for A-restriction. Since a predicate is present now that could be modified by the RC, we would expect welch-RP to be possible. Yet it is not (7): Niemand/Jemand GroBes den/*welchen nobody/somebody big RP kommt zu deinem Fest. comes to your party

(7)

du you

kennst know

This indicates that we have to specify the hypothesis. We could, for instance, refonnulate (5) as (8): 2 In restrictive RC, welch-RP may occur only if an N is present, that can be restricted by the RC.

(8)

(8) seems to be borne out if we look at data like (9) where the matrix DP, apart from the RC, contains only an adjective and where welch-RP may not occur. (9)

Der GroBe, der/*welcher da hinten steht, ist mein Bruder. there back stands is my brother the big RP 'The big one who is standing there is my brother.'

2 The observation, that N must not only be present but also restricted by the RC comes from the observation of amount RC. Here, amount reading is only possible with d-RP (a), while welch-RP (b) forces an identity-of-objects reading (cf. Grosu and Landmann 1 998) (meanings of RC given below). We can test this by inserting the particle so (. in average) in RC, thereby forcing an amount reading . .In those . cases, welch-RP is out.

(i)

a.

Das Geld, das/*welches die Schweizer haben, werden wir nie the money RPneu.sg will . we never the Swiss have haben. have 'We will never have (the amount of) money that the Swiss have.'

The same holds for relativized idiom junks, where the RC is considered an amount RC. Idiomatic reading is only possible with d-RP and not with welch-RP. Since, amount-RC are generally thought to denote degrees (Carlson 1977, Heim 1 987) they do not restrict the apparent head noun but rather operate on a degree-variable. We may therefore conclude, that the condition as to when welch-RP may occur must include that what is restricted by the RC is N.

A note on relative pronouns in Standard German 499

The question now is whether what we have cal led 'A' is indeed of category A and not an N. If we assume this (relying on the fact that A inflects as though an N was present), we would make the prediction that (9) contains a silent N. Then the only reason why welch- could not occur in (9) would be that N is silent. This is backed by the data in (3). As we could see in this example, there seemed to be a relevant difference, expressed by the choice of RP (was vs. d-) between light-headed RC where a property was contextually provided (d-RP) and where it was not (wa�-RP). This difference is reflected by the fact that in (3 b), the D of the matrix might be reduced, while in (3a) or (3c) it may not. This indicates that the DP in (3b) contains a noun and thus we could assume a silent N to be present, whilst in (3a) and we would think it to be absent. Thus, we should maybe rephrase (8) as ( 1 0): ( 1 0)

In restrictive RC, welch-RP may occur only if RC-modified N is there and not silent.

Now, if the distribution of welch- really were a case of phonetic presence of N, we would expect all the relevant contexts to be DPs or QPs that contain an N-position which may be filled phonetically but does not have to be. If it is filled phonetically, welch-RP may appear, if not, it must not. Examples (3), (4) and (9) go along with this hypothesis. However, we have already come across one example that does not behave along these lines, namely (6). As shown in (6a), the QPs niemand and jemand, do not contain an N slot that can be optionally phonetically filled, which means that the N­ position is probably not available because it is already occupied. This claim is supported by the observation that adjectives in cases like (7) can show only one inflectional pattern. Thus we might assume, as has been done before,3 that the N part of niemandl jemand raises from N and is incorporated in the Q-head. But then there is no case of silent N and therefore ( 1 0) cannot be maintained. Further evidence against ( 1 0) comes from other cases, where N cannot be silent because it can never be non­ silent, l ike, for instance in the case of wh-words. In SG, wh- words can appear either as wh-elements that have undergone wh-movement ( 1 1 ) or else as specific indefinite pronouns ( 1 2). In both cases they do not allow for overt N-restriction ( I t a) and ( l 2a) and in both cases welch-RP is not possible (cf. ( l I b) and ( 1 2b».

3

See Emonds 1 976 for English quantifiers somebody/nobody.

500 Martin Prinzhorn and Viola Schmitt ( l l ) a. *Wen Mann findest du toll? who man consider you great b. Wen, der/*welcher morgen kommt, findest du who RP tomorrow comes find you 'Who that will come tomorrow do you consider stupid?'

bifid? stupid

( 1 2) a. *Ich habe wen Mann getroffen. I have wh-pron man met b. Ich habe wen getroffen der/*welcher dich kennt. I have pron. met RP you know ' I have met somebody who knows you. ' Another piece of evidence that (10) is on the wrong track stems from cleft­ constructions. In SG, those may only occur with d-RP. Debatable as their analysis may be (cf. Higgins 1 973), what is clear is that the restrictive RC in the cleft construction in ( 1 3) cannot be thought to modify a si lent N, because there is simply no space for it: ( 1 3) a.

Es war HANS, der/ *welcher die Leiche fand. the corpse found it was Hans RP 'It was HANS, who found the body.'

In sum, all we know for sure is that something of category N must be present for welch-RP to appear, but so far we have fai led to determine what 'presence' means. There are two types of examples where welch-RP is excluded. One type appears to contain a silent N (3b), or is at least compatible with such an assumption (4b). The other type does not allow for a silent N, since the N position is occupied « 6), ( 1 1 ), ( 1 2» , probably by the trace of N. So we basically end. up with two different structures that do riot allow for welch-RP ( 1 4): ( 1 4) a. [DP [D' [d-] [NP [N' [N (N) ]]]]] b. [QP [Q [Q + Nil [NP [N ' [N tj ]]]] ] '

The question is what they have in common. Since we assume the relevant level to be LF (with regard to the fact that distributional differences also occur in amount-RC (see note 2» , we can make the question somehow more precise, by asking whether there are parallels between what we find in N position in ( 1 4a) and ( 1 4b) at LF. To investigate this, let us first look at the examp les represented by ( 1 4b). If there is an element that originates in

A note on relative pronouns in Standard German 50 I

N position, we should find Q or WH restricted in some way, and indeed it

is. Let us first look at the QPs like niemand (nobody). The domain of Q is restricted insofar, as it only contains humans. This is exemplified in ( 1 5): ( I 5) A : Auf dem Bett liegt niemand. on the bed lies nobody B: #That is false, there are cushions lying on the bed.

The same holds for the wh-words we have encountered. In cases l ike ( 1 1) or ( 1 2 ) their set is the case of humans (examples om itted here for reasons of space). 4 In the case of niemandljemand it even seems possible to identify the N-part morphologically as mand which is historically related to the pronoun man (one). Interestingly, the pronoun man shows the same restrictions, insofar as it cannot refer to non-humans (16): ( 1 6) A: Heute steht man auf dem Tisch today stands one on the table B: # This is untrue, no vase is on the table. If man really were a pronoun and thus an individual variable, it would be strange for it to restrict a Q. SO man is a very strange pronoun, if it is one at all . Syntactically, at first sight, it behaves like other pronouns, insofar as it can appear as an argumentS and can be antecedent to reflexives. Just like other pronouns, it cannot be modified by A and cannot occur with D. Semantically, however, it is a non-trivial object. Although it is at first sight not referential, it does not behave like an individual variable, insofar as it does not really seem to depend on an assignment and it cannot be bound. In addition, it cannot bind non-reflexives. What we would like to claim here is that man is a nominal pronom inal. It is pronominal with respect to distribution and because it does not refer, and it is nominal singe it cannot display
We are omitting the wh-word was here, because its treatment needs more space (see Heim 1 987). S For some reason, it may only occur as a syntactic subject.

502

Martin Prinzhorn and Viola Schmitt

( I 7) A : There are many women in this room. B: N iemand lacht. noone laughs A: But the man over there is laughing. So what we would assume for those quantifiers is that at LF N is filled by a nominal pronoun (which might be parallel to the semi-lexical heads introduced by Van Riemsdijk 1 995), and although we will not present evidence for this, we will, on the basis of the parallels observed above, put forward the speculation that the same holds for the wh-words in ( 1 1 ) and (12).6 Let us now look at ( l 4a). The question here is different: N seems to be non-empty, but we find no morphological indication of what it could be. Should we assume that (3b) really contains the lexical N sandwich at LF (and that for some reason it is not spelled out)? On the basis of our data we will argue that it is not, for if sandwich really were present at LF, welch-RP should be possible. That would mean that in some respect silent N is not equivalent to non-silent N. In ( I S) we find further syntactic evidence for th is. Taking the context of (3b), where we talk about sandwiches, the cardinal determiner ein should look identical in ( I Sa) and ( 1 8b), if sandwich was present in ( I Sb), but it does not: ( I S) a.

Ein Sandwich, das ich gegessen habe hat komisch geschmeckt. a sandwich RP I ate have has strange tasted 'One sandwich that I ate tasted strange. ' b. Ein*(-es), das ich gegessen habe, hat komisch geschmeckt.

So we can conclude that sandwich itself is not present. What seems likely is that there is a variable which is assigned a value by the context. But since we found evidence, that it somehow needs to be present syntactically (3 b), we need to find out what the syntactic representation of this variable is. One thing that is clear is that it needs to be able to bear
Whenever a specific set is given, the partitive wh-word preferred.

6

welch-

seems to be

A note on relative pronouns in Standard German 503

( 1 9)

keine Freude mehr kennen. Er, der/*welcher siindigt, soll more know he RP sins should no joy ' He who sins shall not experience happiness anymore.'

So, the two different sets of data could be brought together along the following lines. Both appear with what seems like a nominal pronoun, which does not refer to a specific property. While this pronoun can be contextually bound in the cases like (3), it cannot be bound in cases like (6). Thus, the criterium for the occurrence of welch-RP can be given as follows (but does not yet account for clefts): (20)

In restrictive RC, welch-RP can occur only if the N modified by RC is lexical ( = non-pronominal).

4. Conclusion

In this paper have we tried to show that there are distributional differences in the case of RPs in SG. We further tried to outline that the reason for this could lie in the presence of N, where presence of N means that N must not be pronominal. To relate this to the internal properties of the welch-RP is a matter of future research. References Bayer, Josef. 1 984. COMP in Bavarian Syntax. The Linguistic Review 3 : 209-274. Brugger, Gerhard and Martin Prinzhom. 1 996. Some Properties of German Determiners. Ms, Universitat Wien Carlson, Greg. 1 977. Amount Relatives. Language 53, 520-543 Citko, Barbara. 1 999. Light-Headed Relatives. Working Papers in Linguistics 6. Pennsylvania: University of Pennsylvania. Emonds, Joseph. 1976. A Transformational Approach to English Syntax : Root, Structure-Preserving and Local Transformations. New York: Academic Press Grosu, Alexander and Fred Landman. 1 998. Strange relatives of the third kind. Natural Language Semantics, 6, 125-1 70. Heim, Irene. 1 987. Where does the Defmiteness Restriction Apply? Evidence from the Definiteness of Variables. In Eric J. Reuland, Alice G.B. ter Meulen (eds.) The representation of (in) definiteness. Cambridge MA: MIT Press, 2 1-42 Higgins, Roger. 1 973 . The Pseudo-Cleft Construction in English. PhD Dissertation, MIT Riemsdijk, Henk van. 1 998. Categorial feature magnetism: The endoc€mtricity and distribution of projections. The Journal of Comparative Germanic Linguistics

2: 1 -48

504

Martin Prinzhorn and Viola Schmitt

Riemsdijk, Henk van. 2005. Silent Nouns and the Spurious Indefinite Article in Dutch. In: Mila Vulchanoya and Tor A. A farli (eds.) Grammar & Beyond. Oslo: Novus, 163- 1 78 Schmitt, Viola. 2005. Hessian relative clauses. MA-thesis, Univ. Wien.

Agreeing to bind Eric Reuland 1.

Background I

Much of the linguistic research since the fifties focused on acquiring an understanding of language through the detailed study of individual languages and the types of variation they exhibit. The requirement of explanatory adequacy served as an important constraint on theorizing. No rule or parameter could be admitted that would be incompatible with leamability of the class of grammars embodying them. This line of research led to a great amount of insight into the structure of natural language, and to a considerable number of substantive universals, occasionally abstract, yet precise (see Everaert and Van Riemsdijk, 2005, for a fascinating collection of results). In the eighties, the strengthening of the ties between cognitive science and linguistics led to novel questions. As observed in Chomsky (200 1/2004), for the first time did it become feasible to seriously address the question of how language is rooted in man's cognitive system, thus moving beyond explanatory adequacy as earlier conceived. It is a merit of connectionism, and cognitive grammar that they put this issue to the fore. However there is little reason to expect substantial progress from these directions. Connectionist proposals are generally marred by linguistic naivete, and the potential of cognitive grammar tends to be limited by its reliance on high-level functional concepts and by ignoring the underlying computational mechanisms. If one wishes to explore how language is rooted in man�s cognitive system, the minimalist program as conceived in Chomsky ( 1 995) and subsequent work is an important step forward, due to its attention to the 1 Dear Henk, this squib is for you

as a token of friendship and of appreciation of

what you contributed to the Dutch linguistic community and to linguistics worldwide. You are a binding force who does not agree to be bound. You will undoubtedly continue to contribute, but let us also be clear about what this occasion means, and strengthen our detennination to fight the blind forces of market ideology in education. Support by the Netherlands Organization for Scientific Research NWO is gratefully acknowledged (grant Dr. NV -04-09).

506 Eric Reuland

details of the computational system underlying human language, and is systematic conceptual parsimony. If no principle is to be accepted as a principle of language unless shown to be inevitable for empirical reasons, this has a strong impact on the quest for universals. It forces one to assess all macro-universals that resulted from previous theorizing, under the presumption that, if correct, they should be made to follow from elementary properties of the computational system and the way it operates on the primitive elements of the vocabulary of language. Against this background I will briefly discuss an issue that comes up in the theory of binding. 2.

Binding

In the canonical binding theory (Chomsky 1 98 1 ), henceforth CBT, arguments are classified as R-expressions, pronominais, or anaphors. This classification is based on two designated features: [±anaphoric] and [±pronominal] . Binding theory is concerned with relations of anaphors, pronominals, and R-expressions to possible antecedents (Chomsky 1 98 1 :6). R-expressions are referentially independent in the sense that they cannot be bound, pronouns may, but need not be bound, and anaphors cannot be interpreted independently, hence they must be bound. The notion of binding itself is defined in terms of two ingredients: co-indexing and c­ command, as in ( 1 ). (1)

a

binds b iff a c-commands b and a and b are co-indexed

The fundamental fact to be captured by any binding theory is that anaphors and pronominals impose specific locality conditions on their binders. A binder of an anaphor may not be 'too far away', the binding of a pronominal may not be 'too nearby '. One of the important issues is how these locality conditions are to be captured. (2) presents the binding conditions of CBT, with governing category as defined in Chomsky ( 1 98 1 ): (2)

Binding Conditions (A) An anaphor is bound in its governing category (B) A pronominal is free in its governing category (C) An R-expression is free

The binding conditions in (2) are typical examples of what I earlier called macro-universals. In the light of subsequent developments the conditions as formulated here still constitute a surprisingly good first approximation.

Agreeing to bind 507

Yet there are important reasons to rethink the CBT. One of these reasons is that in the course of the late eighties and beginning of the nineties the central status of the binding theory within grammatical theory as an ingredient of the theory of movement and of empty categories got lost, a development that culminated in the copy-theory of moment, which provided a very different perspective on empty categories. Furthermore a range of empirical challenges was uncovered, see e.g. Everaert ( 1 986), and the various contributions in Koster and Reuland (199 1 ) for overviews. 3.

Rethinking the canonical binding theory

Reinhart and Reuland ( 1 993) provide a number of empirical reasons for a modular approach to binding, separating conditions on reflexivity as a property of predicates from the effects of conditions on syntactic chain formation. In both areas cross-linguistic variation necessitates developing analyses at a much lower level of granularity than before. Finely grained properties of computations play an important role in accounting for the many facts about anaphor-binding (condition A) that have been uncovered over the last decades, such as the contrast between simplex anaphors (SE­ anaphors) and complex anaphors (for instance. Dutch zich, Norwegian seg, Icelandic sig, versus Dutch zichzelf, Norwegian seg selv, Icelandic sjalfan sig, to mention a few cases in Germanic), the role of reflexive c1itics in Romance or Slavic, verbal markings of reflexivity in languages as diverse as Georgian, Telugu and Sakha, etc. Equally important is the logophoric interpretation of free anaphoric forms in languages varying from English and Iceland ic on the one hand, to Chinese and Japanese on the other side of the spectrum. Also canonical condition B effects show an unexpected amount of cross-linguistic variation (locally bound pronominals versus anaphors in Dutch versus Frisian and across German dialects, see, for instance, Reuland and Reinhart 1 995). In view of such diversity, features such as [±anaphoric ] and [±pronominal] lose there empirical basis as theoretical primitives. It seems one has no choice but either accept a great variety of idiosyncratic variation or else rethink the fundamentals of CBT, and deconstruct its macro-universals. Rather than stipulating that an element is an anaphor or pronominal with concomitant binding properties, its behaviour has to be explained in terms of its morpho-syntactic feature composition, and the way in which the computational system makes these features interact with the environment. Thus, the interpretive properties of R-expressions should follow from their lexical and grammatical features. The properties of pronominals (/, you, he, etc.) should fol low from the fact

508 Eric Reuland

that they are only specified for person, gender and number (the -features) and grammatical category. Similarly, the properties of complex anaphors should follow from the properties of their composite parts, and the properties of typical SE-anaphors from the fact that they are underspecified for cI>-features (see below). A significant step forward is Chomsky's ( 1 995) proposal that syntax obeys the inclusiveness condition. It consists of elementary operations such as attract, check/agree, copy, and erase applying to a strictly morpho­ syntactic vocabulary. Indices are not morphosyntactic objects. Hence, they have no place in the grammar. 2 However, eliminating them leaves us with the question of how encode the type of relations indices used to. There are two options. The computational system of human language, CHL, is a mechanical combinatorial system of Lexical Items (LI 's), defining a mapping between representations in two language-external systems, the sensori-motor system and the interpretation system as in (3) (Chomsky 1 995). CHL can access properties of the sensori-motor system only in so far as these are legible at the Phonetic Fonn (PF) interface, and properties of the interpretation system only in so far as these are legible at the Conceptual-Intentional (CI) interface.

(3 )

CI

PF

Sensori-motor system -dedicated

L I' +dedicated-JI C HL

Interpretation system (IS) -dedicated

Disregarding the possibility that binding is a PF phenomenon, both the C-I interface (involving the IS system), and CHL are possible candidates. Chomsky ( 1 995) proposed that binding conditions apply at the C-I interface. However, as shown in Reuland (200 1 ), there is . a syntactic residue to binding. One important reason resides ill locality . . Locality is the hallmark of syntactic operations such as movement and ;a.g�ee ..> .' .". . The abundance of cases with binding into syntactic islands, such as the adjunct in (4), shows that variable-binding relations are not generally subject to locality:

(4 )

Every linguist should resist when faceless forces threaten his field.

2 Given the problematic status of indices as hybrid objects, not quite syntactic, and

not quite semantic this is an independently welcome step.

Agreeing to bind 509

Since binding conditions A and B, however conceived, do reflect locality, one must either duplicate locality conditions at the C-I interface specifically for binding, or else explain the locality conditions on binding as consequences of syntactic computation. 3 So, binding must be deconstructed into a C-I-component expressing variable binding in general 4 , and a syntactic component capturing locality (both as regards reflexivizing operations and as regards non-reflexivizing anaphors), along the lines set out in Reuland (200 1 ). In order to avoid duplicating (parts of) binding theory, CHL should contain no statement specific to binding. In current terms, the binding dependencies expressed in CHL should ideally be m echanical consequences of probe-goal relations, embodied in AgreelMove, with probes ideally restricted to phase heads (Chomsky 2005). Locality conditions on binding should follow from locality of M ove/Agree. 4. Agreeing to bind

These issues are addressed in Reuland (200 1 ) by analyzing binding of SE­ anaphors (such as Dutch zich, Icelandic sig, etc.) in terms of the system of Chomsky ( 1 995). Cognate pairs of formal features, for instance interpretable versus un-interpretable -features, enter into checking configurations. Members of a cognate feature pair are erased up to recoverability. On the basis of this recovery relation checking/erasure relations mechanically establish a formal dependency between SE-anaphors and their antecedents, but not between pronominals and their antecedents. Where syntactic encoding of the dependency between an anaphor and its antecedent is possible, economy rules out using the pronominal, thus deriving complementarity between bound pronominals and anaphors where it obtains. This derivation made use of specific mechanisms in Chomsky ( 1 995), including chain composition, formal feature movement, checking and deletion/erasure. In subsequent work, Chomsky (200 1 , 2005) developed a more parsimonious system, dispensing with feature movement, and relying 3

The cross-linguistic variation in local binding of 3rd person pronominals in Dutch versus Frisian and across German dialects is governed by a purely syntactic factor such as Case, see Reuland and Reinhart ( 1 995), 4Reflected in the logical-syntax based definition of binding given in (i) (Reinhart 2000): (i)

a

A-binds P iff a is the sister of a A-predicate whose operator binds p

"1

I

5 1 0 Eric Reuland

on the more basic notion of Agree. Fulfilling a promissory note of Reuland (200 1 ), I will now show that the dependency of SE-anaphors can be encoded in such a more parsimonious system as well . To this end I will explore the theory of Pesetsky and Torrego (2004a,b, henceforth P&T), and specifically their approach of feature valuation as feature sharing, and their analysis of structural Case as unvalued Tense. As is uncontroversial, SE-anaphors such as Dutch zich, Icelandic sig, etc., are not specified for the features number and gender. They are specified for person. Reuland (to appear) discusses various types of underspecification and concludes that SE-anaphors carry number and gender features that are unvalued in the sense of Chomsky (2001 , 2005) and P&T. Consider now sentences like (5), using ECM for a context where theta­ properties of predicates do not interfere with anaphor selection. (5)

Iedere professional voelde [zjch aan de kant geschoven] . Every professional felt himself to the side pushed

The question is how zich is related to iedere professional. P&T' s framework makes the following specific claims about feature values: •

o

•

Both interpretable and uninterpretable features come as valued and unvalued Agree involves valuation and feature sharing � a feature chain with one valued instance is valued and every feature must end up with at least one interpretable instance Structural Nominative Case on the external argument DP (EA) is unvalued T

Consider now the more abstract structure in (6): (6)

[Tns [EA [y* [ Y SE .... ]]]]

A Subject-V*-T-dependency is established by the following steps: •

•

Tns's unvalued interpretable T-feature probes and finds EA's uninterpretable and unvalued T-feature � Agree � link Tns 's unvalued T-feature probes again and finds V*'s valued uninterpretable T-feature --+ Agree --+ valuation of T on Tns and Subject

Agreeing to bind 5 1 1

Note that the Subject-verb dependency is encoded without depending on '-feature agreement'. On the basis of these steps CHL expresses a dependency between SE and its antecedent mechanically under the following conditions: •

•

• • • •

•

•

•

SE-anaphors have unvalued interpretable -features in addition to unvalued un interpretable structural accusative Case V*'s (object) EPP feature probes and finds SE as a goal -+ SE moves to the edge of V* Tns has unvalued un interpretable €l>-features Subject DP has valued interpretable tl>-features V* has unvalued uninterpretable -features The Tns-V*-DP T-dependency established in (6) extends to a lP­ feature dependency A -feature dependency cannot be computed independently since we have Tns [u] , EA [v�], V* [u€l>] � Tns's first probing would be successful leaving V* unlinked. In the configuration Tns [u], SE [u] , EA [val], V* [u<1>], Tns [u<1>] probes and finds SE [u] -+ Agree -+ link The remainder of the derivation follows with the proviso stated that the T-dependency extends to the full �-feature dependency.

The dependencies are summarized in (7), with EA providing the required valued and interpretable instance of [€l>] : (7)

[TnsucD [SE UI� [EA valflJ [V* utll [ V (SE lieD)

••••

]]]]]

In order to interpret this syntactic dependency as a binding relation, the following assumption about the format of Lexical ltems suffices, with p the instructions for pronunciation, g the instructions for grammatical computation, and i the instructions for interpretation. (8) a. Fonnat of Lexical Items: LI = { p,g,i} b. Fonnat of SE {p,g, u�} c. Valuing SE -+ {p,g, val �EA } =

After valuing SE, the 'instructions for interpretation' of SE are the same as for the -features of EA. This entails that the dependency will be interpreted as a binding relation. If binding of SE-anaphors is syntactically encoded as Agree, it is expected that the requirement that the binder c-command the bindee is

5 1 2 Eric Reuland

derivative of the requirement that the SE-anaphor be a target for a (c­ commanding) probe. Thus, one should find cases where the bindee can be probed by a relevant head, but its eventual antecedent does not c-command it. 5 As pointed out in Chomsky (2005), the relevant check is provided by sentences with a nominative complement of an un-accusative or passive verb, that has not raised to [Spec, IP] position due to the presence of an expletive in [Spec, IP], and with a 'subject-oriented' reflexive (R ) in a PP that must have been merged after and hence higher than the complement, as sketched in (9): (9)

EXPL Tns [[V DPNOM ] [pp P R . . . ]]

Here the probe Tns c-commands R, but not DPNOM• As shown in ( 1 0), 6 Norwegian and Icelandic provide precisely such configurations. ( l O) a. Det hie introdusert en mannj it became introduced a man b. Thad kom mai1uri med bamin There arrived a man with children

for segi selv/*hamj selv to himself (Norwegian) (Icelandic) sin/*hansi SIG

As discussed earlier, the possibility of syntactically encoding the dependency with seg/sin, blocks binding of the corresponding pronominal. By way of conclusion we can say that the elimination of indices provided the impetus to reintegrate the syntactic part of binding with the core properties of the grammar. Although our discussion has necessarily been limited, I trust it is easy to see from an account like this one, that small changes in feature composition or small changes in structure may have considerable impact on binding options. Thus, even where macro­ variation is observed, one should look for the details. References ·

1 98 1 . Lectures on Government and Binding� Dri�dre�ht: Foris Chomsky, Noam. 1 995 . The Minimalist Program. Cambridge, Mass.: MIT Press Chomsky, N oam . 200 1 . Derivation by Phase. In Ken Hale: a Life in Language, ed. Chomsky, Noam.

Michael Kenstowicz. Cambridge, Mass.: MIT Press

S l am grateful to Noam Chomsky (p.c.) for bringing this consequence to my attention.

6

Thanks to 0ystein Nilsen. Halld6r Sigurdsson and Sigga Sigurj6nsd6ttir (p.c.) for

providing me with these facts.

Agreeing to bind 5 1 3 2004. Beyond Explanatory Adequacy. In Structures and Beyond. The Cartography of Syntactic Structure, Vol 3, ed. Adriana Belletti. Oxford:

Chomsky. Noam.

Oxford University Press.

Chomsky, Noam.

2005. On Phases. Ms. MIT 1 986. The Syntax ofRejlexivization. Dordrecht: Foris Everaert, Martin, and Henk van Riemsdijk (eds.) 2005. The Blackwell Companion to Syntax, volumes I-V. Oxford: B lackwell Koster, Jan, and Eric Reuland. 1 99 1 . Long Distance Anaphora. Cambridge: Everaert, Martin.

Cambridge University Press

Pesetsky, David and Torrego, Esther Syntactic Categories. In

2004a. Tense, Case, and the Nature of The Syntax of Time. ed. Jacqueline Gueron and .

Jacqueline Lecarme. Cambridge: MIT Press.

Pesetsky, David and Torrego, Esther.

2004b.

The Syntax of Valuation and the

Interpretability of Features. Ms. MIT and UMasslBoston.

Reinhart, Tanya.

2000.

Strategies of Anaphora Resolution. In

ed. Hans Bennis, Martin Everaert and Eric Reuland,

Royal Netherlands Academy of Arts and Sciences

Reinhart, Tanya and Eric Reuland.

1 993.

Reflexivity.

720

Interface Strategies, 295-324. Amsterdam :

Linguistic Inquiry 24: 657-

200 1 . Primitives of Binding. Linguistic Inquiry 32, 439-492. How Are They Encoded? Towards a Derivation-Based Typology. Ms. Utrecht University Reuland, Eric and Tanya R einhart 1995. Pronouns, Anaphors and Case. In Studies in Comparative Germanic Syntax, ed. Hubert Haider, Susan Olsen and Sten Vikner, 241 -269. Dordrecht: Kluwer. Reuland, Eric.

Reuland, Eric. To appear. Anaphoric Dependencies:

.

Positive polarity and evaluation Hendrik C. van Riemsdijk 1.

It has quite often been observed that most languages, and maybe even all of them, have formal ways of characterizing certain clauses as positive as opposed to negative. ) Normally, it is the absence2 of negation that indicates positive polarity. There are, however, a number of emphatic, contrastive, and -as I will suggest below- evaluative constructions in which we find traces of a positive element that I will argue to be present in the underlying representations3 of positive sentences. In section 2, I will present some arguments that show that Dutch favors an analysis according to which POS and NEG are treated analogously in the base component. 4 Dutch provides us with a quite ideal case because positive polarity is very often expressed formally and this formal expression turns out to be very coherent: in most cases, the word wei fulfils this role. It seems reasonable to say that there is hardly anyth ing more elementary than positive clauses. Nevertheless, the study of this phenomenon has been neglected in the linguistic literature. Why is th is so? In Transformational Grammar, the study of patterns in which negation is ) This paper was written in French as a term ' paper when Henk was visiting the Universite de Paris VII, Vincennes in 1 970. The original title was Positivite et evaluation. The author name used here corresponds to the name used in the manuscript. At that time Henk was affiliated to the Department of General Linguistics at the University of Amsterdam. This translation is produced by Norbert Corver and Jenny Doetjes (with some typographical assistance by Hans Broekhuis) on the basis of a copy of the original manuscript. 2 Intuitively, it would perhaps be more natural to say that it is the presence of negation that indicates negative polarity. J I will have little to say here about the exact nature of those underlying representations and, in particular, about the relations between positive polarity perfonnatives and factivity. 4 This solution requires rather complex rules that delete POS in the majority of cases. It is clear, though, that the selectional restrictions one would need in order to express the relation between POS, analyzed as an adverb, and the rest of the clause would be very complicated as well.

i

Positive polarity and evaluation 5 1 5

not explicitly present started only recently. For positive clauses, the absence of an explicit marker is the nonnal situation. The cases in which a trace of positive polarity is present are rare and they are consequently characterized by marginal syntactic and semantic phenomena regarding positive polarity as such and by a great heterogeneity in the forms in which it manifests itself in different languages. To give a few examples: very often POS is realized by what is etymologically the adverb of bon or beau (bien in French, schon in German, wei in Dutch etc). In English, one often finds the (auxiliary) verb do when no other modal verb (can, should, etc.) is present. In certain English dialects, so and too (=I- also) sometimes function as positive particles. In many Germanic languages, it is possible to place a contrastive accent on the auxiliary if it is present; otherwise the accent is located on the main verb. In section 2, I will illustrate the main characteristics of the behavior of the positive element wei in Dutch, and in section 3 I will discuss some strange facts regarding the distribution of wei in combination with predicative adjectives and some consequences of these facts for the semantics of adjectives. 2.

The peculiar behavior of weI sketched out in this section can partially be interpreted as arguments in support of a coherent treatment of POS and NEG, i .e. against an analysis of wei as an adverb. 2. 1.

Generally, wei is attested in discourse -monologues or dialogues- of the following type: (I)

aP # - a

{� } )

Explanation: • if (l = wei (POS) then - (l = niet (NEG) if (l = niet then - (l = wei • # designates the place where the optional change of speaker (i .e. tum-taking) takes place. Nonnally, this is a boundary between two sentences. • p and q are different propositions but they are connected to each other by the fact that they can enter discourse together in order to form (part of) a "sensible" message.

5 1 6 Hendrik C. van Riemsdijk •

The brackets indicate the possibility of reducing the redundant parts of the second clause (by means of the transformation Conjunction Reduction)

If one of the two parts of the schema is missing, the part that is present presupposes 5 a content that is syntactically isomorphic but contrasting in the value o a and -optionally- in the part that corresponds to the scope of POSINEG in the present part.

f

wei

(2)

Softenon is niet gevaarl ij k. Het is Softenon is not dangerous It is aff. ' Softenon is not dangerous. But it is dangerous. '

(3 )

je bent niet competent voor taalwetenschappelijke Of Either you are not competent for l inguistic vraagstukken, en dan moet je je mond houden, questions and then must you your mouth keep-shut of je bent het weI. or you are it aft: 'Either you are incompetent in solving linguistic puzzles, and, if so, you should keep your mouth shut, or you are competent. '

(4)

Jan is Jan is

{ �} { } ni t weI

not

aff.

{Wei} can {aff.}

aardig, maar hij kan nice,

but

he

niet not

gevaarlijk. dangerous

lekker koken. deliciously cook.

(i): 'Jan is not nice, but he can cook deliciously.' . (ii): 'Jan is nice, but he. can't cook deliciously.'

:; It is maybe a little dangerous to use the tenn 'presupposition' here given the fact that it is impossible to detect it with the help of the test based on the constancy of presuppositions in the positive and the negative form of a sentence. Ducrol's ( 1 968) research has shown, however, that there are apparently more exceptions to this test.

Positive polarity and evaluation 5 1 7

De bediening is wei bij de prijs inbegrepen. The service is aff. with the price included (Maar de BTW niet.) But the V .A.T. not 'The service is included in the price. (But the V.A.T. is not.)'

(5)

2. 2.

There are certain uses of wei that do not fit well into schema ( 1 )

.

2 .2 1 . .

Wei is often used in an ironical way. This use is explained by the fact that one does not accept the presupposition that results from the deletion of part of the schema. More in particular, one can play with the possibility negation has to presuppose an isomorphic positive sentence, all other elements being variables. One of the variables is replaced by a word which finds itself at the same place in the sentence, but which creates in fact the opposite of a contrast: a reinforcement.

helpt ook geen tluit. (6) A: De pH The (contraception) pilI helps also not-a tlute B : Jouw kondooms wei zeker? Your condoms WEL certainly 'The pill does not help at all . Your condoms are surely much better.' 2.2.2. With a particular intonation, which tends to an accentuation of all the words of the sentence, the presupposition disappears. The natural response to this type of sentence is the repeated affirmation of the same contents. A contrastive, negative response becomes polemic then: (7)

De invasie in Laos is wei een vuile zet, zeg. The invasion in Laos is WEL a dirty move, say. 'The invasion of Laos really is a dirty move?'

2.2.3. There is another construction featuring wei, unaccented this time, which escapes the schema in ( 1 ). This construction will be discussed in section 3.2.

5 1 8 Hendrik C. van Riemsdijk 2. 3.

2.3 . 1. Whatever their derivation, the coordinated structures very clearly show the parallelism between pas and NEG. When the contrast is marked by pas and NEG, the conjunction maar (' but' ) is often superfluous. (8)

Jan gaat niet naar huis, (maar) Piet wei Piet aff. Jan goes not to home, but 'Jan doesn't go home, but Piet does. '

(9)

Jan gaat niet naar Polen, (maar) [i.e. Jan gaat] wei Jan goes not to Poland, hut [i.e. Jan goes] aff. naar de DDR. to the OOR 'Jan does not go to Poland, but Jan does go to the OOR. '

( l 0)

Jan geeft het boek niet aan Piet (maar) [i.e. Jan geeft het boek] Jan gives the book not to Piet but [i.e. Jan gives the book] wei aan Jaap. aff. to Jaap 'Jan doesn't give the book to Piet, hut he will give it to Jaap. '

(1 1)

Het is niet duidelijk of hij wei of niet naar Polen gaat. It is not clear whether he aff. or not to Poland goes ' It is not clear whether he will, or will not go to Poland.'

[i.e. gaat naar huis] . (Le. goes to home)

There are also examples that, although they confirm the parallelism between POS and NEG, could be understood as counterexamp les. It is clear, though, that these examples, which are characterized by the impossibility of wei at a position where niet is easily used, form strong evidence for the need of having a transformation which deletes pas in those cases where this element would be interpreted as contrastive but isn't so. ( 1 2) ( 1 3)

Hij eet en drinkt. ' He eats and drinks.' Hij [eet en drinkt niet/wel] .

Positive polarity and evaluation 5 19

( 1 4)

Hij [eet (0)] en [drinkt

( 1 5)

*Hij feet

( 1 6)

Hij [eet

{ } niet , wei

,

we i

nie

* wel

en drinkt] .

{ }] niet

{ � }].

en drinkt

Sentence ( ] 2) is ambiguous between a reading in which 'eat and drink' is interpreted as a single message regarding the activity described by the two verbs together and a reading in which there are two distinct messages: 'he eats' and 'he drinks' . In the following sentences, the brackets indicate which type we are dealing with. I won 't get into the details of the derivation; it suffices to see that a grammar6 which derives one reading by means of a phrase structure rule and the other by means of Conjunction Reduction is consistent with the distribution of POS and NEG . 2.3 .2.

We also note the paral lelism between P�S and NEG in tags: ( 1 7)

H ij is

geslaagd, (is het)

He has passed, ( 1 8)

Hij is

niet

He has not

6

is

{ }] niet

'I

?

* we

it

geslaagd, (is het) passed,

7

is

{ } niet

?

* wei

it

These can be realized, for example, in a grammar along the lines of Dougherty ( 1 968). The reading 'hij [eet en drinkt]' would be derived from V", and the reading 'hij [eet] en [drinkt]' from S It. 7 Tags have been used by Kraak ( 1 966) as a 'def ming contexr for negation. The argumentation thus becomes circular. Nevertheless, it still serves as a heuristic for the parallelism between POS and NEG .

520 Hendrik C. van Riemsdijk

The presence of weI and niet can be predicted then with respect to the syntactically explicit negative or positive polarity of the part of the sentence the tag is related to. This correctly predicts that a semantic negation that is part of the lexical meaning of for instance the verb does not have any effect. ( 1 9)

Hij is

gezald, (is het)

He has failed, is

{ }? niet

* weI

it

It seems reasonable to assume that zakken ('to fail the exam') is the negative antonym of slagen ('to pass the exam '). 2.3 .3 . The fact that the co-occurrence of wid and niet is impossible in the same simple sentence suggests that POS and must be base-generated in the 8 same position. Given that it is impossible to predict the accent for this type of ungrammatical clauses, I neutralize the difference between weI and wei by writing WEL.

NEG

(20)

* Inez neemt de pil Inez takes

{WEL} {WEL} niet

niet

.

the pill

The facts are the same for incorporated neg�tion: • •

8

niet + indef. art. --+ geen 'no' niet + ooit --+ nooi! 'never'

This has been proposed by Kraak ( 1 966), who gives the following rule:

(i)

s --+

s'

Positive polarity and evaluation 52 1

{WEL} {�EL} nooit

(2 1 )

*Inez neemt de pit

niet

{ WE l

nooit

met

(WELl

L

geen pit niet (22)

*Inez neemt

niet

. nOOit

. nomt

geen pi1

In substandard language, we find double negation, which carries the characteristics of negation with two elements, l ike the French negation ne ... pas. In Afrikaans, this type of negation has become the dominant type. All other instances of co-occurrence can be explained by deriving them from embedded sentences. (23)

Zuster Ursula kan de pit wei niet verdragen. Sister Ursula can the pill WEL not tolerate ' It is quite possible that sister Ursula does not tolerate the pill.'

Here, weI (which has lost its accent because of certain rules concerning the intonation of the sentence) relates to the matrix verb kan. Thus, (23) derives from something like (24): . (24)

Het [Zuster Ursula verdraagt de pil nietl s kan weI.

The inverse pattern is not true: (25)

*Zuster Ursula kan de pil niet wei verdragen.

In certain dialects, (25) can be interpreted as 'not good' (= badly), but in present-day Dutch this use of wei is quite archaic. The other reading is completely ungrammatical. Thi s ungrammaticality must be explained either by the rule that deletes POS or by a constraint that applies to this type of structures. Sentence (26) is also very strange:

522 Hendrik C. van Riemsdijk

*Het is niet zo dat [So Zuster Ursula de pil wei kan verdragenJs. 'It is not the case that . . . . '

(26)

Schema ( l ) predicts that S' presupposes that there is someone else who does not tolerate the pill. If one denies S', it becomes impossible to contrast S' with its presupposition; there is no longer any reason then for wid to not he deleted. The exact formulation of this type of constraints remains obscure, however. 2.3 .4. Klima ( 1 964) included POS and NEG in the class of adverbs that also comprises altijd 'always', zelden ' seldom ' and a whole series of other adverbs. It seems, though, that weI and niet behave somewhat differently from the other adverbs with regard to their placement with respect to a definite direct object, when no other element of the clause carries a contrastive accent. There is an asymmetry between (28a) and (30a). (27) a. Dik zoekt b.

{ ��I } een

geen

maocitaat.

. * D.k zoekt een maocitaat

{Wei} ,

Illet

_

'Dik looks for a quotation from Mao.' (28) a . *De arbeiders haten h.

{�el } met

het kapitalisme. .

De arbeiders haten het kapitalisme

{�e }. l

met

'The workmen hate capitalism.' (29) a. Dik zoekt altijd een maocitaat. h . *Dik zoekt een maocitaat altijd. 'Dik always looks for a quotation from Mao' (30) a. De arbeiders haten altijd het kapitalisme. h. De arheiders haten het kapitalisme altijd. 'The workmen always hate capitalism . '

Positive polarity and evaluation 523

2.3 .5. The position of WEL is sensitive to the difference between nominal phrases and prepositional phrases. (3 1 )

{

, Ze bewondert .

* wel hem

}

* wei hem

.

' She admires him. ' (32)

Ze bewondert

{

??Wel hem, maar niet . .. wei hem

}

.

' She admires him, but not. . . ' (33)

Ze I ijkt

{�

W I op hem

}

wei op hem

.

' She resembles him.' (34)

{

}

??WeI OP hem, maar niet... .. . Ze lijkt wel op hem

The facts for the inverted position are grosso modo as follows: (35 ) (36)

WEL.WEL.

Ze bewondert hem *Ze lijkt op hem

The different placement of adverbs like altijd sometimes causes a change in the presupposition; for wei, the presupposition always remains the same. (37) a. Jan verkracht wel lelijke meisjes 0 . h. Jan verkracht 0 lelijke meisjes weI 'Jan rapes ugly girls.' (38) a. Jan verkracht altijd lelijke meisjes 0 . b. Jan verkracht 0 lelijke meisjes altijd . 'Jan always rapes ugly girls. '

524 Hendrik C. van Riemsdijk

(37a) and (37b) presuppose that Jan does not rape pretty girls. (3 8a) presupposes that Jan never rapes pretty girls, while (3 8b) presupposes that Jan sometimes rapes pretty girls. 3. 3. 1.

The facts show rather clearly that wei has the function of contrasting a clause with another one that can be presupposed. The contrast finds itself within the opposition between positive and negative. We have just seen in section 2.3 .3 that sometimes the contrast is formally present but semantically absent, or even impossible. That is especially striking when one "contrasts" a predicate with its antonym or quasi-antonym. (39)

* Jan is niet liberaal, maar Piet is weI rood.9 'Jan is not liberal, but Piet is red.'

(40)

*Nixon is wei een oorlogsmisdadiger, en Oe Tant is geen vredesapostel. 'Nixon is a war criminal, and Oe Tant is not a peace apostle.'

Sometimes, the relative antonymy or synonymy of two predicates results from their usage in such frames. (4 1 ) (42)

Hij is geen aardige vent, maar hij verteJt wei schunnige moppen. 'He is not a nice guy, but he tells obscene jokes.' ' Hij is wei een aardige vent, maar hij vertelt schunn ige moppen.

In (4 1 ), the evaluation of the capacity to tell obscene jokes is positive, i.e. it is part of the behavior that the speaker expects to be present in people that he considers to be nice. In (42), this is precisely the other way around. 3.2.

We note another instance of this sensitivity to positive or negative evaluation in the case of unaccented wei. There are several distinctions, apparently binary ones, which play an important role. 9

I

use the asterisk here without m aking a distinction between semantic and

syntactic deviance.

Positive polarity and evaluation 525 First of all, there are non-evaluative adjectives like green. These adjectives can become evaluative when they are used metaphorically and/or in particular contexts in which certain qualities, normally neutral ones, are essential for the evaluation of the facts. There is also a certain number of adjectives which are very often evaluative but whose polarization, IO i.e. their having a positive or a negative connotation, depends on certain extra­ linguistic factors and varies a lot from person to person. This holds, for example, for Marxist, conservative, pious. Another class of adjectives is essentially evaluative in the sense that it contains adjectives such as good, pretty, nice, pleasant etc. and their negative antonyms, respectively, which are used for the moral, aesthetic, sentimental etc. evaluation. This last class comprises two degrees of emphasis that are clearly different. Besides the unmarked adjectives mentioned above, there are marked adjectives such asfantastic, great, very good, miserable, terrible, very bad, etc. Without taking a position on the exact nature of these semantic properties of adjectives, nor on the representation that must be assigned to them, we can say that they have much in common with the elements POS and NEG. The confusion in the use of the terms 'positive' and 'negative' for the evaluation, as well as for the syntactic facts regarding POS and NEG, is certainly not an accident. This similarity is confirmed by the fact that there are rather severe restrictions on the co-occurrence of wei with II evaluative adjectives. As a matter of fact, wei is only allowed with non­ emphatically positive adjectives; the meaning is that of a careful affirmation that the positive quality in question can be attributed to what is spoken about. We observe the following distribution : * fantastisch

( 43)

Dit boek is wei

' fantastic'

goed

' good'

* slecht

t bad'

* miserabel

' miserablet

* groen

t green'

'This book is . . . . ' 10

See Gruber ( 1 967)

and Givan ( 1 970).

I I There is nothing surprising abeut the fact that

selectional restrictions; it is precisely

a

wei does not show the same

characteristic property of contrastive

sentences that they permit constructions that are not allowed in non-contrastive sentences.

1f .�. .

'.'.. "

;r

526 Hendrik C. van Riemsdijk

(44)

?Dit grasveId is weI groen. ' This greensward is green. '

(45 )

±Jan i s wei marxistisch. 12 'Jan is a Marxist'

(46)

Jan is wei

{ } * gek

' foolish'

gek

' funny'

Often, the negation is explicit in the adjectives. That does not mean, however, that weI in all those cases is impossible, because the negation can yield a positive negative adjective. (47)

Dit boek is wei

{

interessant

}

* oninteressant

'This book is quite (un)interesting.' (48)

De sfeer was er wei

{

* gedwongen

}

ongedwongen

'The atmosphere was quite forced/relaxed there.' The constituents within which the evaluation takes place can be complex: the adjective can be composite, it can be modified by a degree word (very, not too, etc.) and, furthennore, it is possible to have a prepositional constituent at the position of the adjective: .

(49)

Jan is weI

' Jan is . . . . '

12

±

i n dicates variation in acceptability.

.

.

J ,

'

Positive polarity and evaluation 527 The question of where the origin, i.e. the exact position, of the evaluation

s ign is thus remains open, and as long as we do not have an exact formulation, it is impossible to attack the problem of a general formulation of the constraints. So far, the following constraints are needed :

1. 2. 3.

* wei + non-evaluative adjective *wel + negative evaluation *wel + marked evaluation1 3

To these constraints we can at least add the one which blocks

4.

* ... .

(25) and (26):

NIET . . . WEL . . . . .

In order to better understand the schema of constraints, we must also

exam ine other restrictions on 'evaluability'. I mention three of those : •

(50)

Imperatives become strange with negative qualities.

Wees

Be •

Questions

{ {

lief, aardig, voorzichtig . . . .

. , ?? . . stout, ?? . . vervelerrl, ?? . . onvoorzlchtlg . . .

nice, kind, careful . .

}

I.

.. }

naughty, nasty, careless . . . formed

with

hoe

'how '

practically

always

concern

evaluation, except if one expl icitly indicates another quality like

dimension, price, weight etc. If the evaluation is already present in the

clause, the question becomes unacceptable.

(5 1 )

*Hoe is die

{

gOede

}

slechte

film?

'How is that good/bad movie? ' •

Comparatives that are modified by degree words have a strange

distribution. 14

13 There are, as a matter of fact, indications that 2 and 3 have to be treated as two distinct semantic features. The ambiguity of the word terrible, for example, refers only to the polarization 'positive/negative' and not to the feature 'marked! unmarked' . 14 Since the grammaticality judgments for (52) are imprecise and display a lot of variation, I will indicate them in a very rudimentary way.

528 Hendrik C. van Riemsdijk

Zijn tweede boek was

(52)

ongelofelijk

{?

ve l

�

. . welmg

j}

?Onbelangrijker minder belangrijk

.. beIangrIJker

)

dan het eerstc .

'His second book was incredibly much/little more/less (un)important than the first one.' The problem with onbelangryker is apparently semantic in nature too, since ongedwongener is clearly better (cf. (48» . Only when we will approximately know how to count the "+" and the "-", the POS and the NEG etc., we may have hope to discover further regularities. This will allow us to simplify and generalize the constraints and to formulate a coherent solution. Although the problems concerning evaluation as sketched above concern the extra-linguistic domain, we can expect that some of the serious problems that follow from a syncategorematic l5 analysis of evaluation can be avoided in a theory that uses semantic evaluative features whose positive or negative specification depends partially on extra-linguistic factors. References Bierwisch, Manfred. 1 967. Some semantic features of Gennan adjectivals.

Foundations afLanguage 3, 1 -36.

Carstairs, An drew . 197 1 . Syncategorematic words. Linguistic Inquiry 11. 1 , 1 07-

1 1 0.

Dougherty, Ray. 1 968. A transformational · grammar of . coordinate conjoined structures. Unpublished doctoral dissertation, MIT. Ducrot, Oswald 1 968. La descri ption semantique des enonces fram;ais et la notion de presupposition. L 'Homme VIII, I . Givon, Talmy. 1 970. Notes on the semantic structure - of English adjectives. Language 46, 4. Gruber, Jeffrey. 1967. Functions of the lexicon in fonnal descriptive grammars. SOC TM-3770/000/00. Santa Monica, California. Katz, Jerrold. 1 966. The philosophy o/Ianguage. New York: Harper and Row . Klima, Edward S. 1 964. Negation in English. The structure of Language, Jerry Fodor and Jerrold Katz (eds.), Englewood Cliffs. New Jersey. Kraak, Albert. 1 966. Negatieve zinnen. Hilversum: W. de Haan. Sampson, Geoffrey. 1 970. Good. Linguistic Inquiry 1 .2, 257-260.

J5

See

Katz ( 1 966), Sampson (1970), Carstairs ( 1 97 1).

Phase theory and the privilege of the root Luigi Rizzi 1.

The Spell-out mechanism

A salient difference between root and embedded clauses is that the declarative complementizer is typically null in main clauses, and typically expressed (if it is expressed at all) in embedded clauses. For instance it is always null in main clauses in French and English, and it is obligatory in embedded clauses in French, and optionally expressed in some embedded environments and obl igatory otherwise in English:

(1) (2)

(*That) Bill thinks (that) Mary is sick (*Que) Pierre pense *(que) Marie est malade

Why is it so? One possible line of analysis is that this contrast is a particular case of a more general tendency not to pronounce clause initial material, material at the edge of the root category. This tendency is called the "privilege of the root" in Rizzi ( 1 992) and following work, and its �onsequences are explored in child and adult grammars in the reference quoted and elsewhere. As is observed in Rizzi (200Sc), a reference where the ideas presented more sketchily here are developed in more detail, Phase Theory offers a natural mechanism to understand the privilege of the root. Within main­ stream approaches to Phase Theory (Chomsky 200 1 , 2005, Nissenbaum ' 200 1 ), when the category defining a phase is merged (C or v, in this system), its featural requirements are satisfied, e.g. by moving an element to its Spec, and then the complement of the phase head is sent to the interfaces, to Spell-out in particular. So, we have a mechanism like the following: (3)

Send to Spell-out the complement of a phase head.

Let us consider how this mechanism permits successive cyclic movement (the vP phase is not immediately relevant, so it is omitted here) in the derivation of a sentence like (4) (traces are represented as copies within angle brackets):

530 Luigi Rizzi

(4) Who do you think that B ill saw? (5) a. C [Bill saw who] b. Who C [Bill saw <who>] c. Who C [B ill saw <who>]

�

Move � Send to Spell-out

The phase head C is merged with the IP, as in (Sa), and then the wh element is moved to Spec of C, as in (5b). At this point, only the complement of C, the IP underscored in (5c), is sent to Spell-out under (3), so that the moved wh element who remains accessible to narrow syntactic computation and can be moved to the higher C system as new material is merged in the next phase. If the spell-out mechanism (3 ) had affected the whole projection of the phase head, all its content. would be frozen, and successive cyclic movement would be incorrectly ruled out (see the references quoted for detailed discussion). An interesting by-product of (3 ), as stated, is that it immediately accounts for the main-embedded asymmetries in ( 1 )-(2): the embedded C is sent to spell-out at the matrix phase (unless a language specific rule pennits its zero realisation, as in certain structural contexts in English), but the main C, part of the edge of the root phase, is not sent to spell-out at all under (3): only its complement is. So, as stated, the spell-out principle (3) provides a simple explanation for ( 1 )-(2). Nevertheless, it raises obvious problems. For instance, when the wh element is moved to the main C-system in (4), it would end up in the edge of the root phase, so it should also be left unpronounced, incorrectly. For this kind of reasons, Nissenbaum proposes to complete (3) with the assumption that root clauses are introduced by a special [+R] head, and the following holds: (6)

Send to Spell-out the complement of [+R] .

This auxiliary clause correctly ensures the p�oriu�ciation of the -Ieft­ peripheral material in the main clause: who in (4), overt topics, etc. But the drawback is that the simple account of the asymmetry in ( 1 )-(2) is lost. 2.

Phase Theory and Cartography

Things change if we combine Phase Theory with the cartographic approach, the attempt to draw detailed maps of syntactic structures (Belletti 2004, Cinqu e 1 999, 2002, Rizzi 2004). According to the analysis presented

Phase theory and the privilege ofthe root 53 1

in Rizzi ( 1 997), the C-system is defined by two types of heads, Force and Fin(iteness),. delimiting a space which is used to host various kinds of left­ peripheral operators, and the discourse-related positions of topic and focus, among others (positions irrelevant for the present discussion are omitted): (7)

Force . . . Top . . . Foe . . . Fin . . .

Suppose now, as is natural, that the unmarked root category is the Force phrase, and Force is the head of the C-system which normally defines the phase. Suppose also that wh elements end up in Spec of Focus, as is shown by the fact that, in languages with overt Foe markers, wh elements carry such markers (Aboh 1 998, 2004), and by m uch other evidence (see Rizzi 1 997 for discussion). So, when the root Force head is merged with the structure, say, in (4), its complement including the rest of the left periphery (e.g . who in Spec-Foe), is sent to spell-out under (3), so that the wh operator is pronounced, and the special clause (5) is not needed (in this approach, the 'escape hatch' for extraction from embedded clauses in (4)­ (5), etc. should presumably be the Spec of the embedded Force head). On the other hand, the root Force marker is not sent to spell-out under (3), whence the main-embedded asymmetry in ( 1 )-(2). Asymmetries of the kind illustrated by that and other declarative complementizers exist also for other kinds of complementizers. For instance, Lasnik and Saito ( 1 992) observe that the question particle ka in Japanese is obligatory in embedded questions, but optional in main questions: (8) a. John-wa JohnToP b. Mary-ga Marynom

doko-ni where-to [John-ga lohnnom

ikimasita (ka) went Q nani-o katta ka ] whatacc bought Q

siritagatte iru koto want-to-know fact

Notice that here the asymmetry is weaker than in ( 1 )-(2), as the main question marker may optionally be expressed. This could reflect an inherent difference in markedness between declarative and question markers (the former being the unmarked case), but there may also be more interesting cartograph ic reasons for the difference. Rizzi (200 1 ) argues that in Italian the interrogative complementizer se (if) appears in a lower position than the declarative complementizer che (that), as is shown, for instance, by the fact that the former can be preceded

532 Luigi Rizzi

(and followed) by left-dislocated topics, while the latter (in my variety of Italian) can only be followed by topics: (9) a. Gianni si domanda, questo libro, se te 10 potrebbe dare 'Gianni wonders, this book, if he could give it to you' b. * Gianni crede, questo libro, che te 10 potrebbe dare 'Gianni believes, this book, that he could give it to you' Another even more direct type of evidence supporting this conclusion is provided by the complementizer order que si (that if) in certain Spanish indirect questions (see Rizzi 200 1 for other types of evidence and discussion) ( 1 0)

Me preguntaron (que) si tus amigos ya te visitaron en Granada 'They asked me that if your friends had already visited you in Sufter ( 1 994:349) Granada'

So, the reference quoted proposes a partial map like the following, where Int{errogative) is the position corresponding to se, and various topic position are possible under Force: (1 1)

Force . . . Int . . .

Other languages, like Dutch, illustrate the opposite order of dat (if that: Hans Broekhuis, p.c.); I don't think this should be interpreted as requiring a parametrisation of the order of heads, but rather is one of the many indications of the 'versatility' of elements corresponding to that, which may encode force, but also finiteness and focus (Mioto 1 999) in d ifferent languages. Now, if ( I I ) basically corresponds to the structure of main clauses as well, the optionality of ka observed in Japanese main questions can be expressed as follows: the structure can project up to the force head, or stop at Int; in the former case, lea is sent to spell-out under (3); in the latter case, under the assumption that the root category always defines a phase, ka ends up in the phase edge, and is omitted. In case of an embedded question like (8b), lea is always sent to spell-out at the main clause phase. We leave open whether the alternation between whether/if and zero in embedded and main interrogatives in English is amenable to the same kind of analysis (which should account for the obligatory absence of such elements in main questions ).

Phase theory and the privilege ofthe root 533

This analysis opens up the following possibility. Different languages may be able to 'truncate' the CP system at different levels, hence admit different kinds of root categories in addition to the universally available Force. Combining this with the adopted version of phase theory, this may offer a way to express the parametrisation of the omissibility of clause initial elements, which in fact varies to some extent across languages. The consequences of the truncation approach are explored for adult grammatical systems and for child language in Rizzi (2005a-c). Here we will only look at the Topic Drop parameter. 3.

Topic Drop

Ross ( 1 982) observed that colloquial German allows for the possibility of dropping a topic (see also Huang 1 984 for discussion). In fact, colloquial varieties of most Germanic V-2 languages (not all, as West Flemish does not: Haegeman 1 996) seem to have this option. I wil l follow here the description of a variety of colloquial German provided by Cardinaletti 1 990, illustrating what I will call the 'pure' version of Topic Drop. In this variety, a subject can be dropped in initial (V-2) position ( 1 2a), but not in . the middle field ( 1 2b-c), nor in embedded positions, in embedded V-2 or V-final clauses ( I 2d-e). ( 1 2)

Topic Drop in Colloquial German: a. (lch) habe es gestem gekauft I ' have it yesterday bought b. Gestern habe *(ich) es gekauft yesterday have I it bought c. Wann hat *(er) angerufen? When has he telephoned? d. Hans glaubt *(ich) habe es gestem gekauft . Hans believes I have it yesterday bought e. Hans glaubt dass *(ich) es gestem gekauft habe Hans believes that I it yesterday bought have

A topicalized object can also be dropped ( 1 3 ), while an expletive subject cannot be dropped ( 1 4a-b):

(13)

(Das) this

habe ich Ida!! gestern gekauft have I yesterday bought

534 Luigi Rizzi

( 1 4) a. *(Es) it b. *(Es) it

wurde viel getanzt was a lot danced hat viel geregnet has a lot rained'

The latter constraint follows from the fact that the Top position is specialized to host D-linked, or 'partitive' elements (Rizzi 200Sb), and expletives do not qualify as they lack referential properties. The contrast between initial and non-initial position in ( 1 2) clearly shows that Topic Drop is a particular manifestation of the privilege of the root, so that it would be desirable to connect it to the approach sketched out in the previous sections. The hypothesis that different languages may choose root categories 'truncating' the full cartographic structure at different levels suggests an obvious possibility: Topic Drop languages may be languages in which the structure can be truncated at TopP; i.e., consider a formulation of the Topic Drop parameter as the presence/absence of the following option in a grammatical system: ( 1 5)

Top may be a possible root category

So, all languages have Force as a possible root, and different languages may vary as to what other categories (lnt, Top, etc.) may be taken as roots. Now, it is natural to assume that the root category always defines a phase: phases are defined as syntactic objects enjoying a certain degree of autonomy at the interfaces (Chomsky 200 1 ), and the root is the autonomous syntactic object par excellence. So, th� root TopP will be a phase in languages selecting option ( 1 5). Then, in the derivation ,of ( 1 2a) or ( 1 3), if the structure is truncated at TopP, the Specrop is not sent to spell-out, hence we have Topic Drop. If the structure is projected up to the Force phrase, the topic will be in the complement of the root phase head (F<>rc�, ) n this case), , We thus have 'ilie, i,ollseived optional and therefore it will be pronounced.' . ' , ' , : : " ' , '- , ' , drop. What about the fact that the inflected verb is not dropped in ( 1 2a), along w ith the topic? If in V-2 constructions the inflected verb always moved to the head of whatever left peripheral position is activated (Top in ( 1 2a), etc.), it would be in the edge of the root phase, hence it should behave on a par with the topic in permitting the lack of spell-out, contrary to fact. So, this suggests that the inflected verb does not 'follow' the activated left-peripheral position in V-2, but stops in a lower position.

Phase theory and the privilege ofthe root 535

In fact Koster (2003), building on the classical analysis of Den Besten ( 1 977/83) has proposed that the movement of the inflected verb to C in V-2 is motivated by scope-d iscourse requirements, basically by the necessity of moving the T-bearing element to an appropriate scope position. Combining this idea with the cartography of the left periphery, it seems natural to assume that the head-movement part of V-2 takes place to the T-related head of the C-system, Fin, and that it stops there (the fact that a single left peripheral Spec can be activated may be understood as a 'bottleneck effect', if Fin formally requires to have its Spec position filled; the attracted phrase then continues to its final destination, Top, Foc, etc., which is required by its discourse status; no other element can be directly attracted to the left periphery as the intervening Spec-Fin would trigger a minimality effect; see Haegeman 1 996 for discussion and an implementations of this idea through an EPP feature in Fin). The crucial thing for our purposes here is that the inflected verb stops in a position too low to benefit from the privilege of the root, hence it is pronounced. Other Germanic varieties (colloquial Swedish for instance: Rizzi 1 992) also permit the dropping of an initial expletive. Such expletives, which are neither topic nor focus, presumably fill an A position in the C-system immediately above the Fin projection, possibly the Spec an Agr-like projection (Poletto 2000, Schoorlemmer 2004); root expletive drop can thus be accommodated by assum ing that the languages in question allow a truncation at the level of the projection immediately containing this A position, "in addition to the truncation of TopP, as in ' pure' Topic Drop languages. References Aboh. Enoch. 1 998. From the Syntax of Gungbe to the grammar ofGbe, Doctoral Thesis, University of Geneva. Aboh, Enoch. 2004. The Morphosyntax of Complement-Head Sequences. Oxford University Press, New York. Belletti, Adriana (ed.). 2004. Structures and Beyond. The Cartography of Syntactic Structures, Vol. 3, Oxford University Press, New York. Besten, Hans den. 1 977/1 983 . On the Interaction of Root Transfonnations and Lexical Deletive Rules, in W. Abraham, (ed.). On the Formal Syntax of Westgermania, Amsterdam: Benjamins, 47- 1 3 1 . Cardinaletti, Anna. 1990. Pronomi nulli e pieonastici nelle lingue gennaniche e romanze, PhD diss., University of Venice. Cardinaletti, Anna. 2004. Toward a Cartography of Subject Positions, in Rizzi, ed.

(2004a). 1 1 5- 1 65.

536 Luigi Rizzi Chomsky, Noam. 200 1 . Derivation by Phase. In M. Kenstowicz (ed.). Ken Hale: A Life in Language, MIT Press, Cambridge, Mass. Chomsky, Noam. 2004. Beyond Explanatory Adequacy, in Belletti, ed. (2004). Chomsky, Noam. 2005 . On Phases, ms., MIT. Cinque, Guglielmo. 1 999. Adverbs and Functional Heads, Oxford University Press, New York. Cinque, Guglielmo (ed.). 2002. The structure of DP and IP - The Cartography of Syntactic Structures, Vol. 1 . Oxford University Press, New York. Haegeman, Liliane. 1 996. Verb second, the split CP and initial null subjects in early Dutch fmite clauses. GenGenP 4, 1 33- 1 75. Huang, James. 1 982 . Logical Relations in Chinese and the Theory qf Grammar, PhD d issertation, MIT, Cambridge, Massachusetts. Kayne Richard S. 1 994. The Antisymmetry of Syntax, MIT Press, Cambridge, Mass. Koster, Jan. 2003. All Languages are Tense Second, in Jan Koster and Henk van Riemsdijk, eds., Germania et Alia: A Linguistic Webschrift for Hans den

Besten. Lasnik, Howard and Mamoru Saito. 1 992 Move Alpha, Cambridge, Massachusetts: MIT Press. Mioto, Carlos. 1 999., A periferia esquerda no portugues brasileiro ms., Universidade Federal de Santa Catarina, University of Siena. Nissenbaum, Jon. 2000. Investigations of Covert Phrasal Movement, PhD Dissertation, MIT, Cambridge, MA. Poletto, Cecilia. 2000. The Higher Functional Field in the Northern Italian Dialects, Oxford University Press, New York. Rizzi, Luigi. 1 992. Early Null Subjects and Root Null Subjects, GenGenP 1 992. published in T. Hoekstra and B. Schwartz, eds., Language Acquisition Studies in Generative Grammar, John Benjamins, AmsterdamlPhiiadelphia; also in Rizzi (2000). Rizzi, Luigi. 1 997. The Fine Structure of the Left Periphery, in L. Haegeman, ed. Elements of Grammar, Kluwer, Dordrecht; also in Rizzi (2000a). Rizzi, Luigi. 2000. Comparative Syntax and Language Acquisition, Routledge, London. Rizzi, Luigi. 200 1 . On the position INT(errogative). in the Left Periphery of the C lause, in G. Cinque, G. Salvi, eds. Current Studies in Italian Syntax, Elsevier, Oxford, pp. 287-296. Rizzi, Luigi. (ed.) 2004. The Structure of CP and IP - The Cartography of Syntactic Structures, Vo1.2, Oxford University Press, New York. Rizzi, Luigi 2005a. On the Grammatical Basis of Language Development: A Case Study, in G. Cinque. R. Kayne (eds.) Handbook of Comparative Syntax, Oxford University Press, New Y orklOxford. Rizzi, Luigi. 2005b. On Some Properties of Subjects and Topics, in L. Bruge et al. (eds), Proceedings of the XXX Incontro di Grammatica Generativa, Cafoscarina, Venezia. .

Phase theory and the privilege ofthe root 537 Rizzi, Luigi. 2005c. Grammatically-Based Target-Inconsistencies in Child Language. In Deen et at. (eds.)., The Proceedings ofthe Inaugural Conference

on Generative Approaches to Language Acquisition - North America (GALANA). UCONN / MIT Working Papers in Linguistics. Cambridge, MA: MIT Press. Ross, John Robert. ] 982. Pronoun Deleting Processes in Gennan, paper presented at the annual meeting of the LSA, San Diego, California. Schoorlemmer, Eric. 2004. Double Agreement in Complex Inversion, paper presented at Going Romance 2004. Suffer, Margarita. 1 994. V-movement and the Licensing of ArgumentaI Wh Phrases in Spanish, Natural Language and LinguistiC Theory, 12, 335-372 .

On the role of parameters in Universal Grammar: a reply to Newmeyer Ian Roberts and Anders Holmberg 1.

Introduction'

Newmeyer (2004) argues that the notion of parameter of Universal Grammar has no role to play in accounting for cross-linguistic differences in syntax, and that, instead, "language-particular differences are captured by differences in language-particular rules" ( 1 83). He tries to show that parameter-approaches "have failed to live up to their promise" ( 1 8 1 ) and that "the hopeful vision of UG as providing a small number of principles each admitting of a small number of parameter settings is simply not workable" ( 1 85). In this paper, we defend the principles-and-parameters model of cross­ linguistic variation. We propose that Newmeyer's arguments against it are based on m isunderstandings either of theory or of data, are conceptually misconceived, illogical or simply false. The rule-based alternative he suggests is hard to evaluate, since few actual instances of it are offered, and none is discussed in any detail. In a couple of instances that Newmeyer discusses, his rule-based account is a notational variant of a parameter­ based theory. In other cases it represents a retreat to observational adequacy. The first part of the paper replies to Newmeyer's critique of parameters (his pp. 1 98-2 1 5). The second part demonstrates an example of the efficacy of the parameter-based approach, drawing on work by Platzack and Holmberg on Scandinavian. 2.

Newmeyer's critique of parameters

Newmeyer begins by enumerating eight ways in which parameters might be thought to be attractive, as follows:

I The research reported here was carried out under the auspices of th e project Null Subjects and the Structure of Parametric Theory funded by the Arts and Humanities Research Council of Great Britain ( Grant No. APN 1 4458).

On the role o/parameters in Universal Grammar 539

(1)

a.

Parameters are descriptively simple, whereas rules are (generally) not. -

b.

Parameters have binary settings (an idea which i s inapplicable to rules).

c.

Parameters are small in number; the number of rules is open­ ended.

d.

Parameters are hierarchically/implicationally related, thereby accounting for both order of first-language acquisition and typological generalizations (there is nothing comparable for rules).

e.

Parameters are abstract entities with a rich deductive structure, making

possible the prediction of (unexpected) clusterings of

morphosyntactic

properties.

f.

Parameters and the set of their possible settings are innate (and

g.

Parameter settings are easily learned, while rules are learned with

therefore-univer�al). Rules are not (normally) assumed to be drawn from an innate set.

h.

greater difficulty.

Parametric change i s markedly different from rule-based change

(such as grammaticalisation and morphological change).

( 1 85)

Let us look at Newmeyer' s attempted refutations of each of these points in tum.

aJ Parameters are descriptively simple. Here Newmeyer's point seems to be that there is no j ustification for introducing the notion of parameter into the theory since their postulation leads to a more complex system than one which only has language-specific rules. However, Newmeyer' s theory includes rules which are equivalent to parameter settings. For example Newmeyer proposes to handle head­ complement order not by a parameter with two settings, as in the classical

formulations of the head-parameter, but by language-particular rules of the form ( l a,b), where ( l a) is operative in Japanese, ( l b) in English (p.

(1)

a.

b.

1 84).

Complements are to the left of the head.

Complements are to the right of the head.

Given that

UG

specifies that a phrase consists of a head and a single

complement, as Newmeyer aSsumes, and given that the order cannot be left

unspecified, it follows that there are only two possible rules specifying head-complement order, and each language has to pick one of them . In case

540 Ian Roberts and Anders Holmberg

head-complement order is relativized to categories (for instance, complements can precede nouns but follow verbs in the same language, a fact noted by Newmeyer), there can only be two rules times the number of complement-taking categories; still a finite and relatively small number. Thus some of Newmeyer's language-specific rules amount to a very close notational variant of parameter settings; after all, a parameter can always be reformulated as a set of competing rules. In the case of head-complement order, what is language-specific is the choice of rule from a set of possible rules allowed by VG, exactly as in the case of parameter-setting. We conclude that the model sketched by Newmeyer is not formally simpler than standard principles-and-parameters theory.2 b) Binarity. Newmeyer claims that "there is little evident binarity in morphosyntax" ( 1 9 1 ) . This overlooks the fact, evident from the most cursory glance at the literature on syntactic typology since Greenberg ( 1 966), that many forms of word-order variation are best phrased, at least at the observational level, as dyads of type OVNO, NAdj/AdjN, etc.; as discussed above, Newmeyer himself proposes a binary set of competing rules to replace the head parameter. Similarly, parameters which deal in the presence vs. absence of a property (nul l subjects, overt wh-movement, etc.) necessarily deal in binary oppositions. One example is given of a non­ binary parameter: Manzini and Wexler's ( 1 987) parameter for binding domains, where each parametrised definition of governing category is in a subset relation to the next: binding domains can be determined by Inft, Tense, referential Tense, or root Tense, for example. However, this can be trivially restated as a network of implic�tionally related binary parameters, as in (2):

(2) a. Is the binding domain determined by lnft? YESINO b. If NO, is the binding domain determined by Tense? YESINO c. If NO, is the binding domain determined by referential Tense? YESINO. d. If no, is the binding domain is determined by root tense? YESINO We have here a further instance of a parametric hierarchy of the type discussed by Baker (200 1 ) (on which more below), with the interesting 2 We do not necessarily subscribe to the view that head-complement order is

determined by the classical head-complement parameter, but this does not affect the points being made here in relation to Newmeyer' s rule-based alternative.

On the role ofparameters in Universal Grammar 54 1

additional property that each position on the hierarchy is related to the next in terms of the Subset Principle: the highest position defines the smallest gram mar, and each further position going down the hierarchy represents a superset grammar in relation to the immediately higher one. It is therefore easy to see how acquirers might proceed down the hierarchy; each step down the hierarchy is motivated purely by positive evidence moving them towards a superset grammar (see Manzini and Wexler ( 1 987) for discussion). More generally, the question of binarity is arguably more a matter of formulation than anything else. It may be useful to formulate parameters as binary options, since this creates the possibility of seeing a set of parameter values as a kind of 'bit string'; see Clark and Roberts ( 1 993) for some discussion of the potential utility of this. The only really substantive claim behind a binary formulation of parameters is that the values are discrete: . there are no clines, squishes or continua. c) Smallness of number. Here Newmeyer takes on the question of how many parameters there are. He points out that there may be hundreds or even thousands of parameters, if one extrapolates from the literature on comparative syntax of the past fifteen years or so, and concludes:

If the number of parameters needed grammars of the world's languages, idiolects is in the thousands (or, worse, them to an innate UG to my mind plausibility. ( 1 96)

to handle the different dialects and (possibly) millions), then ascribing loses all semblance of

Perhaps, but noone has ever suggested that there are millions of parameters. There may be millions of possible grammatical systems, but only twenty independent binary parameters are needed to produce that order of magnitude of grammatical systems. Nearly all estimates of the number of parameters in the literature (Lightfoot ( 1 999), Kayne (2000), Roberts and Roussou (2003 , Chapter 5» judge the correct figure to be in the region of 50- 1 00. This is a plausible, if conservative, conjecture. Here too Newmeyer seems to think that rules are necessary to any formulation of UG, while parameters are an optional extra. For example, he states that "I see no reason to conclude that the number of the former [parameters AHIIR] is significantly less than the number of the latter [rules [AHlIR]" ( 1 96). Actually, there is one obvious reason: parameters allow us to collapse cross-l inguistic differences into single abstract -

542 Ian Roberts and Anders Holmberg

properties of grammars, while language-specific rules are just that: language-specific. A final point here: it may seem strange that evolution endowed the language faculty with 50- 1 00 choice points. However, if we think of parameters along the lines suggested in Roberts and Roussou (2003, Chapter 5) and elsewhere, then it may be that parameters are not really primitives of UG, but rather represent points of underspecification which must be filled in in order for the system to become operative. In that case, nothing is being added by assuming parametric variation; in fact, it may be compatible with assuming the most minimal initial state we can . This idea would provide a strong conceptual basis for our account of cross-linguistic variation, especially in the context of a minimalist conception of U G. d) Hierarchies. Here Newmeyer discusses Baker's (200 1 ) Parametric Hierarchy (PH) in some detail. He points out one or two empirical difficulties with it (e.g. the fact that, as formulated, it predicts that only SVO languages can be null-subject languages, which certainly appears to be false, given Irish at least). These difficulties seem to us to be difficulties of formulation, rather than difficulties with the concept of parameter hierarchy or the notion of parameter itself. But Newmeyer concludes "No hierarchy of the general form of the PH is capable of representing the parametric choices that the child is hypothesized to make" (20 1 ). This sweeping conclusion seems to us not to be justified by the observation, essentially, that the null-subject parameter is almost certainly ranked too low on the hierarchy as Baker presented it. Furthermore, there is no mention here of language-specific rules as an alternative. A moment's reflection reveals why this must be: language-specific rules make absolutely no predictions about complex patterns of relatedness of the type Baker's PH tries to make. As things stand, they predict that languages may vary without assignable limits. They may lack the empirical difficulties that beset something like Baker' s PH, because, lacking any constraint on their formulation of any kind, they are able to predict everything and therefore nothing. e) Clustering. Here Newmeyer makes two main points: that clustering effects within a single language can be predicted by language-specific rules and that the clustering effects often claimed to hold for well-known parameters such as the null-subject parameter and the head-complement parameter are illusory. He also makes some further related points regarding acquisition.

On the role ofparameters in Universal Grammar 543

The first point concerns the capacity of language-specific rules to predict clustering effects. In this connection, Newmeyer cites the phrase­ structure and transformational rules proposed in Chomsky ( 1 957) for the English auxiliary system: (3) a. AUX � TNS (M) (have + en) (be + ing) b. affixal element + verbal element � verbal element + affixal element Newmeyer states that from these rules "a host of seemingly unrelated properties follows: the fact that on the surface perfect and progressive in English are overlapping and d iscontinuous, the impossibility of double modals, the site for the insertion of supportive do, some of the possibilities of contraction, and more" (202). Aside from the fact that several of these further properties require further specification and do not fal l out directly · from the rules in (3), what is striking here is how a fairly robust cross­ linguistic generalisation regarding the ordering of Tense, Mood and Aspect (TMA) markers is being missed. As Cinque (1999: 1 53ff.) shows, many languages show a very similar ordering of TMA elements. The list in (4) illustrates a small subset of the languages listed by Cinque: (4) English Spanish Welsh Modern Greek Hindi Finnish Hungarian Mongolian . Abkhaz Lezgi�lO

TIMod-Pertect-Progressive-Voice T-Perfect-Progressive-Voice T-Perfect-Progressive T-Asp-Voice T(Past)-T(Anterior}-Progressive TlMood-Perfect-Voice Mood spcechAct-T-ModRoot MoodspcechAct-T(Past)-T(Future)-Progressive-Voice MoodIT(Past)-T(FuturelAnterior)-Perfect-Progressive-Voice MoodlT(Past}-T(Future}-Asp

There is a clear cross-linguistic pattern here; what seems to vary in fact involves how certain features are differently syncretised on certain position� cross-linguistically (Cinque's interpretation of the data is rather different). It is clear that English fits the cross-linguistic pattern rather well. So Newmeyer's example neatly illustrates how a reliance on language­ specific rules will lead one to miss significant cross-linguistic

544 Ian Roberts and Anders Holmberg

generalisations. A parameter-based approach, on the other hand, will naturally lead one to look for similarities and differences in other systems. Newmeyer � s second point is that the well-known correlations proposed by Rizzi ( 1 982) in connection with the null-subject parameter have been shown not to hold, and that this is to be expected since the notion of parameter argued for by Rizzi is illusory. Newmeyer's version of Rizzi's proposal is that the possibility of null thematic subjects in tensed clauses, null nonthematic subjects, 'free' subject inversion and apparent that-trace v iolations were typologically connected. The strongest possible g loss one can put on this is that any language must have all or none of these properties, irrespective of any other property of the system ? Gilligan's ( 1 987) cross-linguistic survey revealed that these very tight correlations did not hold if one surveyed a typologically diverse range of a hundred or so languages. Instead, only four implicational statements survive and five language types are attested whose existence was not predicted. Newmeyer sees the results of Gilligan's survey (which, like him, we will take at face value for the sake of the discussion) as damning for Rizzi's proposal. But consider what has actually been shown: an original very strong correlation was postulated on the basis partly of theoretical considerations and partly the close comparison of a small number of closely related languages. When a very large number of genetically and typologically highly diverse languages were compared for the 'same' properties, with no control as to the other typological features of these languages, the original correlations were shown not to hold in their original form, although four implicational statements could still be gleaned and five unsuspected language types observed. T
3

Rizzi never actually made this strong claim, but in fact identified two " related but autonomous p arameters" (1 982: 143), which give rise to a four-way typology. For the purposes of the present discussion we will continue to entertain the stronger hypothesis reported in the text, however, since this is what Newmeyer discusses.

On the role ofparameters in Universal Grammar 545

Newmeyer goes on to point out that the minimalist approach to parameters, which involves seeing them as inherently connected to features of functional heads, "makes it all but impossible to predict any significant degree of clustering" (208). This is clearly false: Rizzi ( 1 982 : 1 43) identified the core property allowing null subjects with a [+pronoun] feature of the functional head Inft (this feature may or may not be referential, giving rise to the two subparameters mentioned above). In MP terms, this can easily be restated as the presence of a D-feature on T, or perhaps, following Alexiadou and Anagnostopoulou ( 1 998), on the verbal inflection itself. The clustering of properties, and indeed the two subparameters, can then be stated almost exactly as Rizzi stated them. In this particular case, the restatement in MP terms is very straightforward, contrary to Newmeyer's claim. We will see a further example of clustering in Section 2 below. A final point concerns first-language acquisition of the null subject parameter: Newmeyer says "[i]t is very difficult for me to imagine how the child could ever come to set this particular parameter" (206). Perhaps this is hard for Newmeyer to imagine: much of what small children achieve in acquiring language is indeed extraordinary. But they do it: the evidence from the language-acquisition literature that young children acquiring null­ subject languages set the parameter very early in acquisition is unequivocal, as every study since Hyams ( 1 986) has shown (see Guasti (2002) for an overview). Wexler ( 1 998) enumerates the null-subject parameter among those which appear to be set at or before the time of the first multiword utterances in the course of normal acquisition. It is hard to see how this is achieved, whether we formulate the relevant grammatical properties in parametric terms or otherwise, but this is � empirical challenge linguistic theory must meet. And it seems clear that the postulation of language­ . specific rules in this domain wi1l get us nowhere, once again.

j) Innateness/universality. Here the central point Newmeyer wishes to make is that there is no need for prespecified options if UG leaves a range of possibilities open and positive evidence does the rest. He states "I have never seen a poverty of the stimulus-based argument for such an assumption [of prespecified options AHlIR] and doubt that one can be constructed" (2 1 1 -2). In fact, Lightfoot ( 1 99 1 , 1 999) argues at length for a selective theory of learning, showing the advantages such an approach has in dealing with complex learning problems and relating the idea specifically to the poverty of the stimulus. -

546 Jan Roberts and Anders Holmberg g) Learnability. Newmeyer's central objection here is that the postulation of any parameter or parameter-setting on the part of the acquirer seems at first sight to entail a kind of regress: "parameter-setting presupposes some non-negligible degree of prior structural assignment" (2 1 2). An objection of this general form can be made regarding almost any aspect of the acquisition of grammar: some parts of the grammar must be acquired before others. And the objection can be met in a very straightforward way, whether or not one is assuming a parameter-setting model of acquisition: some things must be learned before others, and it is part of the task of language-acquisition research to determine exactly what the "learning path" is. Very interesting progress has been made in charting the learning path connected to the parameters governing metrical phonology by Dresher ( 1 999) and the references given there. It is untrue to c laim that no progress at all has been made in this area by assuming a parametric approach: at the very least Wexler's Very Early Parameter Setting hypothesis tells us that many important parameters are fixed very early in acquisition. h) Diachrony. Newmeyer alludes to the six properties of parametric change enumerated by Lightfoot ( 1 99 1 ) and criticised by Harris and Campbell ( 1 995). However, he seems to take it for granted that there must be a difference between parametric change and other kinds of morphosyntactic change such as grammaticalisation. However, Roberts and Rou ssou (2003) devote an entire monograph to showing how grammaticalisation can be understood as a fairly standard kind of parametric change involving categorial reanalysis. Moreover, Roberts (in preparation) shows that there are real ly only two kinds of syntactic change: parameter change and changes to lexical entries of lexical heads. Given the hypothesis that parameters are associated with lexically encoded properties of functional heads, this reduces to the idea that change affects lexical entries, either of lexical heads or of functional heads. The latter case is familiar parameter change, the former lexical change. There is no deep problem here for the concept of parameter, and no case for replacing it with language-specific rules. 3.

Cross-linguistic correlations: a case study

As discussed by Newmeyer, the existence of cross-linguistic correlations of syntactic properties is crucial to parametric theory, since only they provide empirical evidence in favour of parameters, against l anguage-specific rules. However, according to Newmeyer "two decades of intensive research has

On the role a/parameters in Universal Grammar 547

failed to reveal the existence of the hoped for correlations" (2 1 1 ). As discussed above in connection with the null-subject parameter, we don 't think this conclusion is warranted. Let us consider a specific case from the principles-and-parameters literature of a set of putative cross-linguistic correlations, and try to assess both the validity of the correlations, and their impl ications for linguistic theory. As first discussed by Platzack 1 987, there is a syntactic division among the Scandinavian languages between what he called Insular and Mainland Scandinavian. The Insu lar Scandinavian (ISc) languages are Icelandic, Faroese, medieval varieties of the Mainland Scandinavian languages, and the A lvdalen dialect in Sweden. The Mainland Scandinavian (MSc) languages are all other varieties of Danish, Swedish and Norwegian. ISc and MSc differ with respect to the fol lowing five properties (the a-examples are Icelandic, the b- and c-examples Swedish): (5)

ISc but not MSc allows a null a. Nil hafa e komia b. Nu har *(det) komm it come now have EXP

nonreferential subject. margir srudentar. manga studenter. many students

(6)

ISc but not MSc has non-nominative subjects. a. Honum var gefinn hestur. horse he-DAT was given b. *Honom blev givet en hast. him was given a horse

(7)

ISc but not MSc has so-called stylistic fronting (the fronted category is in boldface): a. Fram hefur komia aa fiska6 hefur veria i leyfisleysi. forth has come that fished has been illegally h. *Fram har kommit att fiskat har blivit olagligt. forth has come that fished has been illegally c. Det har kommit fram att det har blivit fiskat olagligt EXP has come forth that EXP has been fished illegally 'It has been revealed that illegal fishing has taken place.'

548 Ian Roberts and Anders Holmberg

(8)

ISc but not MSc has V-to-T in embedded clauses (both are V2 in main clauses): a. I>ao er greinilegt aa Anders talar ekki islensku. it is obvious that Anders speaks not Icelandic b. Det ar uppenbart att Anders inte talar isUindska. it is obvious that Anders not speaks Icelandic 'It's obvious that Anders doesn't speak Icelandic.'

(9)

ISc but not MSc has relatively rich subject verb agreement. a. E.g tala/�u talar/via t(ilum islensku. b. Jag talar/du talar/vi talar islandska. ' I speak/you speak/we speak Icelandic. '

The claim made by Platzack 1 987 is that these differences are the result of different settings of one binary parameter. The precise formulation of the parameter has since undergone certain changes and refinements reflecting the evolution of syntactic theory; see Falk 1 993, Holmberg and Platzack 1 995, 2005. However, the basic idea has remained the same since Platzack 1 987, and can be expressed as follows: ( I O) (a) In all the languages a finite sentence must have a nominal category in SpecIP or 1. 4 (b) In all the languages I has an EPP-feature which requires a specifier for I . (c) ISc has a nominal Agr in I, spelled out as subject-verb agreement, MSc doesn't. That is to say, the parameter is the feature value of Agr: nominal Qr non­ nominal.5 The conditions . ( l Oa) and ( l Ob) may · be ' universal, or more plausibly, are themselves settings of two independent parameters. ,

cl

: In Holmberg and Platzack 2 00 5; Oa) is fonnulated tri ::�pedtically require a nominative NP in I or SpecIP in fmite clauses. In Platzack (2004) and Holmberg and Platzack (2005) what is required is a person feature in SpecIP or I. In tenns of a theory with valued and unvalued features (following Chomsky 200 1 ), assume that finite C contains an unvalued person-feature [u-person] which needs a person­ marked category in SpecIP or in I to get valued. ISc has a person-marked Agr in I, spelled out as agreement, but MSc doesn't, so a person-marked NP has to move to SpecIP; the facts in (5)-(9) follow as laid out in the text. S Plausibly Agr is completely absent from I in MSc. This would be a special case of 'non-nominal Agr' . 4

On the role ofparameters in Universal Grammar 549

The effect of ( 1 Oa,b,c) in the case of MSc is that the highest nominal argument invariably moves to SpecIP, or else a nominal expletive is merged there (or moved there; see Holmberg 2002), satisfying ( l Oa) and ( l Ob) at the same time. In ISc Agr in I will always satisfy ( l Oa), while ( l Ob) will be satisfied by moving some category to SpecIP, subject only to loca lity. Often the category closest to I is the nominative subject, in which case it will move to SpecIP, the result then being superficially identical to a MSc sentence. But if the nominative subject isn't closest to I, some other category ends up in SpecIP; this is how oblique subjects as well as stylistic fronting are derived. The Scandinavian languages are V2 languages. This means that I in main clauses moves to C . In main clauses ( l Ob) can therefore be satisfied by moving a topicalized or focused category to SpecCP; this is how (Sa) is derived. In MSc, too, I moves to C in main clauses, so ( l Ob) is satisfied by the adverbial in SpecCP in (5b). Condition ( l Oa) still re�uires a nominal category in SpecIP, though, hence the obligatory expletive. The contrast between (8a,b) follows if nominal Agr in I has the effect of attracting V to I; see Roberts 1993, Rohrbacher 1994, Vikner 1 995. We return to this issue below. The reason why we have picked this example of a parameter is that the correlation of the properties (5)-(9) has stood the test of time quite well. As noted in Platzack and Holmberg 1 989, Old French, Middle English, and Yiddish exhibit the ISc cluster of properties, while Modem English exhibits the MSc cluster, as does Modern French, with some caveats to which we return below. Diachronic investigations have shown that the appearance of the Modem MSc cluster correlates rather well with the loss of agreement, particularly person agreement; see Falk 1 993 ; on French and English, see Roberts 1 993. Notably, since 1 987 the Mainland Scandinavian dialects have been subject to a significant amount of research, uncovering an impressive range of syntactic variation; see Holmberg and Sandstrom 1 996, Holmberg and Platzack 2 005, and the papers in Vangsnes et al. 2004. However, as regards the properties (5-9) the dialects behave as predicted by the theory sketched above in ( 1 0): there is one known dialect, the A lvdalen dialect, which belongs in the ISc camp. All other dialects, as far as we know, clearly exhibit the MSc pattern. Given the range of dialectal variation found in

6

The EPP condition is suspended in the case of yes/no questions and a few other

verb initial structures -

.

5 50 Ian Roberts and Anders Holmberg

other syntactic domains, this is a striking generalization, which can be explained by the parameter-based theory sketched above in ( 1 0). 7 The theory does not predict that the ISc properties will always cluster. It allows for the possibility that a language has the positive setting for ( I Oc), yet does not, have� say, oblique subjects. This is so if allowing oblique subjects depends on other parameters in addition to ( l Oc); to start with, the language should have non-structural case. 8 Styl istic fronting, too, may depend on other parameters than ( 1 Oc). Parametric theory predicts that a cluster dependent on a given parameter setting will reappear in language after language, all else being equal. But all else is seldom equal. The best chance of observing cross-linguistic correlations is in closely related languages or dialects, but even then other parameter settings may complicate the picture, masking the effects of a parameter setting. The place of V-to-l in this connection has been subject to much controversy. One often noted problem is that Modem French patterns like MSc with respect to (5,6,7) but has V-to-I. This has generated much discussion about just how much verb-agreement is required for V-to-l to be licit; French has some subject-verb agreement, but not much (see Roberts 1 993, Rohrbacher 1994, Vikner 1 995b, Koeneman 2000). In recent work B iberauer and Roberts 2005 have argued that V-to-I may be triggered by other verb-inflection-related properties than agreement. They note that the Romance languages � including French, have a richer tense system than the Germanic languages, and they propose that rich tense, just as well as rich agreement, may trigger V-to-I. So while V-to-I in the ISc languages may be an effect of nominal Agr, this is not a criterial property. A further complication is that most varieties of Fa.roese don't have V-to-I, although they do exhibit other ISc properties. In any case, the theory in ( 1 0) is not dependent on V-to-I being included in the cluster of properties. The only dialect apart from A lvdalen which has retained subject-verb agreement is the Hallingdalen dialect in Norway, which, however, only has number agreement. This is not sufficient to set the parameter the ISc way, if [person] has the pivotal role alluded to in footnote 4. There is at least one dialect where V-to-I is reported to occur in embedded clauses: the dialect of Kronoby in Finland, but where no other ISc properties are evident. See the text below on the role of V-to-I . 8 There are dialects in Sweden which have dative case, but have not been reported to have oblique subjects or other ISc-like properties. Whether sentence-initial oblique NPs in the ISc languages are actually subjects rather than, say, fronted topics in constructions corresponding to (6a) in all varieties of ISc is a highly controversial question. Insofar as the NPs satisfy the EPP it does not matter for our purposes whether they exhibit all the canonical subject properties or not. 7

On the role ofparameters in Universal Grammar 55 1

Now consider (5)-(9) in a rule-based theory. Assume, for the sake of argument, that UG does not specify ( I Oa,b), and that the parameter ( I Oc) does not exist either. Instead, the contrast between (5a,b) would be the result of ISc having, say, a language-specific rule of expletive deletion, absent from MSc. Similarly, the contrast between (6a,b) would be the result of ISc having a ru le of oblique NP movement to specIP, absent from MSc (also crucially in those varieties of MSc which have a dative case). The contrast between (7a,b) would be the result of IS e having a rule of stylistic fronting, absent from MSc, and the contrast between (8,b) a result of ISc having a rule moving V to I, absent from MSc. The contrast between (9a,b), finally, is a matter of ISc having a paradigm of inflected verb forms absent from MSc. The theory is observationally adequate, obviously, but · makes no pred ictions whatsoever regarding the correlation of the properties (5-9). I t implies that the properties are independent, and that any correlations are accidental. On the basis of what we know about these languages at present it seems highly plausible that the properties are not independent. A parameter-based theory roughly along the lines of Platzack 1 987 can explain the correlations. One aim of Newmeyer 2004 is to argue that typo logical generalizations can be explained by "independently needed principles of performance" (p. 1 8 1 ), mainly following Hawkins 2004. As we doubt that any such principles can explain the correlations in (5-9), we conclude that the parameter-based theory has been vindicated. 4. Conclusion

We recognize the importance of distinguishing linguistic variation which is due to extragrammatical factors from variation which is parametric and . thus predictable from properties of UO (see Barbiers 2005 and Adger and Smith 2005 for interesting recent contributions). Newmeyer's approach, which is to deny the existence of parameters, is ill-conceived and not likely to further our understanding of the interplay of UG and extragrammatical factors in explaining cross-linguistic variation. To pursue Baker's metaphorical correlation with chemistry a little further, one may see something l ike Baker's PH as a precursor to a periodic table for languages; we agree with Baker that this is an optimistic view but one worth striving to maintain. Against this background, Newmeyer appears like a 1 9th-century chemist arguing against the abstractions of the early conceptions of the periodic table and in favour of a return to traditional theories of the nature of the elements, humours, etc. Fortunately, chemists chose the optimistic,

552 ian Roberts and Anders Holmberg

difficult, abstract path. We don 't see any good reason why to day ' s li ngu ists should not follow their example. References Adger, David and Jennifer Smith. 2005. Variation and the minimalist program. In Syntax and variation. Reconciling the biological and the social, ed. by Leonie Com ips and Karen P. Corrigan. 1 49- 1 78. Amsterdam: John Benjamins. Alexiadou, Artemis and Elena Anagnostopoulou. 1 998. Parametrizing AGR: word order, V-movement and EPP-checking. Natural Language and Linguistic Theory 5: 1 - 32. Baker, Mark. 200 1 . The Atoms ofLanguage. New York: Basic Books. Barbiers, Sjef. 2005. Word order variation in three-verb clusters and the division of labour between generative linguistics and sociolinguistics. In Syntax and variation. Reconciling the biological and the social, ed. by Leonie Com ips and Karen P. Corrigan. 233-264. Amsterdam: John Benjamins. Biberauer, Theresa and Ian Roberts. 2005. Subjects, Tense and verb-movement in Gennanic and Romance. Talk given at the 5th Asian GLOW Meeting, Jawaharlal Nehru University, Delhi. Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton. Chomsky, Noam. 200 1a. Derivation by phase. In Ken Hale: A Life in Language, M. Kenstowicz, 1 -59. Cambridge Mass.: MIT Press. Cinque, Guglielmo. 1 999. Adverbs and functional heads: a cross-linguistic perspective. New York: Oxford University Press. Clark, Robin and Ian Roberts. 1 993. A computational approach to language learnability and language change. Linguistic Inquiry 24:299-345. Dresher, Elan. 1 999. Charting the learning path: cues to parameter setting. Linguistic Inquiry 30. 1 : 27 - 67. Falk, Cecilia. 1 993 . Non-referential subjects in the history of Swedish. Doctoral dissertation, Department of Scandinavian languages, Lund University. Gilligan, Gary. 1 987. A cross-linguistic approach to the pro-drop parameter. PhD Dissertation, University of Southern California. Greenberg, Joseph. 1966. Some universals of grammar with particular reference to the order of meaningful elements. In Universals of Language, ed. by Joseph Greenberg, 73- 1 1 3 . Cambridge, MA: MIT Press. Guasti, Maria-Teresa. 2002. Language Acquisition. Cambridge, MA: MIT Press. Harris, Alice and Lyle CampbelL 1 995. Historical syntax in cross-linguistic perspective. Cambridge: Cambridge University Press. Hawkins, John A. 2004. Efficiency and complexity in grammars. Oxford: Oxford University Press. Holmberg, Anders. 2002. Expletives and agreement in Scandinavian passives. Journal ofcomparative Germanic linguistics 4 : 85- 128. Holmberg, Anders and Christer Platzack. 1 995. The role of inflection in Scandinavian syntax. Oxford and New York: Oxford University Press.

On the role ofparameters in Universal Grammar 553 Holmberg, Anders and Christer Platzack. 2005. The Scandinavian languages. In The Oxford handbook of comparative syntax, ed. by Guglielmo Cinque and Richard Kayne, 420-458. Oxford: Oxford University Press. Holmberg, Anders and GOrel SandstrOm. 1 996. Scandinavian possessive constructions from a Northern Swedish viewpoint. In Microparametric syntax and dialect variation, ed. by James Black and Virginia Motapanyane. Amsterdam: John Benjamins. Hyams, Nina. 1 986. Language acquisition and the theory of parameters. Reidel: Dordrecht. Kayne, Richard S. 2000. Parameters and Universals. New York: Oxford University Press. Koeneman, Olaf. 2000. The flexible nature ofverb movement. LOT, University of Utrecht. Lightfoot. David. 1 99 1 . How to set parameters. MIT Press. Lightfoot, David. 1 999. The development oflanguage. Oxford: Blackwell. Manzini, Rita and Kenneth Wexler. 1 987. Parameters, binding and learning theory.

Linguistic Inquiry 1 8:41 3-444.

Newmeyer, Frederick J. 2004. Against a parameter-setting approach to language variation. In Language Variation Yearbook, Volume 4, ed. by Pierre Pica, Johan Rooryck and J. van Craenenbroek, 1 8 1 -234. Amsterdam: Benjamins. Platzack, Christer. 1 987. The Scandinavian languages and the null-subject param eter. Natural Language and Linguistic Theory 5: 377-401 . Platzack, Christer. 2004. Agreement and the person phrase hypothesis. Working papers in Scandinavian syntax 73, 83- 1 12. Department of Scandinavian Languages, University of Lund. Rizzi, Luigi. 1 982. Topics in Italian Syntax. Dordrecht: Foris. Roberts, Ian. 1 993. Verbs and diachronic syntax. A comparative history of English and French. Dordrecht. Kluwer. Roberts, Ian. In preparation. Diachronic Syntax. Oxford: Oxford University Press. Roberts, Ian and Anna Roussou. 2003. Syntactic change: a minimalist approach to grammaticalisation. Cambridge: Cambridge University Press. Rohrbachert Bernhard. 1 994. The Germanic languages and the full paradigm: a theory of V-to-I raising. Ph.D. dissertation, University of Massachusetts, Amherst. Vangsnest 0ystein, Anders Holmberg and Lars-Olof Delsing (eds.). 2003. Dialektsyntaktiska studier av den nordiska nominalfrasen. Oslo: Novus. Vikner, Sten. 1 995. VO-to-IO movement and inflection for person in all tenses. Working papers in Scandinavian syntax 55: 1 -27. Department of Scandinavian Languages, University of Lund. Wexler, KeIUleth. 1 998. Very early parameter setting and the unique checking constraint: A new explanation of the optional infinitive stage. Lingua 1 06,

1 998.

Welsh VP-ellipsis and the representation of aspect Alain Rouveret 1 . VP ellipsis in Welsh: some preliminary observations Welsh VP-ellipsis (henceforth VPE) roughly displays the same distribution and range of discourse functions as English VPE. In the general case, auxiliary and modal verbs are retained, lexical verbs are replaced by

gwneud 'do ' .

(l)

Mi

geith Mair aros y nos

PRT can

Mair stay

a

geith Sioned hefyd

the night and

can

Sioned too

' Mair can stay the night and Sioned can also.' (2)

Prynodd Sion y llyfr hwn bought

a

Sion the book this and

gwnaeth

Mair hefyd

did

Mair too

'Sion bought this book and Mair did too.' The contrast between

( 1 ) and (2)

is highly reminiscent of the one di splayed

by English auxiliaries and main verbs. Careful examination, however, reveals that this parallel ism is spurious and that Welsh

gwneud behaves

quite differently from English do. First,

gwneud can be found in all finite sentence types, not just

interrogative and negative clauses, but also affinnative declarative ones, not just VP frontingIVP ellipsis constructions, but also sentences containing no topicalized nor contrastively focused constituent. It seems reasonable to assign to the gwneud occurring in (3 )

Mi

wneith y ceffyl

PRT

did

(2) the same analysis as to

the one in (3).

fwyta'r moron i gyd

the horse eat the carrots all

'The horse ate all the carrots. ' Second, the extension of gwneud-substitution i n Welsh differs from what is observed in English and in other languages. It is obligatory with transitive eventive predicates, cf.

(4):

Welsh VP-ellipsis and the representation ofaspect 555

(4)

·

·

*Prynodd Sian y llyfr hwn bought Sian the book this

a and

phrynodd Mair hefyd bought Mair too

In this respect, Welsh differs from Hebrew (cf. Doron 1 990), Irish (McCloskey 1 99 1 ), and Finnish (Holmberg 200 1 ), where the ellipsis site can be 'governed ' by a raised finite transitive verb. Why is (4) ungrammatical? It corresponds to the output of a derivation which one could expect to be we1l-formed in a language in which both VPE and verb raising are available. Contrary to English, Welsh falls under this characterization, since main verbs also move to the inflectional domain. In the representation associated with (4), namely (5), the target of el lipsis is the VP complement of finite I, emptied of the bare verb which has inoved to I to pick up the inflectional endings residing there or check the corresponding features. One of the tasks facing the analysis is to determine why (5) is ill-formed, whereas (6), the representation corresponding to (2), is perfect (in (5) and (6), the antecedent phrase is bracketed and the elided VP is explicitly indicated by striking through the unpronounced material): (5 ) (6)

Prynodd Sian [ t y lIyfr hwn ] a phrynodd Mair t y Il),fr ltV/R hefyd Prynodd Sian [ t y lIyfr hwn ] a gwnaeth Mair t hf)'FH:l )' Ilyfr IIlli'R hefyd

It is interesting to note that the languages which display well-formed equivalents of (4) generally do not present counterparts of (2). This holds true for Hebrew which does not have do support As for Irish, the equivalents of (2) seem to be very rare and lim ited to some dialects. This suggests that the ungrammaticality of (4) in Welsh is tied up with the availability of (2). But Welsh also displays VPE sentences in which the main verb of the first conjunct must be repeated ill the second. This phenomenon, which has been studied and analyzed by Jones ( 1 999), definitely confirms the existence of a link between the availability of gwneud 'do' and the iII­ formedness of constructions l ike (4). When for some reason, gwneud cannot be resorted to, the main verb can and must be repeated in the second conj unct. Among the verbs for which gwneud is not available as a substitute, one finds predicates such as gwybod 'to know' . Interestingly, gwneud cannot be used in a simple declarative sentence whose main verb is gwybod, cf. (7c). -

·

.

556 Alain Rouveret

(7)

a.

Mi

wyddet ti

bopeth

a

wyddet ti

bopeth

a

knew y ou everything and 'Y ou knew everything and I did too.'

PRT b. *Mi

PRT c. *Mi

PRT

knew

wnait did

you

ti

you

everything and

wybod bopeth know

gwyddwn

hefYd

too

knew

gwnawn

i hefyd

did

I

too

everything

At this point, it is legitimate to conclude that having access to the inflectional domain is a

fonn to

necessary,

but not sufficient condition for a verbal

non-su bstitu ti on by gwneud. In non elided sentences, both the predicates of the prynu c lass and those of the gwybod class raise to the inflectional domain, as the fonn of the fi rst conjunct in (2) and (7a) suffice to establish. Yet, resort to gwneud is obligatory when VPE involves prynu, excluded when it involves gwybod. An additional factor must be taken into consideration . Some predicates are inherently incompatible with gwneud, as shown by their inability to Occur 'su rviv e'

ellipsis, i.e. for its

in simple (non elided) gwneud-initial sentences, cf. (7c). These predicates invariably survive ellipsis.

2. VPE and classes of verbs Let us consider more closely the predicates which

show an exc epti onal VPE constructions (for which the resort to gwneud is either excluded or optional). Jones ( 1 999) distinguishes two classes. Class 1 is constituted by verbs which show a specific use of tenses. In Welsh, verbs in the present tense do not allow a referential p resent time behavior in

interpretation: either they refer to the future or denote an habitual activity.

Gwybod ' to know' is exceptional in this respect, as well as gweld ' to see', clywed 'to hear', credu 'to believe', coelio ' to believe', which assign a present time i nterpretati on to -ith fOl:1lls , and an ':irnperfect,:interpretation to -ai forms and hoffi 'to like', "licio 'to like', whose ,forms "-ai/';'asai have no past use. I The resort to gwneud is excluded in this case.

I The su ffix -ith is one of the 3rd person singular endings of verbs in the present tense and is used here to refer to the whole present tense paradigm ; -af is the third person singular ending of verbs in the imperfect tense; -asa; is the 3rd person singular of the p luperfect.

Welsh VP-ellipsis and the representation ofaspect 557

(8) a. Mi

wela' i o a gweli see I it and see 'I see it and you see it too.' · ? h . Mi wela' i 0 a gwnei PRT see I it and does PRT

· ·

,

hefyd di you too hefyd di you too

Class 2 consists of three verbs, which show a regular use of tenses, but display irregular morphophonemics: mynd 'to go', dod 'to come', cael 'to get'. With these verbs, gwneud-substitution is optional. yn 61 a dof i hefyd (9) a. Mi ddoi di PRT come you back and come I too ' You' ll come back and I will too.' b. Mi ddoi di yn 61 a gwna i hefyd PRT come you back and do I too

The general claim I wish to develop is that the variation observed between class 1 verbs and the other predicates should be traced back to the '. . interaction between the Aktionsart profile of the verbal root, grammatical Aspect and the structural make-up of the vP phase in Welsh. In order to : substantiate this claim, it is necessary to introduce the following general assumptions. The first one is that the predicate phrase is headed by a light v, which encodes the properties of Voice, Agentivity, Transitivity, cf. Chomsky 1 995 . The second one is that the phrase structure of vP includes at . least one non-lexical projection between vP and VP, which, following Travis 2000, I will identify as the projection of Aspect. I will assume that Aspect represents grammatical Aspect, not Aktionsart. The resulting .architecture is the one schematized in ( 1 0): .

.•

·

(1 0) 2. 1.

... T [vp V [AspP [ Aspect ] [vp [v Aktionsart ]]]] Stative predicates

The semantic asymmetry between the verbs belonging to class 1 and the other ones is reminiscent of the contrast found in many languages between stative predicates and eventive ones in the present tense. It is often the case, in particular in the languages where tense has no morphological realization (cf. Dechaine 1 993 on Haitian), that the state denoted by stative predicates is interpreted as holding during a time interval which includes the speech �ime, whereas eventive predicates are assigned a present time interpretation only if the event is construed as a state, either as a temporary state relative

558 Alain Rouveret

to a bounded interval (continuous interpretation) or as a permanent state (generic interpretation). In Welsh, whereas eventive predicates in the simple present tense are interpreted as generic/habitual or as referring to the future, the predicates included in class 1 have a present time reference. I will consider that they qualify as 'stative predicates', provided that ' stative' is understood here as 'non eventive' and that one acknowledges the possibility that stative predicates not included in class 1 exist. I will assume that when V is a stative predicate, the governing Aspect category contains a zero aspectual morpheme. (1 1)

...

T [vp V [ASpP [ASP 0 ] [ .. Stative Verb .

'"

]] ]

We obv iously want the lack of overt realization of Asp to be principled and constrained. A very tentative option consists in assuming that, by itself, (2) has no designated meaning and inherits it from the temporal or the lexical environment (in a way to be made precise): (2) is inserted when its value is recoverable from the local context. This is the case for example when there is a 'natural connection' between the viewpoint aspect carried by the sentence and the situation aspect inherent to the predicate. In the other situations, when a mismatch exists between the two, overt realization of Asp is obligatory. When a sentence containing a stative predicate is assigned an imperfective interpretation, (2) is a legitimate choice. But the overt realization of the aspectual morpheme is necessary to give an imperfective profile to a construction containing an eventive predicate or to select the progressive value of a continuous predicate (in those cases, Welsh resorts to the verb bod 'to be' , combined with the particle yn). 2. 2. Gwneud-constructions

There is a large distributional and interpretive overlap between the gwneud + VNP construction (cf. (3» and the simple form construction? Gwneud­ constructions don't allow the present time referential interpretation in the present tense, simple forms constructions don't either. Gwneud cannot cooccur with an aspect marker, simple forms obviously cannot either:

2 VNP

here is the abbreviation of Verb-Noun Phrase. Whenever a verbal root is

trapped inside the VP, it is realized as a verb-noun. Whether the verb-noun

in this

case is a default form resulting from the addition of features to the verbal root on the PF side or is derived through affixation in the syntax, as assumed in ( 1 4), can't decide here.

I

Welsh VP-ellipsis and the representation of aspect 559

( 1 2)

*Mi wneith Sian fod yn gweithio PRT does Sian be PROG work

But gwneud-constructions block the habitual reading in the present and past tenses. Only simple fonns can be used in this case or periphrastic forms involving the habitual fonns of bad 'be', namely bydd or byddai. pob dydd ( l 3 ) a. ?*Fe wnai e fynd i 'r ysgol to the church every day PRT did he go b. al i'r ysgol bob dydd went to the church every day 'He/she went to the church every day' mynd i'r ysgol bob dydd c. Byddai 'n PROG go to the church every day was ' He/she went to the church every day' The availability of the imperfective interpretation is presumably a necessary condition to derive the habitual reading and perfective gwneud cannot be construed as an imperfective. It is safe to conclude that gwneud is not just a support-element aJlowing the spell-out of the tense and agreement features of the category T: it carries an aspectuaI value, namely [perfective], and provides finite clauses with a 'processual profile' (cf. Fife 1 990). In (14), gwneud is directly merged into the v position. 0 raises to v and merges with gwneud. The resulting combination is [perfective], hence [-habitual], [-continuous], [-progressive] .

(14)

.. . T [vp [v gwneud ] [ASpP [Asp 0 ] [bnP Aff [Rootp gwe ith- . . . ]]]]

(1 4) thus illustrates a situation where 0 receives its interpretation from the local context (through merger with gwneud). 2.3. Simplefarms ofnon-stative predicates The habitual reading of eventive predicates which is blocked in gwneud­ constructions is available in the present and the imperfect in sentences where the predicate is a simple form of the verb. The same constructions also allow the punctual event reading - an event which is situated in the . future when the verb is in the present tense and in the past when the verb is past. I will assume that simple forms are not defective for Aspect, but coexist with a phonologically null aspectual morpheme in all their uses.

560 A lain Rouveret ( 1 5)

.. . T [vp V [AspP [Asp 0 ]

[RootP gweith- '"

]]]

gwneud is not inserted into v, hence does not impose a processual profile to the sentence. If one assumes that [±habitual] constitutes an autonomous aspectuaJ feature and that 0 has no aspectual value by itself, both the habitual interpretation and the punctual event interpretation should be available in the absence of gwneud. 3. A

pbasal app roach to VPE

The differential behavior of the various classes of verbs in VPE constructions can be subsumed under ( 1 6). ( 1 6)

Only the verbal forms which have access to v (whether they move to v or are directly merged there) survive ellipsis.

( 1 6) can be made to follow from Chomsky's 200 1 theory of phases, in partiCUlar from the PIC, provided that an additional assumption concerning ellipsis is made, namely ( 1 7): 3 ( 1 7)

An ellipsis is a vP phase that is spelled out as null at PF.

The PIC states that the complement of v, namely VP (or AspP in ( 1 1 ), ( 1 4), ( 1 5» , stops being accessible to syntactic operations when the derivation has reached the next highest phase, namely CP, because at that time, the complement of v has already been sent to PF and LF. If the phonological component is accessed cyclically, in a way which is determined by the PIC, the decision to spell out the complement of a vP phase or not must have been taken when the next highest phasal head, C, is merged. As for the verbal elements which are merged into v or have moved to v during the vP phase, they are not affected by this decision, they remain accessible and are free to move to T in order to satisfy its morpho-syntactic requirements. It still must be explained why eventive predicates cannot be repeated in VPE constructions and must be substituted by gwneud. We have to assume that in this case, economy considerations dictate that the merger of gwneud 3

( 1 7) reproduces Holmberg's 200 1 characterization of ellipsis structures. The difference between the two accounts is that Holmberg argues in favor of a LF­ copying analysis of VP ellipsis, whereas I claim that a PF-deletion approach makes the right predictions for the Welsh data (cf. section 4 below). Crucial to this account is the claim that gwneud and the main verb belong to the same phase.

Welsh VP-ellipsis and the representation of aspect 56 1

at the v level (and its subsequent movement to T) is less costly than the movement of the verbal root to v. 4 The reason why these economy considerations appear to be relevant in VPE constructions and not in ordinary sentences (if they were, all verb initial sentences would be gwneud constructions) could be traced back to the fact that in the former, there is a strong preference for not repeating any of the lexical material which is already found in the antecedent clause. 4. Some

theoretical implications

In the VPE constructions involving one of the morphologically irregular verbs included in class 2, two options are available: either repetition of the verb or gwneud-substitution (cf. (9a)-(9b» . One way to interpret this dual behavior is to propose that it reflects the dual status of the irregular verbal morphology. Irregular morphology can be treated exactly as the regular one, ie. as a grammatical dimension whose locus is T. If this option is chosen, the derivation follows the same route as in the cases involving a non-stative predicate: gwneud- substitution is obligatory because the corresponding predicates are eventive. But irregular verbal morphology can also be assigned a lexical status (exactly as number in Arabic broken plurals). When treated as a lexical property, it cannot be dissociated from the verbal root. Either the verbal form is inserted fully inflected into syntactic derivations, or the inflectional morphology itself originates within the vP and its merger with the verbal root occurs within the vP. In this case, there is no possibility for gwneud to -be inserted into v. Instead, the irregular form is already morphologically complete at the vP level. If the preceding analysis is on the right track, a clear picture of the division of labor between the various components of the Computational . System emerges. In the phasal approach, either the main verb in the elided clause raises to v or gwneud is merged into v and the main verb is deleted together with the VP. V-to-v movement necessarily takes place in the overt syntax, because the verbal form in v must subsequently move to T. On the other hand, the fact that ordinary (eventive) predicates and morphologically irregular verbs are treated differently is a strong indication that VPE cannot exclusively be a matter of semantic isomorphism between the two 4 This situation recalls expletive insertion, as Chomsky characterizes it. There is a

preference for Merge over Move. The merger of the expletive is less costly than the movement of an argument to the m erger site.

562 Alain Rouveret

conj uncts. Sensitivity to morphological irregularity is no doubt a strong argument in favor of the PF approach to advocated by Chomsky 1 995.

VPE

References Chom sky Noam. 1 995. The Minimalist Program. Cam bridge, Mass.: MIT Press. Chomsky, Noam. 200 1 . Derivation by Phase. In Ken Hale. A Life in Language, Michael Kenstowicz (ed.), 1 -52. Cambridge, Mass.: MIT Press. Dechaine, Rose-Marie. 1993. Predicates Across Categories: Towards a Category­ Neutral Syntax. Ph.D., University of Massachusetts, Amherst. Doron, Edith. 1 990. V-Movement and VP-Ellipsis. Unpublished ms., Hebrew University of Jerusalem. Fife, John. 1 990. The Semantics o/the Welsh Verb. A Cognitive Approach. Cardiff: University of Wales Press. Holmberg, A. 200 1 . The Syntax of Yes and No in Finnish. Studia Linguistica 55, 1 40- 174. Jones, Bob Morris. 1 999. The Welsh Answering System. BerlinlNew York: Mouton de Gruyter. McC loskey, James. 1 99 1 . Clause Structure, Ellipsis, and Proper Government in Irish. Lingua 85, 259-302. Travis, Lisa. 2000. Event Structure in Syntax. In Events as Grammatical Objects : the Converging Perspectives of Lexical Semantics and Syntax, Carol Tenny and James Pustejovsky (eds.). S tanford : CSLI Publications. ,

A new perspective on event participants in psychological states and events Bozena Rozwadowska 1.

Introduction

In this squib I relate certain characteristics of the so-called stative subclass of Object Experiencer (OE) predicates in Polish highlighted in Bialy (2005) to argument structure properties of nominals derived from them. I will demonstrate that the observations made and conclusions reached independently by various scholars for different purposes are fairly consistent, which reinforces their validity and might provide supporting evidence for the importance of finer event structure distinctions at the lexicon-syntax interface in the spirit of Levin and Rappaport Hovav (1 999) (henceforth L&RH), Rappaport Hovav and Levin 1 998 (henceforth RH&L), Arad ( 1 998, 1 999), or Pylkkanen (2000), among others. Specifically, I will address the issue of the relevance of the number and status of event participants in those distinctions. I will also show that there is a correlation between the argument structure properties and the aspectual properties of stative OE predicates. 2.

Initial boundary eventualities

[n Rozwadowska (2003) it is suggested that psych-predicates denote complex initial-boundary events, i.e. a culmination followed by a state, and are a m irror image of telic action predicates, i.e. a process followed by a culmination. In ( l ) there is an example of a Subject Experiencer psych-verb denoting a mental state: a Polish verb rozumiec 'understand' and its prefixed perfective variant zrozumiec: ( l ) a.

Marek zrozumial to zadanie. Marc understand past-perf this problem ' Marc got to understand this problem. ' b. Marek rozumia! to zadanie. Marc understandpast-im perf this problem ' Marc understood this problem.'

564 Boiena Rozwadowska

c.

to zadanie. Marek rozumie Marc understandpres-imperf this problem ' Marc understands this problem. '

(A)tel icity tests, when applied to the above pair, give the following results: (2) a.

Marek zrozumial to zadanie w godzin�/godzin�. Marc understandpast_perf this problem in hour/hour ' Mark got to understand this problem in an hour /*for an hour' . b. Marek rozumial to zadanie godzin�/*w godzin� Mark understood this problem for an hour!in an hour

The perfective variant accepts the time span (TS) adverbial, whereas the imperfective one co-occurs with the durative adverbial. However, with zrozumiec, the TS adverbial measures the time before the event occurred, as with atelic predicates and not as with telic predicates. The perfective variant is not atelic either, as indicated by its incompatibility with the durative adverbial in (2a). Time-related entailments of accomplishments and psych­ predicates give reverse patterns, summarized in (3a) and (3b) respectively: (3) a. Activity/accomplishment: Imperfective at ( does not entail perfective at t ' < I Perfective at time I entails imperfective at t ' < t b. Initial boundary! resulting state: Perfective at t entails imperfective at I ' > t Perfective at t does not entail imperfective at t ' < I Imperfective at I entails perfective at ( ' < t The conclusion is then that Polish initial boundary predicates do not pass standard telicity tests and have reverse entailments to those of telic predicates. Related to that is the difference in behavior between terminal vs. initial boundary verbs with respect to other constructions sensitive to the temporal development of events, illustrated in (4-6): (4) a. On pisa1imperf, pisa1inperf aZ napisalperf ten list. 'He was writing, writing until (finally) he wrote this letter.' b. On pisal ten list, ale go nie napisal 'He was writing this letter, but he has not written it. ' # c. On napisal ten list i nadal go pisze. 'He has written this letter and he is still writing it.'

Event participants in psychological states and events 565

(5) a. *On rozumia1im perf, rozumialimperf aZ zrozumiatperfto zadanie. ' He understood until (finally) he got to understand this problem.' b. *On rozum ial to zadanie, ale go nie zrozumial. ' He understood that problem, but he has not understood it. ' c. On zrozumial to zadanie i nadal je rozumie. ' He has understood this problem and he still understands it. ' 'write' (6) a. *Napisal ten list i nadal go pisze. ' He has written this letter and he is sti ll writing it.' b. Zrozumial ten problem i nadal go rozumie. 'understand' ' He has understood this problem and he still understands it. ' c. Pokochal ji\. i nadal kocha. ' love' ' He started/got to love her and he still loves her.' ' interest' d. Ta ksiilZka go zainteresowala i nadal go interesuje. 'This book started to interest him and it still interests him. ' Another characteristic of initial-boundary predicates is their co-occurrence with result-state (RS) adverbials (cf. Pilion 1999). In Polish there is a distinct prepositional phrase bounding the duration of a state resulting from some initial change, illustrated in (7): I na cal)' semestr (7) a. Marysia rozchorowala si\: reflel onto the whole semester and Mary i I Ipast-p erf musiala wzi�c urIop. muslpast take leave 'Mary became ill for the whole semester and had to take a leave.' b. Ten problem zainteresowal ich tylko na chwil�. this problem interes!past_perf them only onto while 'This problem got their interest only for a while. '

Note that the RS adverbial is incompatible with verbs in the imperfective: (8) a. * Marysia chorowala na caly semestr i musiala Mary i llpast-i mperf onto the whole semester and mustpast wzii\.c urlop. leave take b. *Ten problem interesowal ich tylko na chwilcr. this problem interestpast-imperf them only onto while

566 Botena Rozwadowska

To conclude, perfective verbs denoting telic events co-occur with the TS adverbials, which measure the duration of the process leading up to its termination point, whereas perfective verbs denoting initial boundal)' events co-occur with the RS adverbials, which measure the duration of the state resulting from its inception. Neither RS- nor TS-adverbials co-occur with atelic or imperfective eventualities. 3.

Stative and eventive ObjExp verbs

In Rozwadowska (2003) it is suggested that all psych verbs (both Subject Experiencer verbs and Object Experiencer verbs) follow the above pattern and denote complex events with the culmination point at the beginning followed by a state. B ialy (2005 : 1 5 1 - 1 52) convincingly argues that in fact there is a difference between SubjExp verbs (taken to be always states) and stative ObjExp verbs on one hand and non-stative ObjExp verbs on the other hand. Bialy demonstrates that the pattern discussed above is characteristic only of the members of the former class, whereas the non­ stative ObjExp behave in a different way. This observation is one of his arguments for the subdivision of the Polish ObjExp predicates into the stative and non-stative (eventive) subclasses. The following non-stative ObjExp verbs illustrate this contrast: (9) a. ??Takie powitanie zdenerwowalo ich na 5 minut. 'Such a welcome angered them for 5 minutes.' 'n b. Reakcja Tomka oburzyla Zosi� na caly dzien. 'Tom's reaction irritated Sophie for the whole day.' As illustrated in (9), non-stative ObjExp verbs in the perfective do not admit the RS adverbial. Similarly, Bialy (2005 : 1 5 1 ) shows that the · entailments of non-stative ObjExp verbs are not exactly like those in (3) · . above: ( 1 0) a. Halas przerazil Mari�. the noise horrifypast-perf Mary b. Tomek zdenerwowal Zosi�. Tom irritatepast-perf Sophie

=/} Halas przeraZa Mari�. the noise horrifypres-imperf Mal)' Tomek denerwuje Zosi�. Tom irritatepres-imperf Sophie

Inspired by the ideas developed in L&RH, Bialy argues that stative ObjExp verbs (as well as SubjExp verbs) denote temporally dependent eventualities which unfold within the same temporal contours (i.e. at the same time),

Event participants in psychological states and events 567

whereas non-stative ObjExp describe temporally independent, sequential sub-events. He postulates (2005 : 1 55) that temporal dependency of L&RH and the coexistence of the two sub-eventualities of stative ObjExp verbs (as understood by Arad 1 998, 1 999) are equivalent and adopts the configurations as in ( 1 1 ): ( 1 1 ) a.

Stative ObjExp: t(

tn

(e\)

b. Non-stative ObjExp:

< ------

tn

Now, if we combine all the assumptions of the adopted approaches to event structure, we can reason as follows. Co-identification entails that there is only one event variable in the event structure, and thus only one structure participant (one event identifier) in the sense of RH&L. In a complex event with two temporally independent sub-events there are two event-variables, and thus two structure participants (two event identifiers). This view also entails that the two co-existing sub-eventualities share the same structure participant, namely the Experiencer argument. At the same time Arad's approach to stative ObjExp verbs implies that the trigger of the emotion (i.e. the Target/Stimulus participant) is present during the eventuality. Interestingly, this approach to ObjExp verbs seems to be correlated with the argument realization of nominals derived from the respective suh-types of psych verbs. One observation made by Bialy is that eventive psych verbs allow nominalizations with the przez 'by' -phrase as in (1 2), but that stative psych verbs do not, as in ( 1 3): ( 1 2) a.

przeraZenie Marka przez zlodziejal*zlodziejem scaring Markgen by thief/thiefinstr b. zdenerwowanie nauczycielki przez ucznial*uczniem by pupil/pupilinsfr teachergen making-angry

568 Botena Rozwadowska

Zosi�*przez Zosi� ( 1 3 ) a. zainteresowanie Marka interest Markgen ZosiainsJby Zosia 'John's interest in Zosia.' b. zachwyt Tomka Mari�*przez Mari� admiration Tomekgen Maryinstr/by Mary 'Tomek's admiration for Mary.' Bialy takes the presence of the przez 'by' phrase as indicative of a transitive complex event. In other words, those eventualities which allow the agentive phrase belong to the subclass of temporally independent complex events. Indeed (as confirmed by native speakers' intuitions) the nominalizations in ( 1 2) marginally allow the przez-phrase on the agentive reading, whereas such agentive reading is unavailable with the nominalizations in ( 1 3). Whi le the agentive reading is irrelevant for the psych phenomenon as such, it can serve as one of the tests for the division of psych roots into two different subclasses, one allowing the agentive shift easier than the other. What is more important, I believe, is the contrast between the impossibility of expressing the TIS argument (when it denotes an individual) in the instrumental case with the eventive class and its obligatoriness with the stative class. There is a clear complementary distribution in this respect, visible in ( 1 2- 1 3) above. Moreover, with stative ObjExp nominals, the TIS participant expressed in the instrumental case is obligatory in nominalizations. In contrast to that, the nominalizations of non-stative ObjExp verbs do not require the TIS participant to be overtly realized when they denote the resultant state: Przeratenie Marka wszystkim si� udzielilo. 'Mark's was shared by everybody.' b. Zdenerwowanie Marii byto czytelne dla k�dego. 'Mary' s was clear to everybody.'

( 1 4) a.

,

, ;

dlJgb'. . "

' "

( 1 5) a. Zainteresowanie Marka· *(historiV nie trwalo ' Mark's interest * (in history) did not last for a long time.' b. Zachwyt Tomka *(KasiV byl dla nas niezrozumialy. 'Tom's admiration *(of Kate) was beyond our understanding. ' If stative causativity involves coexistence of two eventualities and if the Stimulus, as Arad claims, has to co-occur simultaneously with the mental state in order for it to hold, its obligatoriness with nominalizations of the type illustrated in ( 1 3 ) and ( 1 5) follows. However, assuming RH&L's

Event participants in psychological states and events 569

approach, this participant is not a structure participant but a constant participant. This is not the case with temporally independent complex events. Here, the non-Experiencer argument is not obligatory in the nominalization of the second sub-event, because the relevant structure participant for its identification is only the Experiencer. The Stimulus is involved only in the first, causing sub-event. That the Stimulus is not necessary with the derived nominals in ( 1 2) and ( 1 4) is compatible with the idea that causation is independent of the mental state that follows the change of state. In other words, in this situation the Stimulus does not necessarily need to co-occur simultaneously with the mental state. So, the difference between complex sequential psych events and simple co-existing events is reflected in the contrastive argument realization 10 nom inalizations. So far, I have shown that the argument realization patterns of nominals derived from the two subclasses of ObjExp verbs are different. There is a clear complementary distribution both in the pattern itself and in the optionality/obligatoriness of the non-Experiencer (TIS) participant. Interestingly this contrast in nominalizations is correlated with the behavior of psych-verbs with respect to the aspectual tests discussed above. This leads to the conclusion that argument realization in nominals is sensitive to aspectual and event structure distinctions and provides additional support for theories which assume that the event structure level is of crucial importance at the lexicon-syntax interface. 4.

Psych nominals and structure participants

The second point I would like to make is the status of the participants of all psych eventual ities, irrespective of whether they belong to the stative or the non-stative type. The analysis of psych nominalizations developed extensively in Rozwadowska ( 1 997) shows that in that class we have a variety of argument-taking e-nominals (i.e. event nominals) derived from transitive psych verbs where there is no przez-phrase, and where the Experiencer must be realized as the possessive genitive, the other participant (TIS participant) appearing in an oblique case or in a prepositional phrase. The important generalization is that this pattern holds for all types of psych nominalizations (i.e. those related to SubjExp verbs and to ObjExp verbs). Note also that Bialy's examples quoted in ( 1 2) above refer to the possible agentive interpretation of psych-nominals, and can serve only as a diagnostic for the possibility of the agentive shift, which is an additional

570 Bozena Rozwadowska

pattern, on top of the primary non-agentive psychological interpretation, where the przez-phrase is not possible. The conclusion from the analysis of psych e-nominals developed in Rozwadowska ( 1 997) is that their argument realization is similar to that of intransitive e-nominals despite their dyadic nature. This leads to the distinction between external (non-psych) events and internal (psych-) events. The difference between the two is in the event identification (or its location). In other words, a psychological experience is located solely in the Experiencer participant, in a similar way as an intransitive activity event is identified by its Agent. In that sense, psych eventualities are ' intransitive' at the level of event structure. This ' intransitivity' was then understood as the number of event identifiers (i.e. participants which serve as the location of respective eventualities). In other words, psych eventualities are single participant eventualities with respect to event identification (like intransitive actions). The rules that capture the distribution of arguments in derived nominals are formulated as in ( 1 6) (c.f. Rozwadowska 1 997: 55): ( 1 6) a. by-phrase = an ergative marker (in the sense of Williams 1 987) licensed by complex event (in the sense of Grimshaw 1 990) iff the event is external b. Possessive phrase can accommodate only a single event participant The conclusion that the realization of arguments in derived nominals is sensitive to the event type (external vs. internal) and to the number of participants identifying an event is, I believe, very similar to RH&L's division of event participants into ' structure' participants and ' constant' participants. In other words, event-identifiers in Rozwadowska ( 1 997) can be viewed as structure participants in the theory developed in RH&L. All the above remarks provide a synthesis of various independent investigations. They seem to be consistent and convergent in many respects. Their mutual compatibility provides additional support for the importance of finer event structure distinctions and the importance of the number of event participants involved in various types of events. This brief synthesis of the state of the art in the domain related to psych verbs invites further research in that area.

Event participants in psychological states and events 57]

References Arad, Maya. 1 998. VP structure and the syntax-lexicon interface. Doctoral Dissertation, University College London. Arad, Maya. 1999. What counts as a class? The case of psych verbs. MIT Working Papers in Linguistics 35, 1 -23 . Cambridge, Mass: MIT Press. Bialy, Adam. 2005. Polish Psych Predicates at the Lexicon-Syntax Interface in Cross-linguistic Perspective. Frankfurt am Main: Peter Lang. Grimshaw, Jane. 1 990. Argument Structure. Cambridge, Mass.: The MIT Press. Levin, Beth and Maika Rappaport Hovav. 1999. Two structures for compositionally derived events. Proceedings of SALT 9, 199-223. Cornell Linguistics Circle Publications, Cornell University, Ithaca, NY. Pifion, Christopher. 1 999. Durative adverbials for result states. In Proceedings of the 18th West Coast Conference on Formal Linguistics, ed. by Sonya Bird, Andrew Carnie, Jason D. Haugen and Peter Norquest, 420-433 . Somerville, MA: Cascadilla Press. Pylkkanen, Liina. 2000) On stativity and causation. In Events as Grammatical Objects, ed. by Carol Tenny and James Pustejovsky, 4 1 7-444. Stanford: CSLI Publications. Rappaport Hovav, MaIka and Beth Levin. 1 998. Building verb meanings. In The Projection of Arguments. Lexical and Compositional Factors, ed. by Miriam Butt and Wilhelm Geuder, 97- 1 34. Stanford: CSLI Publications. Rozwadowska, Bozena. 1997. Towards a Unified Account of Nominalizations: External and Internal Eventualities. WrocJaw: Wydawnictwo Uniwersytetu Wroclawskiego. Rozwadowska, Bozena. 2003. Initial boundary and telicity in the semantics of perfectivity. In Investigations into Formal Slavic Linguistics, ed. by Peter Kosta. Joanna Blaszczak, Jens FraSek, Ljudmila Geist, Marzena .zygis, 859872. Berlin: Peter Lang. Williams, Edwin. 1 987. English as an Ergative Language: Theta structure of derived nouns. Chicago Linguistic Society 23.

A glimpse of doubly-filled COMPs in Swiss German Manuela Schonenberger 1.

Introduction1

This paper examines the distribution of doubly-filled COMPS (DFCs) in a small corpus of Swiss German containing spontaneous speech data. The corpus was originally compiled to study the acquisition of verb placement in Swiss German, and contains mainly child but also some adult data. The focus here lies on the acquisition of DFCs. To study this question an examination of the occurrence of DFCs in the input was performed. The specific aim was to investigate whether the distribution of DFCs in the adult as well as the child data could be governed by prosody rather than semantics, and this indeed seems to be the case. 2.

Doubly-filled

COMPS

DFCs are attested in many Germanic dialects, as in the examples in ( I ), taken from Penner and Bader ( 1 995 : 1 20). As shown by Lightfoot ( 1 979: 3 2 1 f) a.o. DFCs also occurred in diachronic data of English, illustrated in (2). ( I ) a. dea Hund [dea wo gestern d'Katz bissn hot] the dog who WOrel yesterday the-cat bitten has (Bavarian; Bayer 1 984) 'the dog that bit the cat yesterday' b. vi vet [hvem som snakker med Marit] : we know who that talks with Mary . . ' We know who is talking with Mary.' (Norwegiart; Taraldsen 1 986 ) ' ' ' ' c. kweten nie [wat dan · d'joengers geeten · eeri]t I know not what that the-children eaten have ' I don't know what the children have eaten . ' (West Flemish; Haegeman 1 992) .

I I am grateful to the following people for clarification of a number of questions: Agnes Bende-Farkas and liirgen Pafel concerning D-linking, and Matthias Jilka

and Astrid Krahenmann concerning prosody.

Doubly-filled comps in Swiss German 5 73

d . ik weet niet [wie of (dat) het gedaan heeft] I know not who if that it done has 'I don't know who did it.' (Dutch; Koster 1 986) (2) a. only the sight of hire [whom that I serve] (c. 1 3 86 Chaucer, Knight's tale) b. men shall wei knowe [who that I am] ( 1 485 Caxton, R67) There are two hypotheses about the distribution of DFCs in Swiss German. Penner and Bader ( 1 995) and Penner (1 996) state that DFCs in Bemese are optional with light wh-phrases (3a), but more or less obligatory with heavy wh-phrases (3b). (dass) er wohnt]. (3) a. I weiss [wo I know where that he lives 'I know where he lives.' b. I ha ne gfragt [i welem Huus ?\dass) er wohnt] . I have him asked in which house that he lives 'I asked him in which house he lives.' Cooper ( 1 995 ) proposes for Zurich German that DFCs are obligatory in embedded clauses which cannot be unambiguously identified as such, i.e. where embedded word order could be confused with that of matrix clauses. In (4a) the finite verb does not immediately follow the wh-phrase, while in (4b) it does. Thus (4b) looks like V2 if dass is not used. Ich wott wUsse [wo (dass) e vorhlisch z'Ubemachte] . . intend2sg to-overnight I want know where that 'I want to know where you intend to stay overnight. ' b. Ich wott wUsse [wo *(dass) e Ubemachtisch]. I want know where that ovemight2sg I want to know where you are staying overnight. ' '

(4) a.

I analysed the data in my corpus to find out whether either of these hypotheses can be extended to Lucemese. Since long wh-phrases, e.g. example (3 b), are often D-linked I tried to detennine whether the notion of

574 Manuela Schonenberger

D-linking might also play a role in the distribution of DFCs, rather than just 2 the sheer weight of a wh-phrase. 2. 1. Corpus data on adult Lucernese

I n my acquisition corpus on Swiss German I transcribed part of the input one of the children received. The input sample contains 20 1 wh­ complements produced by adult speakers of Lucernese. Most of these were produced by the child's mother. The verb placement used in wh­ complements is generally verb-final, but verb movement in wh­ complements is an option in Swiss German: 6 of the 20 1 examples show the V2 pattern. It is unclear which matrix verbs selecting a wh-complement license V2 besides the verb-final pattern.3 Still, the verb-final pattern is commonly used in wh-complements in contrast to Belfast English, which generally allows V2 in wh-complements independent of the matrix verb (Henry 1 995). In this study only finite verb-final wh-complements are considered: 32 examples of a total of 1 95 examples show DFCs. According to Penner and Bader ( 1 995) and Penner ( 1 996) the length of the wh-phrase plays a role in whether or not a DFC is used. No definition of ' light' and 'heavy' is given. I f the weight can be measured in number of syllables then the wh-phrases can be classified accordingly. I therefore subdivided the examples produced 2 Extendin g Pesetsky s ( 1 987) concept of D(iscourse)-linking from DPs to clauses, Penner and Bader ( 1 995 : 1 09) propose that in the alternating examples in (i), only (ib) can be D-linked. A D-linked clause fonns part of the conversational background. If the question of life on Jupiter is already being discussed, (ib) is felicitous and (ia) is not. They note that in the case of declarative clauses introduced by the complementizer dass 'that' it might be more appropriate to talk about presupposed clauses rather than D -linked clauses. '

(i)

a. b.

3

I gloube [es git Uibe uf em Jupiter]. ' I believe there is life on Jupiter.' I gloube [dass es uf em Jupiter Uibe git] . ' I believe that there is life on Jupiter. '

Wh-complements selected by verbs such as frooge 'ask' and ramIe 'guess' are compatible with V2. These verbs select interrogative complements. On the other hand, wh-complements of verbs such as e.g. dejiniere 'define' or usrechne ' calculate' can also show V2. There is no consensus among native speakers when V2 sounds acceptable in a wh-complement and out of context these are generally j udged to be ungrammatical. The only context in which Swiss Gennans generally allow V2 is in wh-complements selected by weisch 'knOW2sg' (see also Van Riemsdijk 2000).

Doubly-filled comps in Swiss German 575

by the adults into three categories: (i) monosyllabic wh-phrases (e.g. war 'who' , wo 'where', wie 'how'), (ii) bisyllabic wh-phrases (e.g. worum ' why', wivil 'how many', wie wiit 'how wide'), and (iii) polysyllabic wh­ phrases (e.g. weli Frou 'which woman', wa fore Gschicht 'what kind of story'). The data are summarized in Table 1 .

Table 1 : OFes in finite verb-final wh-complements in adult Lucemese wh-phrase - DFC + DFC

monosy llabic 1 49 1

bisyllabic 10 15

polysyllabic 4 16

Since only one of a total of 1 5 0 wh-complements with a monosyllabic wh­ phrase occurs with a DFC its use with 'very short' wh-phrases seems to be ruled out in Lucemese. Bisyllabic wh-phrases sometimes occur with DFCs and polysyllabic wh-phrases often occur with DFCs. If heaviness of wh­ phrases can indeed be translated as the number of syllables, then the contrast between the following pairs of examples can be explained: both wh-phrases in (5) have two syl lables, and both wh-phrases in (6) have three syllables. Thus wh-phrases with two or three syllables are neither heavy nor light and are therefore optionally compatible with a DFC. The question then is how many syllables make a wh-phrase heavy. Even in (7) no DFC is used, although the wh-phrase consists of five syllables. Note that example (3b) illustrating heaviness in Bemese consists of only four syllables. (5)

a.

Weisch du [worum s Rossli warm isch]? know you why the little horse warm is ' Do you know why the little horse is warm?' b. Etz chum-i drus [wiso dass' zwee Biige ufem Bild now come-I through why that-it two stacks on-the picture hat] . has 'Now I understand why there are two stacks in the picture.'

gem hasch] . (6) a. Ich weiss ebe nod [weles Tier du I know EBE not which animal you dear have ' I don't know which animal you like.' gwiisst [wo dore dass-r miiend fahre]? b. Wie hand-r how have-youpl known where in that-youpl must drive]? 'How did you know where you had to drive in?'

576 Manuela Schonenberger

(7)

Si wett no lose [was fares Gschichtli mer hut she wants still listen what for-a fairy-tale we today am morge verzellt hand] . at-the morning told have ' She still wants to hear what kind of fairy-tale we told this morning.'

The fact that some non- light wh-phrases co-occur with dass does not seem to be correlated with the linear position the finite verb occupies: All of the 1 95 wh-complements are unambiguously non-V2 independent of the presence or absence of dass. Thus Cooper's proposal for Zurich German does not carry over to Lucernese. Let us now address the question of whether the occurrence of dass could depend on D-linking. In other words, does dass only occur in D­ linked phrases? Although this question is difficult to tackle, it appears that this hypothesis is not confirmed by the data. For instance, worum 'why', which is usually interpreted as non-D-linked, sometimes co-occurs with dass (Sa) and sometimes does not, as in (Sa) above. On the other hand, weles Tier 'which animal' could be taken to be D-linked, but once it co­ occurs with dass (8b) and once it does not, as in (6a) above. Moreover, the examples (6a) and (8b) are both produced by the mother and uttered in the same situation. (8) a. Verzell de Manuela [worum dass 't's ufschribsch] . Manuela why that-you-it down-write tell thedat 'Tell Manuela why you are writing it down. ' b. Chasch j o fliischtere [weles Tier da�s d u gem hasch] . Can2sg JO whisper which animal that you dear have 'You can whisper which animal you like' Since neither the heaviness of wh-phrases nor the notioI.1 of D-Iinking seems to be able to account for the data, I try to pursue a different approach. It appears that the presence or absence of dass partially depends on whether the constituent following the wh-phrase i s a clitic. In the case of monosyllabic wh-phrases a clitic forms a prosodic word with it, which is supported by the fact that an intrusive n is used between the wh-phrase ending in a vowel and the clitic starting with a vowel, as shown in (9a) and (9b). The non-cl itic ire 'her' does not form a prosodic word with the wh­ phrase wie 'how' in (9c). An intrusive n is not permitted (9c').

Doubly-jilled camps in Swiss German 577

lose [wie-n-er redet]. (9) a. T'Manuela wett the-Manuela wants listen how-youpl speak ' Manuela wants to listen to how you speak. ' b. Die hat gwUsst [wo-n-er isch] . this one has known where-he is 'This one knew where he is.' [wie ires MeitH heisst] . c. Und etz weiss ich nUm and now know I no longer how her girl is-called ' And now I no longer know what her daughter is called.' c' . *Und etz weiss ich nUm [wie-n-ires MeitH heisst] . In the case of wh-phrases consisting of more than one syllable, a clitic seems to be unable to form a prosodic word with the wh-phrase: doss is inserted to accommodate the subject clitic in the examples in ( 1 0) and to form a prosodic word with it: ( 1 0) a.

Hat zersch mtiese lehre [wie starch dass-r isch] . has first must leam how strong that-he is ' He first had to leam how strong he is. ' b. Weisch [was for Ziit dass' isch]? know2sg what for time that-it is 'Do you know what time it is?'

The use of a DFC in (8b), but not in (6a), is not covered by this hypothesis. At first, example ( 1 1 ) seems to be problematic as well. However, in this case the insertion of dass breaks the sequence of four consonants (final 'ts' in ' Satz' and initial 'tm' in 't'Moira'), which would be difficult to pronounce. Note that the cHtic article t forms a prosodic word with the noun Moira and not with the preceding dass. On the other hand, one could assume that the insertion of doss is optional in these contexts. (1 1)

dass t' Moira Die tuet ufschriibe [was fOr Satz this one does down-write what for sentences that the-Moira macht] . makes ' She writes down what kind of sentences Moira produces. '

Without going into further detail I assume that prosody is the driving force behind DFCs. I now tum to the acquisition of DFCs in Lucemese.

578 Manuela Schonenberger 2. 2.

Acquisition data on DFCs

Penner's ( 1 996) study is based on diary data and elicitation data of a child acquiring B ernese, covering the emergence of the first embedded clauses at age 1 ; 1 1 up to the acquisition of DFCs and adverbial correlates around age 5 ;0 . Penner notes that child J. is unable to reproduce DFCs in an experimental task at age 4;0, but that ten months later he is able to do so. Shortly afterwards 1. starts to produces DFCs in spontaneous speech. In my acquisition data of two children, Moira and Eliza, acquiring Lucemese it is more difficult to determine when exactly the children start to produce DFCs productively. One of the problems is that both children move the finite verb in all embedded contexts before age 4; 1 1 (see Schonenberger 200 1 ). Verb movement in wh-complements is applied irrespective of the matrix verb. It looks as if the children had set the parameter for Belfast English. Before age 5;0 Moira produces three examples with a OFC, shown in ( 1 2), Eliza none. Two of Moira's examples, ( I 2a) and ( 1 2b), are incompatible with DFCs, regardless of the non-target-consistent verb placement. ( 1 2) a. %Da ghort's nod [wenn dess chunt opper eifach this one hears-it not when that comes somebody simply is Schloss]. (Moira: 4;06) to the castle 'This one does not hear when somebody simply walks up to the castle.' dass gseht uus wie a de Rahel b . %es Ross [wo eri Hoor] (Moira: 4;07) a horse which that looks out like to the Rabel her hair 'a horse which looks like Rabel's hair' % sage [weles Tier dass nimmsch du au c. Chasch au Can2sg also say which animal that take you also no] . (Moira: 4; 1 0) still 'You can tell me which animal you are also going to take.' Since OFCs are assumed to appear around age 5;0 I started to elicit OFCs from the children by asking them to repeat a target sentence which contained a OFe. The elicitation experiment was first used when the children were 4; 1 1 . Several sessions followed over the coming months and the children became more and more successful at repeating DFCs. One of the errors which frequently occurred in the early sessions was that both

Doubly-filled comps

in Swiss German 57 9

chi ldren left out dass and moved the finite verb instead, giving rise to V2. In a brief experiment carried out at Moira's kindergarten the elicitation of DFCs also provoked verb-placement errors in some of the children. After age 5 ;0 Moira gradually switches to the verb-final pattern and OFes start to appear in her spontaneous speech. In Table 2 all her wh­ complements which are not V2 are considered. The data collection spans from age 3 ; 1 0-8;0 I . The distribution of Moira's DFCs closely resembles that of the adult grammar. Eliza's corpus is much smaller than Moira's: There are 58 wh-complements in which the verb placement is not V2. In most of these the wh-phrase is monosyllabic (55 examples). The remaining three contain bisyllabic wh-phrases. In two of these Eliza uses a DFC, and the constituent following dass is a subject clitic. Table 2: OFCs in Moira's finite non-V2 wh-complements

wh-phrase - DFC + DFC

monosyllabic 483 2

bisyllabic 8 22

polysyllabic 8 28

Moira very occasionally produces embedded sentences which are unusual because they contain a superfluous s, shown in ( 1 3). In 6 of a total of 9 examples this s occurs in relative clauses introduced by woo It is not at all clear what this s stands for. Three options come to mind: (i) s is a resumptive pronoun, (ii) s expresses complementizer agreement. (iii) s is the remnant of a partially silent days. (Moira: 4;07) ( 1 3 ) a. % mit dem [wo s hat scho opper] with that which S has already somebody (Moira: 5 ;07) b. % [ob s du weltst au Chind] if s you want also children (Moira: 6,'02) c. %Jede tor! uslese [wie lang s er torn. everybody can choose how long S he may [wenn s ne oppis weh d . %da sageI vii Chind this say many children when S lhemdal something- harm (Moira: 7;05) tuet] does If s were a resumptive pronoun then its occurrence would be particularly mysterious in the examples which are not relative clauses. Moreover, in Lucernese resumptive pronouns usually occur in oblique relative clauses, but not in non-oblique relative clauses-an option which would however be

5 80 Manuela Schonenberger

available in Bemese object relative clauses. Complementizer agreement would straightforwardly explain ( l 3 b) in which Moira tries to imitate a native speaker of a German dialect, but would be difficult to extend to the other examples. The third hypothesis seems to be sJightly more promising. Although relative clauses headed by wo are incompatible with dass in the target grammar and Moira produces a DFC with the homophonous wh­ phrase wo only once, there is at least one example with an actual DFC in a relative clause in her data (cf. [ 1 2b D. Moreover, this line of reasoning could also be applied to ( 1 3d) containing the complementizer wenn 'wenn, if; she produces one example in such a context (cf. [ 1 2a]). Example ( l 3c) would in fact be compatible with a DFe, which would further support this hypothesis. 3.

Conclusion

I have tried to show that the occurrence of doubly-filled COMPs in Swiss German wh-complements is dependent on prosody. Not only does the length of the wh-phrase play a role but so does whether or not the constituent following it is stressed. In my corpus data on adult Lucemese, DFCs are not attested with monosyllabic wh-phrases; instead they seem to be used when the wh-phrase consists of more than one syllable and the constituent following it is a subject (or object) clitic. Children are understood to acquire DFCs around the age of 5;0 and in this respect my Lucemese children's use of DFCs is target-consistent and does not seem to be delayed. This is in sharp contrast to their acquisition of verb placement, which they are apparently slow to acquire. References Bayer, Josef. 1 984. Towards an explanation of certain that-t phenomena: The COMP-node in Bavarian. In Sentential complementation, ed. by Wim de Geest and Yvan Putseys, 23-32. Dordrecht: Foris. Cooper, Kathrin. 1 995 . Nun subjects and clitics in Zurich_ German. [n Topics in Swiss German Syntax, ed. by Zvi Penner, 59-72. Bern: Lang. Haegeman, Liliane. 1 992. Theory and description in generative syntax. A case study in West Flemish. Cambridge: Cambridge University Press. Henry, Alison. 1 995. Belfast English and Standard English. Dialect variation and parameter setting. New York: Oxford University Press. Koster, Jan. 1 986. Domains and dynasties. Dordrecht: Foris. Lightfoot, David W. 1979. Principles of diachronic syntax. Cambridge: Cambridge University Press.

Doubly-filled comps in Swiss German 581 Penner, Zvi. 1 99 6 . From empty to doubly-filled complementizers. A case study in the acquisition of subordination in Bernese Swiss German. Fachgruppe Sprachwissenschaft der Universitat Konstanz. Arbeitspapier Nr. 77. Penner, Zvi and Thomas Bader. 1 995. Issues in the syntax of subordination: A comparative study of the complementizer system in Germanic, Romance and Semitic languages with special reference to Bemese Swiss German. In Topics in Swiss German Syntax, ed. by Zvi Penner, 73-290. Bern: Lang. Pesetsky, David. 1 987. Wh-in-situ: movement and unselective binding. In The representations of {in)definiteness, ed. by Eric Reuland and Alice ter Meulen, 98-129. Cambridge, Mass.: MIT Press. Riemsdijk, Henk C. van. 2000. WH-prefIxes: The case of w iiisch i n Swiss German . In Naturally! Linguistic studies in honour of Wolfgang Ulrich Dressler presented on the occasion of his 6(jh birthday, ed. by Chris Schaner-Wolles, John Rennison, and Friedrich Neuharth, 423-43 1 . Torino: Rosenberg and Sellier. Schonenberger, Manuela. 200 1 Embedded V-to-C in child grammar. The acquisition o/verb placement in Swiss German. Dordrecht: Kluwer. Taraldsen, Tarald. 1 986. On Verb Second and the functional content of syntactic categories. In Verb Second phenomena in Germanic languages, ed. by Hubert Haider and Martin Prinzhom, 7-25. Dordrecht: Foris. .

Missing prepositions in Dutch free relatives Chris Sijtsma

1.

Introduction

In this squib, I will present some Dutch data concerning free relative c lauses with verbs selecting prepositional objects. I wiJI demonstrate that it is the matrix preposition that goes missing. Although none of the analyses in the literature can account for this fact, I will not offer an explanation. Instead, I will leave this as a challenge for retired linguistic professors. 2. The Analyses for Free Relatives

Bresnan and Grimshaw ( 1 978) support the Head Hypothesis, which specifies that the relative pronoun, together with the pied piped material, ends up in the projection heading the relative clause. Bresnan and Grimshaw say that the free relative is not a full fledged clause. Because the specifier of CP is missing, the relative pronoun has no other choice but to move to the projection heading the relative clause (Bresnan and Grimshaw's example 1 07). (1)

I will move

[pp to whatever town]i [IP you move [pp tj]] .

Because the moved projection moves into the matrix clause, matching phenomena are accounted for automatically. According to Bresnan and Grimshaw, non-matching languages have CPs as free relative clauses instead of IPs. Groos and Van Riemsdijk ( 1 979) support the COMP Accessibility Hypothesis instead of the Head Hypothesis. Under the COMP Accessibility Hypothesis, in matching languages, the specifier of CP of the free relative clause is accessible to the matrix verb for fulfilling its selectional restrictions. Groos and Van Riemsdijk think that free relative clauses are always full fledged CPs (their example 6d).

Missing prepositions in Dutchfree relative 583

(2) a. Head Hypothesis Zij danst [pp met wie]i [cp [Spec e] j ij gisteren danste [pp til], She dances with whom you yesterday danced. 'She dances with the one with whom you were dancing yesterday. ' b. COMP Accessibility Hypothesis Zij danst [pp e] [cp [Spec pp met wie]i j ij gisteren danste [pp ta] . =

Extraction behaviour in Dutch and Gennan led them to bel ieve that structure (2b) is to be preferred (Groos and Van Riemsdijk's example 25). (3) a. Ik heb [ NP de vis [cp die over was]] opgegeten. I have the fish that left was PRT.-eaten. 'I have eaten the fish that was left. ' b. Ik heb [ NP de vis [cp til] opgegeten [cp die over wask c. *Ik heb [NP til opgegeten [ NP de vis [ cp die over Was]]i.

(4)

a. Ik heb [NP e [cp wat over was]] opgegeten. I have what left was PRT.-eaten. 'I have eaten whatever was left.' b. Ik heb [NP e [cp till opgegeten [cp wat over was]i.

The core grammar has a matching parameter. If this parameter is switched off, matching is not enforced. Because example (2) is ungrammatical in Gennan, although German is a matching language, Groos and Van Riemsdijk propose a filter barring anything from the specifier of CP in free relative constructions, except a wh-word. Harbert ( 1 983) wonders why the specifier of CP is accessible. He argues that the filter that Groos and Van Riemsdijk propose is redundant in many cases. Also, he thinks that, even in non-matching languages, the specifier of CP is accessible from outside. Harbert suggests that the head of the free relative is PRO in a matching position. (5)

V [NP PRO] [cp [SpecWH] . ]] .

.

To escape ungrammaticality, government accesses the specifier of CP, consequently leaving PRO ungoverned. In a non-matching position or a non-matching language, the head of the free relative is pro. This restricts non-matching free relatives to languages and positions that have pro detennination. Also, the empty head must be a noun phrase. Harbert proposes a deletion rule on S-Structure to account for missing prepositions:

584 Chris Sijtsma

P2 can be deleted in the context PI P2 where P I and P2 are identical and L­ adjacent. Not only the phonological content disappears, but the entire P-projection is deleted (Harbert's example 41). S Ofie con [NP PRO [cp [Spec PP sea quien]i til whom you I-dreamed about sofiaste [pp ti]]. dreamed. I dreamed about the person whom you dreamed about. ' b. Sofie con [NP PRO [cp [Sp ec NP quien]i to sofiaste [pp til]·

(6) a.

=

=

The matrix preposition has to be present at D-Structure, because otherwise the selectional restrictions of the matrix verb will not be met. At S­ Structure, the selectional restrictions of the matrix preposition can be met by the NP in the specifier of CP. As far as I am concerned, there are two problems with this proposal . Firstly, the PP trace in the relative clause is left coindexed with an NP at S­ Structure and LF. Secondly, the projection principle forbids the deletion of projections. Sufier (1984) suggests that the Spec C position of the relative clause is never accessible for the matrix verb to fulfil its selectional restrictions. The head of the relative is always pro in a matching a position. Matching is just another case ofpro determination. In ungoverned positions (e.g., infinitives and topicalisation), there is no matching because there is no need for pro determ ination since the NP heading the realtive is PRO. In subject position, pro is already determined by Inft in pro-drop languages. Sufier accounts for missing prepositions by proposing a PF filter: either PI or P2 can be deleted in the context PI P2 where PI and P2 are identical and L-adjacent. Only the phonological content is deleted. Larson (1987) analyses m issing prepositions in English free relatives as a case of antecedent-contained deletion. This process is only allowed if reconstruction is possible at LF (Larson's examples 4 and 5). (7) a. John suspects [NP everyone [cp that Mary does]] b. *John suspects [NPFritz, [cp who Mary does]] The NP everyone in (7a) is quantified. Quantified NPs are raised at LF. When the NP is raised, it drags the relative clause along. Because the relative clause moves out of the containing VP, reconstruction becomes possible. Since the indicated NP in (7b) is not raised at LF, no suitable

Missing prepositions in Dutchfree relative 585

source for reconstruction will be available. Reconstruction of the deleted verb results in an endless loop; hence, the example is ill-formed.

(8) a. [ NP everyonej that Mary [vp e]] John [ vp suspects [ NP ed] b. [NP everyonej that Mary [vp suspects [NP eil]] John [ vp suspects [NP eil]

(9)

a. John [v p suspects [ NPFritz, who Mary [ vp e]]] b. John [vp suspects [ NP Fritz, who Mary [vp suspects [ NP Fritz, who Mary [vp e]]]]]

Because relative pronouns in free relatives are quantified, they are raised at LF. Therefore, Larson suggests that the missing preposition has undergone antecedent-contained deletion. {lO) a. SS: I will live [ pp in [cp [ spec=- NP whatever] town you live [ ppe]]] b. [ cp [Spec NP whateverj] town you live [ pp e]] I will live [ pp in ed c. LF; £Cp [Spec'"' NP whateverj] town you live [ ppin ed] I will live [ pp in ed =

Notice that Larson 's structures seem to lack a noun phrase heading the free relative clause. Instead, the matrix PP seems to directly fulfil its selectional restrictions in the specifier of the embedded CPo Also, his analysis of missing prepositions in English free relatives hinges on the fact that the preposition in the subordinate clause is missing. 3.

Free Relatives and Missing Prepositions in Dutch

In both the English and the Spanish examples with missing prepositions, the prepositions could have been linearly adjacent before deletion. Therefore, it is not easy to discover whether it is the preposition in the matrix clause or the one in the subordinate clause that goes missing. Because PP is a bounding node in Dutch, pied piping is compulsory. However, there exists one class of pronouns that can escape from a PP via [Spec, Pl. These pronouns are called R-pronouns (Van Riemsdijk, 1 978).

(11) a. [ pp [Spec ] van het]

b. [ pp [ spec] van wat] [ pp [Spec ] van dat]

C.

(of it � thereof) � [pp [Spec eri] van til � [pp [ Spec waari] van til (of what � whereof) � [pp [Spec daari] van til (of that � thereof)

586 Chris Sijtsma R-pronouns only exist for [ -human] pronouns. In an appropriate environment, the R-pronoun can move further, leaving the preposition stranded . For [+human] pronouns, pied piping is the only option. [pp Op wie]i wacht jij [ pp til? on whom wait you? 'Who are you waiting for?' b. [ NP WaarJi wacht jij [pp [ Spec til What wait you 'What are you waiting for?' c. [pp Waarop]i wacht jij [pp ta?

( 1 2) a.

This phenomenon gives us a tool for finding out which preposition is deleted in Dutch free relatives. Ik houd (pp van [ NP e] [cP [Spec NP waari] J IJ I like of what you [p p [spec til van [ NP ti]] houdt]]. of like 'I like whatever you like' h. Ik houd van waar jij ¥aft houdt (delete the embedded preposition) (delete the matrix preposition) c. Ik houd ¥aft waar jij van houdt

( ] 3) a.

=

We have to tread carefully, however. In their footnote 25, Groos and Van Riemsdijk ( 1 979) warn us that, in Dutch, different factors having to do with production and perception influence the relative acceptability of certain free relatives with PPs in [Spec, C] . The length of the sentence matters, as does the identity of the matrix and the embedded verb. I conducted a survey under my colleagues, after having made a prediction of the results. Be�ause I myself think it is the embedded preposition that is left behind, eyen when the preposition�s stranded, in my prediction, I distinguish between" different versions'of the filter. The three versions I account for are 'only one preposition is allowed' ( l prep), 'two prepositions are allowed as long as they are not linearly adjacent' (2 prep, non-adj) , and 'two prepositions are allowed to co-occur at all times' (2 prep, adj). Of course, this last one is not a different version of the filter, it is the option of not using a filter at all. I make a further subdivision for R­ raising: 'R-raising is optional' (R-opt) or ' R-raising is preferred' (R-pref), because I observed that for some speakers the R-raising demonstrated in (1 1 ) is truly optional, whereas for others it is the preferred option.

Missing prepositions in Dutch free relative 587

(14)

2 prel, adi 1 Jfep 2 prep. non-adi R-opt R-pref R-opt R-pref R-opt R-pref * * * * ok ok ok ok ok ok ok ok

PREDICTIONS

I like whomever you like

:',

Ik houd van van wie jU houdt jij houdt

Ik houd van wie

I like whatever you like

�-..

* ok * ok

Ik houd van van wat jij houdt Ik houd van wat jti houdt Ik houd van waarvan jij houdt Ik houd waarvan jij houdt

* * * ok

* * ok ok

* ok ok ok

* * ok ok

ok ok ok ok

Strictly speaking, the predictions for the sentence Ik houd WiR waarvan jij do not have to be correct. Because R-raising took place, and because one of the prepositions was left out, there is another variant for this sentence, namely Ik houd van waarwm jij houdt. People m ight differ on which preposition to leave out. Because the preposition 'has undergone pied piping and is therefore close to the matrix preposition, and because I myself think Ik houd van waar jij houdt is very ungrammatical, I will ignore this variant for the moment. In my survey, I included some sentences that made it possible for me to assign a respondent to one of the six groups. The results are shown in 1 table ( 1 5) below. With some idealisation of the data (***, **, *, ? * and ??? are mapped onto *, and ??, ?, # and ok are mapped onto ok), my predication almost came true. There is only one deviant judgement. The sentence Ik houd van wat jij houdt is judged better than expected by the group that prefers R-raising. On the other hand, R-raising is only a preferred option, not a compulsory one. Therefore, one expects the judgements to be less sharply divided. houdt

Per group, 1 took the mean judgement in the following way. I assigned a value to every judgement: ok 1 0, # 9, ? 7, ?? 6, ??? 5, ?* 4, * 3, ** 2 and *** 1 . I then took the mean. I converted the mean back to a judgement in the following way: [ 1 - 1 5 > ***, [1 .5-2.5 > **, [2. 5-3 .5> *, [3.5-4.5> ?*, [4.55 5> n? [5.5-6.5> n, [6.5-7.5> = ?, [7 .5-8.5> # and [8.5-1 0 ]:::: ok. The group 'I prep, R-opt' consisted of 2 persons, the group '2 prep, non­ adj, R-opt' of 6, and the groups '2 prep, adj, R-opt' and '2 prep adj , R-pref of 5 each. I left out one set ofjudgements altogether.

,I

=

=

=

=

=

=

=

=

=

.

.

=

=

=

=

=

=

=

,

588 Chris Sijtsma

( 1 5) 1 prep

RESULTS

I like whomever you like

Ik houd van van wie jij houdt Ik houd van wie jij houdt I like whatever you like

lk houd van van wat.Hi houdt Ik houd van wat jij houdt Ik houd van waarvan jij houdt Ik houd waarvan jij houdt

R-opt ?* ok

?* ok ** ok

R-pref

2 prep, non adj -

R-opt *

R-pref

ok

* # # ok

2 prep, adj R-o pt ok ok

R-pref ok ok

# # ok ok

???

t�;;l ok ok

What is a matter of concern is that the groups ' 1 prep, R-pref and '2prep, non-adj, R-pref do not have any members. I do not know whether this is a coincidence and that, if a sufficiently large number of respondents had been used, these groups would have had members, or that there is an underlying reason for the lack of members in these groups. It might even be the case that there is no such thing as a distinction between optional R­ raising and preferred R-raising. What struck me as peculiar is that 16 out of 18 respondents allow two prepositions to some degree. I am one of the two people who allow only one preposition. Maybe the problem is just that I see myself as the standard instead of as the exception. Let us now turn to the topic of this squib: missing prepositions in Dutch free relatives. The judgements of my respondents are shown in table (16). (16) RESULTS, MISSING PREP.

Pied piping Ik houd van waarvan jij houdt Ik houd van waafYtm j ij houdt Ik houd ¥aR waarvan jij houdt Preposition stranding Ik houd van waar jij van houdt Ik houd van waar jij ¥aR houdt Ik h oud ¥aft waar jij van houd t

1 )rep 2 prel , non-adi R-opt R-pref R-opt R-pref ** # * **

2 prep, adi R-opt R-pref ok ok * *

ok

ok

ok

ok

·.1? .

**

# *

ok *

ok *

ok

ok

ok

'!

The one thing that immediately catches the eye is that deleting the embedded preposition is completely ungrammatical. All respondents agree that the matrix preposition may be deleted, but that the embedded

Missing prepositions in Dutch free relative 589

preposition has to stay. If the matrix preposition is deleted, it does not real ly matter whether the embedde d preposition is pied piped or whether it is left stranded. The only remarkable fact left is that the one-preposition group does not completely reject two prepositions when the embedded preposition is left stranded. The explanation is quite simple. The one-preposition group consists of only two people, me and someone else. The other guy just had it wrong when judging this sentence. No seriously, the one-preposition group only has two members who totally disagree on the sentence lk houd van waar jij van houdt. I think the number of respondents in this group is just too small to draw any further conclusions. For dessert, let us look how free relatives with prepositional objects behave in a potentially ungoverned position. Let us see what happens if we topicalise the free relative clause. Let us say that the judgements do not change after topicalisation, i.e., we use the actual judgements of the oon­ topicalised sentences as a prediction for the topicalised ones.

ok

ok

ok

ok

# ok

ok

ok

ok ok ok ok

ok ok ok ?

ok **

??

#

Now let us tum to the actual judgements as presented below.

590 Chris Sytsma

I really cannot explain the judgement *Waarvanjij houdt, houd ik in the '2 prep, adj, R-pref group. If we disregard this oddball judgement and if we also leave the '2 prep, adj, R-opt' group aside, a disturbing pattern emerges. Could it be that the topicalisation structure is derived from left dislocation under deletion of (part of) the resumptive prepositional object?

( 1 9) a. b. c. d.

*Waarvan j ij houdt, EIaaf houd ik van Waarvan jij houdt, Eiaar\'aH houd ik Waar jij van houdt, EIaaf houd ik van *Waar jij van houdt, deaNaR houd ik

We m ight hypothesise that we can delete the part of the resumptive prepositional object in [Spec, C] in the matrix clause if it is 'equal' to the content of [Spec, C] of the relative clause (waarvan B daarvan, waar B daar), yielding the topicalisation structure. This idea does not tally with the fact that, in Dutch left dislocation structures, the preposition of a prepositional object can be doubled, whereas this doubling is completely impossible in topicalisation constructions. (20) a. Van bananen daar houd ik van From bananas there like I from 'Bananas, I 1 ike them' b. *Van bananen houd ik van From bananas like I from 'Bananas I like' The best I can offer is that in Waarjij van houdl, houd ik, the stranded preposition has a feel of being too far out of scope for the matrix verb to be construed as its prepositional object. This leaves, the deviating results for Waarvan jij houdt, houd ik van and Van wie jij hoildt,� houd ik van t9tally unaccounted for. , '; " -

, ''

References Bresnan, Joan, and Jane Grimshaw. 1 978. The Syntax of Free Relatives in English,

Linguistic Inquiry 9, 3 3 1 -3 9t.

Groos, Anneke, and Henk van Riemsdijk. 1 97 9. Matching Effects in Free Relatives: A Parameter of Core Grammar. In A . Belletti, L. Brandi, and L. Rizzi, eds., Theory of Markedness in Generative Grammar, Scualo Nonnale Superiore, Pisa.

Missing prepositions in Dutchfree relative 591 Harbert, Wayne 198 3. On the Nature of the Matching Parameter, The Linguistic Review 2, 237-284. Larson, Richard. 1987. 'Missing Prepositions' and the Analysis of English Free Relative Clauses, Linguistic Inquiry 18, 239-266. Riemsdijk, Henk van. 1978 . A Case Study in Syntactic Markedness: The Binding Nature of Prepositional Phrases, Foris Publications, Dordrecht. Suffer, Margarita. 1 984. Free Relatives and the Matching Parameter, The Linguistic Review 3 363-387. ,

Final sonorant devoicing in early Yokuts field-records N orval Smith

1.

Introduction l

Th e topic of this squib is th e devoicing, or phonetically more correctly, whisp ering of final sonorant devoicing consonants and vowels in various Yokuts dial ects and languages (henceforth, lects) spoken in the southern C entral Valley of California. Up till about 1 849 there were about 50 such Yokuts lects spoken here, despite the large-scale mortality engendered by Spanish m issionization, and an epidemic, probably of malaria (Cook 1 955: 59), earli er on in th e 19 th century, which had together reduced th e population by at l east a half. The Gold Rush, and the associated influx of Euro-American and other immigrants wiped many Yokuts tribelets from th e fac e of the earth in a period of 20 years. In the early 1 900's only about 1 % of the original population survived, at l east half of th e loss being directly attributable to the period between 1 850 and 1 870. As a result of this there are large gaps in our knowledge of Yokuts lectal geography, particularly in the northern San Joaquin Valley. For this reason it is important to be abl e to extract as much infonnation from the m eagre records as possible. Sonorant d evoicing is represented variously in early fi eld-recordings. My conclusion is that some devoicing. is omni-present in Yokuts, but that various sonorants display different patterns. This helps us in the interpretation of the phonological/phonetic structure of dial ects for which we lack accurately transcribed recordings. Stanley Newman, the grand old man of Yokuts linguistic description , refers to this fact most succinctly (Newman, 1 944: 1 6, 26): The dialects also agree in partially whispering I, w, and y in final position: Wikchamni yakaw, "stone," might be written phonetically as yakaw hw. In Chawchila final m and n are similarly whispered . The glottalized consonants, w', y', I', m', n ' , and I)', I r thank William F. Weigel for allowing me to use a copy of his Yokuts word-list extracted from Newman's work.

Final sonorant devoicing in early Yokutsfield-records 593

when they occur finally in a word or in a closed syllable, are heard h , hw or even as _, w, etc. as w _

In Chawchila a word-final vowel may become unvoiced, and, when this takes place, the vowel is assimilated to the preceding vowel. This unvoicing process, however, does not occur regularly and invariably; unvoicing is more common in slow speech than in rapid speech, and final vowels are more apt to be unvoiced in words preceding a pause than in words within the body of a phrase. In the most northerly form of Yokuts, Delta Yokuts, there is also a process by which 1m, nl are changed to Ib, dI in some tects. I will not discuss this here. The strategies used to represent whispered sonorants are three in number: ( 1 ) a. The sonorant is recorded without any indication of whispering. b. The sonorant is recorded with an indication of whispering. c. The sonorant is omitted. In the case of vowels, the last strategy is utilized, but more common is repetition of the second-last vowel in final position. 2. The final whispered sonorants

Just to emphasize the fact that the omISSIon of a consonant does not necessarily mean that it was absent, I refer to the fact that the unfamiliar velar fricative Ix! was often omitted, indicated by h, or replaced by the more familiar k. These two strategies, omission and substitution, have also thus been adopted in the case of the whispered sonorants. I will concentrate on the cases of omission, and those cases where I interpret the representation as an attempt, however unsatisfactory, to represent the whispered sounds. Obviously, the cases where the final sonorants are recorded without further ado do not tell us anything about their phonetic realization. The data, unless otherwise indicated, come from Kroeber ( 1 959). The Noptrintre data is taken from Beeler ( 1 97 1). In this table, and those following, the column headed "example" exemplifies the imperfect transcription of a final sonorant, while the column headed "compare" exemplifies the correct transcription.

594 Norval Smith

Table 1: Omission 11#1 Saratoga Chalostaca Yachikumne B Yachikumne B Yachikumne B /y#/ Saratoga /W#/ Knight's Ferry 0 /n#/ Yachikumne B Yachikumne B Knight's Ferry D Yachikumne B

compare

example

trans lation

chadashiwi thebiCi poot03 ta/axe silki

five nose fire deer groundsquirrel

podo

two

Yawelmani bonoy

hile

where?

Yachikumne A hileu

ottie suyo upe 10-;-0

chief sun moon night

Coconoon oleen Yachikumne A suy6n Yachikumne B hopem Yachikumne A toyin

Example

translation

compare

itzanik holox itexl -silet tsadax4

five fire ash rock ice

Yawelmani yic 'inil Knight's Ferry F hOlo/ Yachikumne A hitel Yachikumne A silel Yawelmani tropol

u/us

black bear

Yachikumne A ului

Yachikamne A tranacwil2 General Yokuts trimetril Yachikumne A hOlol Yachikumne A ta/axil Yachikumne A sitkil

Table 2: Indication of whisper 11#/ Takin Knight's Ferry 0 Yachikumne B Yachikumne B Chalostaca /y#/ Noptrintre

2

I indicate the retroflex consonants of Yokuts as [tr, dr, tr']. c in this fonn represents [�], othetwise [ts]. 3 P for h. 4 dror p.

Final sonorant devoicing in early Yokuts field-records 595

Iw#1 Chalostaca Takin Noptrintre

ceux ottrup lomtoug

Yawelmani cri'iw rabbit Yawelmani 'otr'ow head to the mountain 10m/ow (-ug = Gen. Yok. locative I-wI)

5

Im#1 Noptrintre In#1 Noptrintre Noptrintre

nim, nimg min, ming amin,

7 aming

6

mine

Yawelmani nim

yours his, hers

Yawelmani Yawelmani

min 'amin

Some commentary is apposite here. The representation of a whispered III as x or xl is the most successful in phonetic terms. t is a poor rendition, while k could be an attempt at a perceived [x] . The recorder of the Takin vocabulary was a very poor transcriber. s as a rendition of whispered /y/ is reasonable in terms of the Spanish orthography of the Noptrintre original. The result of whispering the Iyl must have resembled the German ich-Laut. The other three whispered sonorants are transcribed with the help of (equivalents of) [x].

Table 3: Confusion of whispered sounds Im$1 Knight's Ferry 0 Chalostaca /n$/ Chalostaca /0#/ Knight's Ferry D

example

translation

compare

hatilka haatilke

sing! sing!

Yachikumne A hdtimka Yachikumne A hatimka

ukkulku

drink!

Yachikumne A ukunka

ukul

drink

Yawelmani

'ugun

Here we have a number of cases where the whispered sounds, which resemble each other far more than their non-whispered counterparts, display confusion. S

And numerous other examples. Read by Beeler as miug 7 Read by Beeler as amiug

6

.

596 Norval Smith In his edition of Arroyo's Noptrintre grammar from 1 837, and texts from 1819, written down at Mission San Juan Baptista in the then Spanish/Mexican California (Beeler 1 97 1 ), Beeler notes that this Spanish mission linguist usually wrote ug [wx] for final Iw/. He also notes the optional forms cuyo - cuyon 'ankle' and pucho - puchon 'son, cousin', wh ich he attributes to the voicelessness of the final /nJ as in Chawchila. He does not comment on the form nimg, nor on the forms ming, aming, which he reads miug, amiug. T would interpret these as referring directly to the whispering of the final nasals, Iml pronounced as something resembling [mx] , and Inl as something resembling [nx] finally. The final [x] represents an attempt to represent the final (voiceless) whisper. We do know where Noptrintre was spoken natively - on the west side of the San Joaquin river opposite Chawchila. It is therefore not surprising that it seems to resemble Chawchila so much. In fact these two, and Takin, appear to form a sub-group of the Valley Yokuts lects. This has obv ious consequences for the original location of the Takin speakers, who ended up on the Stanislaus River near Knight's Ferry, at the valley edge. 3.

Whispered vowels

Yokuts phonological words may not end in more than one consonant. As we will see two orthographic final consonants is an important indicator of a final-whispered vowel when it occurs in the records. The other indicator is a repeated vowel in the frequently recorded imperative suffix l-k'a1, as well as in other short-vowel suffixes. These are both features only recorded in Chawchila among the modem descriptions of Yokuts lects. So the Chawchila imperative form 'run!' may be optionally [lihimk'a] or [ lihimk'l] where the final vowel is repeated and whispered in the second case. The Chawchila data here is taken from Newman ( 1 944). Table 4: Forms apparently ending in two consonants Merced Falls Takin Takin

example ukunk lehimk hattimk

translation drink! run! sing!

compare [cf. Chawchila trunk'ut [cf. Chawchila lihimk'l] [cf. Chawchila lihimk'l]

8 'Close the door!'. Newman does not give the imperative fonn of 'drink'.

Final sonorant devoicing in early Yokuts field-records 597

Table 5: Explicitly repeated suffix vowel Merced Falls Chalostaca Chalostaca Chalostaca Chalostaca Chalostaca Noptrintre

example luiku luiko9 ukkulku haatilke kudeikelO huiosko huohuicolca

translation eat! eat! drink! sing! sleep! sit! sleep!

The question now arises just how far north these repeated whispered vowels occur. In the northernmost lects they turn out to be absent, as can be seen from Table 6.

Table 6: No repetition Yachikumne A

example ukunka

translation drink!

Yachikumne A Lathrop Lathrop Knight's Ferry F Knight's Ferry 0 Knight's Ferry 0 Knight's Ferry F

hatimka luika ukudka luika hatilka woika huyoska

sing! eat! drink! eat! dance! sleep! sit!

So northern Delta Yokuts lects do not display the phenomenon of final whispered/repeated vowels, as can be seen from (9). The southern Delta Yokuts lects (e.g. Chalostaca), and the northernmost Valley Yokuts lects (Chawchila, Takin, Merced Falls, Noptrintre) do, as we can see in Table 4 and 5 . Because the phenomenon of whispered/repeated vowels is not a widespread feature in Yokuts, it turns out to be a valuable locator for a number of dialects whose precise original whereabouts are not known, either because they were recorded at various of the missions to which Yokuts Indians were taken, like Chalostaca at Santa Cruz, or because they 9 10

The quality of whispered vowels is not so clear as those of non-whispered ones. Here the orthographic i is to be understood as a coda Iy/. leI is the vowel here.

598 Norval Smith

were recorded at various post-mission or post-Gold Rush refuge settlements, like Takin, as mentioned above. The northern floor of the San Joaquin Valley was virtually emptied of the original Yokuts inhabitants. The few that survived mostly moved to the valley margins for self­ protection. Beeler notes the correspondence between Chawchila and Noptrintre in respect of this feature too, pointing out another example of vowel repetition. 4.

Conclusions

Final whispered sonorants appear to be a feature of all northern Yokuts lects. This appears to include the nasals, which are not known to be whispered in lects south of Chawchila. Final whispered vowels appear to be a feature of a more restricted group of lects, basically a band covering about half the length of the valley from Chawchila northwards. I f we take account of the problems the recorders had to face in transcribing these sounds, some of the apparent irrationality of their transcriptions disappears. They turn out not to to have done such a bad job after all, in the sense that their attempts provide us with valuable clues as to what they were trying to represent. References Beeler, Madison S. 1 971. Noptrintre Yokuts, in J. Sawyer (ed.), Studies in American Indian languages. Berkeley:. University of California Press, 1 1-76. Cook, Sherburne F. 1 955. The aboriginal population of the San Joaquin valley, California, Anthropological Records 1 6, 2: 3 1 -78. Kroeber, Al fred L. 1 959. Northern Yokuts, Anthropological Linguistics I, 8: 1 - 1 9. Newman, Stanley. 1 944. Yokuts language o/ California. New York: Viking Fund, Inc.

Cyclic NP structure and trace interpretation Dominique Sportiche

1.

The problem1

There is an asymmetry between A' -movement and A-movement regarding reconstruction, which can be i llustrated as follows: (1) (2)

Which pictures of Bill does it seems to him [t look fuzzy] Pictures of BiIl seems to him [t to look fuzzy]

While coreference between the underscored name and the pronoun is fine in ( 1 ), it is not in (2). One contemporary type of account for ( 1 ), based on Chomsky ( 1 995) is that traces, indicated as t here, are (near2) literal copies, and that copies must be present at the point where Binding conditions are checked. This type of account would wrongly rule out (2), and thus needs to be modified. I will suggest below that traces are not always interpreted as copies. 2. Former proposals

In their 1 98 1 NP-structure paper, Henk van Riemsdijk and Edwin Williams (henceforth VR& W) suggest the following grammatical architecture, postulating an intermediate level of NP-structure between deep structure and surface structure, at which the Binding Conditions are evaluated:

I This is dedicated to Henk van Riemsdijk, as a small token of friendship and appreciation for his pioneering and stimulating work in all spheres of academic life, for his inspiring coup de fourchette, in intellectual matters and elsewhere. 2 The quantificational force of a literal copy must sometimes be tinkered with, as e.g. wh deletion in Chomsky (1995), or trace conversion in Fox (2002).

600 Dominique Sportiche

(3 )

Deep structure NP-movement NP

structure Wh-movement

I Surface structure I Given VR&W's proposal,3 the data above falls out from the architecture of the model. As the binding conditions are assumed to be checked at NP structure, that is post NP movement, but crucially before wh-movement, the input representations for the purpose of evaluating Condition C of the Binding theory are:

(4) (5)

It seems to him [which pictures of Bill look fuzzy] Pictures of Bm seems to him [t to look fuzzy]

This 'timing' idea - metaphorically speaking - derives the condition C violation in (4) and the lack thereof in (5), since in the latter but not in the former, the pronoun c-commands the name. Since it was written, new facts were discovered that cast doubt on such a treatment. First, sometimes, binding conditions are evaluated before NP movement has taken place: A-movement can reconstruct. This happens when a raised subject is understood as not scoping over the matrix clause (arguments to this effect are discussed, referred to and summarized in Sportiche, 2005). Thus if (2) is interpreted as meaning (6), a Condition C effect reemerges.

(6)

It seems to Bill that there are pictures of him which look fuzzy

Similarly, Condition A or Pronominal Binding can be sensitive to premovement structures. Thus, both of the following sentences are fine 3

Discussed in this respect in De Fourier ( 1 980).

Cyclic NP structure and trace interpretation 601

because there has been movement from a position (marked t) c-commanded by the ultimate (underscored) binder: (7)

(8)

pictures of each other seemed to the boys to be t fuzzy pictures of his parents seemed to every boy to be t fuzzy

These facts also show that it would be wrong (in addition to being theoretically undesirable) to suppose that traces of A-movement cannot be copies. Secondly, sometimes, binding conditions must be evaluated after wh­ movement as the following sentences illustrate:

(9)

( 1 0)

(11)

They know which stories about each other I like They know I like which stories about each other *They know that I like these stories about each other

The first sentence is fine, even though it is predicted to behave like the third, if pre wh-movement structures as in the second were the input for the evaluation of the binding conditions. The only other principled attempt I know of is Lebeaux's (1991,1998/ who (in agreement with Belletti and Rizzi 1988) took condition A to be an anywhere condition (it can be satisfied at any point in a derivation), and who took Condition C to be an everywhere condition (it cannot be violated at any point in a derivation). In a technical sense, the former proposal regarding condition A works straightforwardly, but the latter regarding condition C requires some addition: to prevent (5) to always be ruled out because of its (approximate) underlying structure:

(12)

seems to him [pictures of Bill to look fuzzy]

Lebeaux also adopts a timing idea according to which, while a D must be present in D-structure, its NP complement is introduced later in the derivation - a kind of late insertion proposal similar to the one he famously made to deal with the complement/adjunct distinction. Its logic is that A-movement operates on structures lacking the information relevant to check referential dependencies (which Lebeaux 4

Conceptually, Lebeaux's approach is similar to aspects of Williams' (2003) Representation Theory, as well as to Sportiche's (2005) D-splitting proposal. They all substantially differ in detail however.

602 Dominique Sportiche

hypothesizes to be NPs), hence Condition C. The infonnation is added once A-movement has taken place (but of course no later), that is, once a OP has moved into a Case position. In effect, this proposal is partially equivalent to postulating a level of NP structure, that is the level at which NPs are inserted. S As described the proposal would predict, just like VR&W's that A-movement never reconstruct for the purpose of binding, clearly a wrong result: first because A-movement does show Condition C effects; secondly because if the absence of NPs prior to NP-structure is what prevents binding Condition C violations, examples such as (7) and (8) should also be excluded, wrongly. 3.

An Alternative Proposal

I would like to propose an updated version of VR&W' s proposal to deal with this array of data. Here, I will sketch the idea, building on but modifying somewhat independently justified theoretical constructs and indicate in very rough outline some analytic consequences . Well-formed examples such as (9) show that a simple level of NP structure will not work. However, an obvious modification resolves this problem, if we conceive of NP structure not as a level but as check point within a cyclic system. Roughly speaking, imagine that there is one NP structure per clause, that is the binding conditions are cyclically checked. Then (9) is licensed on the second cycle. We specify this proposal further below. The Condition C facts in A-movement cases are telling in that they show that condition C effects in A-movement cases do not arise due to the form of underlying representations (otherwise (2) sh()uld always be out), but rather due to whether underlying representations are interpreted with the exact fonn they have. This results in two conclusions: First the representations on which B inding conditions are checked are the representations that are interpreted. This means that binding conditions are checked at Logical Fonn. Secondly, whichever traces are interpreted, they need not be interpreted as literal copies of their antecedents. Thus in (2), the trace of pictures of Bill cannot be literally interpreted as "pictures of Bill". In particular, lowest traces of movement must always be interpreted since this is how the predicate argument relations (thematic roles) are computed; see 5

The difference with the NP structure proposal is that Lebeaux's would allow the binding conditions to be checked anytime later than NP-structure.

Cyclic NP structure and trace interpretation 603

Sportiche (2005). This means that lowest traces cannot always be treated as 'literal' copies. To reconcile the fact that traces must be interpreted but not necessarily as literal copies of their antecedents, I propose that traces can in principle be interpreted as variables bound by their antecedents, i.e. more or less like bound pronouns. Thus, depending on the relative scope of the subject with respect to the verb seem, the first sentence below is treated as having one of the next two (approximate) representations:

( 1 3)

Pictures of Bill seemed to him to be fuzzy

( 14)

Subject wide scope: no Condition C effect there are pictures of Bill such that it seems to him that they are fuzzy

(15)

Subject narrow scope: Condition C effect it seems to him that there are pictures of Bill which are fuzzy

Putting all these conclusions together, we get the result that the binding conditions are checked cyclically at Logical Form, and that traces can but need not be treated like bound pronouns. To decide how exactly the binding conditions are checked cyclically, the case of wh-movement is also telling since there are cases of wh­ movement which cannot 'remove' condition C effects, regardless of the scope of the moved phrase; cf. ( 16) and ( 1 7) in the latter, the phrase pictures of John may be interpreted either inside or outside of the scope of want- hence c-commanded or not by him. -

( 1 6) (1 7)

*Which picture of John does he deserve to see t. *They wonder how many pictures of John you wants him to take/retake t

Since A-traces can be interpreted in a way equivalent to pronouns, this is mysterious. It would not be if A-bar traces were different in this respect, but how could this difference be captured theoretically? The obvious option is Case: previously we could have said that traces in Case positions are different from traces in non Case position. In a Probe/Goal plus EPP system, we would need to formulate this differently (A-bar traces would be in Specifier position of a head which both has a Case feature and an EPP feature). The problem with such an approach is that it is totally ad hoc:

604 Dominique Sportiche

nothing in these properties (Case or Case +EPP) seems to say anything about why traces should be different. Assume with Chomsky ( 1 999) that representations are incrementally built by iterating Merge and Move (=Remerge) by merging at the edge of already built constituents. The idea that there are PHASES means that syntactic processes (basically Merge and the probe/goal system) only have access to a window of structure of limited size at any point. As a result, conditions operating over unbounded contexts - such as Condition C, the condition on Pronominal Binding, Polarity item licensing etc. . . - cannot be 'strictly' syntactic.6 To simplify exposition, I will assume that Logical Form - the representation used to compute structure dependent meaning - is incrementally built this way to become an in principle unboundedly large representation. Given the above reasoning about A-movement, we saw that it is empirically (and also conceptually) natural to postulate that binding conditions are checked at Logical Form and that this is done cyclically. Putting all this together, it seems natural to suppose that the 'Binding Conditions are checked cyclically, at each phase boundary. In essence, this comes down to adopting an NP-structure like idea, that is the relevance of an intermediate point after all A-movement and all A-bar movement, but view it as part of the cycle. In case of A-movement, the maximal span of an A-chain (the set of successive A-positions occupied by an A-moved DP or a maximal A-chain) is included in a single phase. Coupled with the idea that traces can be interpreted like pronouns, this allows A-movement to destroy configurations violating Condition C. Let us illustrate with example ( 1 7). If we want the subject wide scope reading, the LF we get is (1 8), where PH is a single phase. In this phase, t is a trace, so is interpreted as them yielding the reading there are pictures of Bill such that it seems to him that they are fuzzy.

( 1 8) ( 1 9)

6

Pictures of B ill seemed to·him to be fuzzy [PH Pictures of B ill seemed to him to be t fuzzy]

Unless of course they are made bounded (e.g. by using the equivalent of successive cyclic movement).

Cyclic NP structure and trace interpretation 605

If we want the subject narrow scope reading, we get the LF in ( 1 9), in which the moved subject is not interpreted (surely a possible option7), against which Condition C is checked, yielding a violation. (20)

[PH seemed to him to be pictures of Bill fuzzy]

Let us tum now to a case of wh-movement in (20). If CP is a phase and TP is not, no violation of condition C arises: it would be circumvented by wh­ movement to [spec,CP] (leaving a trace behaving like a pronoun).

(2 1 )

[cp Which picture of John does [TP he like wkisR fliswre sf JehR ]]

If on the other hand Condition C is evaluated at the (non defective) TP boundary, we get the right result (within TP alone, the phrase which picture ofJohn is not a trace). 4.

Some consequences, sketched

First, for this approach to work, it must be that wh-movement (and more generally A-bar movement) from an A-position can never move past an A­ position within a single phase. Suppose this could occur as below:

(22)

[PH

• • •

wh-XP . . . DP . . . . twh.XP

• • •

J

With PH a single phase and DP a pronoun c-commanding XP, XP containing a name coreferential with the pronoun, we would not observe a Condition C effect. The simplest assumption probably is that wh-movement always target a phase external position (much the way it is assumed now: wh-movement is to the edge of a phase whence it moves to a position outside this phase). This conclusion affects in ways that I will not discuss here how we should conceive of syntactic structures (much more articulated structures are needed than standardly assumed), of the inventory of phases (TP, vP with a single specifier), and of the formulation of the Phase Impenetrability Condition (PIC) in ways that recall discussion of the late 1 970's on the choice of bounding nodes (Phases) and the formulation of the Subjacency Condition (PIC). 7 The Principle of Full interpretation, in whatever guise, will require each phrase to be interpreted but can naturally be interpreted as not requiring - although not preventing - any phrase to be interpreted twice.

606 Dominique Sportiche

Secondly, consider what happens to traces of wh-movement: given TP as a phase, a to-be-moved wh-phrase will have to be interpreted within this phases . Moving to the next phase up, the wh-phrase will have to be interpreted again. This means that wh-traces must always be interpreted as copies.9 This creates a problem with Antecedent Contained Deletion (ACD). As Fox (2002) discusses, if traces are interpreted as (near) literal copies, there is no way to get around via QR the infinite regress problem or the problem of identity between a non finite object and one of its strict subparts - ACD raises. This prediction seems correct for wh-movement viz. the (surprising) deviance of [which book that you did� did she like? Turning to QR, empirically this problem does not arise. One reason could be that Fox' (2002) approach to ACD is correct. Alternatively, this could be related to the fact that QR is bounded in a way that wh-movement is not: if QR can operate phase internally, we expect two correlated results: (i) the trace of QR to be treatable like a bound variable (ii) QR to be able to remove Condition C effects. Convincing examples are hard to construct 10 • The fol lowing examples from May and Fiengo ( 1 994) may be an illustration: (2 3 ) (24)

Mary always buys him whatever John' s other friends do Mary gave him [the same thing that John's mother did] for his birthday

Finally, note that the combined RizzilBelletti-Lebeaux's idea that Condition C cannot be violated anywhere while condition A must be satisfied somewhere gets a natural treatment. As an LF is incrementally interpreted cyclically, this interpretatiori will crash as soon as a condition C effect is found and will not crash if an anaphor is provided a proper antecedent on some cycle. Note further that if TP is a phase, a 'clause bound' anaphor will need to find an antecedent within its own TP (or vP ...) . If no valid antecedent is found within its minimal phase, it will have to move out of the phase - say 8

I

Note that this cannot be literally true: the wh-operator cannot be interpreted there . gloss over this problem, resolved in the D-splitting type of approach I defend in

Sportiche (2005).

9 We suppose, consistent with Chomsky's (2005) No Tampering Condition that

interpretation once computed cannot be changed.

lOA discussion of these questions would take beyond the limits of this article. Let it

be noted here that it should be shown that these examples would be deviant without VP Ellipsis.

Cyclic NP structure and trace interpretation 607

pied p ipe under movement - to avoid a crash. This may we ll b e a way to reduce the locality domain on (certain) anaphor binding . to the notion of phase (or cycle), echoing again discussions from the late 1970's and early 1 980's. We are led to postulating periodicity for checking binding conditions, an idea m uch like VR&W's NP structure, but a transform(ation) to Fourier's taste, after all. References Chomsky, Noam. 1995. The Minimalist Program. MIT Press Cambridge, Mass. Chomsky, Noam. 200 1 . Der ivation by phase. In Ken Hale. A life in Language, ed. Michael Kenstowicz, 1 -52. Cambridge (Mass.): MIT Press.2005 On Phases. Unpublished manuscript. MIT. Belletti, Adriana and Luigi Rizzi. 1988. Psych Verbs and Theta Theory. In Natural Language and Linguistic Theory 6, 29 1 -352. Fouri er, Zenon de. 1 980. Remarks on the Extended Linear Model. Unpublished Manuscript. MIT. Fox, Danny. 2002. Antecedent-contained deletion and the copy theory of movement. In Linguistic Inquiry 33:63-96. Lebeaux, David. 1 998. Where does the Binding Theory Apply? unpublished manuscript. NEC Research Technical Institute. Lebeaux, David. 1 99 1 . Relative Clauses, Licensing and the Nature of Deri vations. In Phrase Structure, Heads and Licensing, Susan Rothste i n and Margaret Speas (eds.), Syntax and Semantics 25, Academic Press. May, Robert and Robert F i engo 1 994. Indices and Identity. MIT Press, Cambridge, Mass Riemsdijk, Henk van and Edwin Will i ams. 1 98 1 . NP-Structure. In The Linguistic Review 1 , 1 7 1 -2 1 7. Sportiche, Dominique. 2005. Division of Labor between Merge and Move: Strict Locality of Selection and Apparent Reconstruction Paradoxes, in Division of Linguistic Labor: the La Bretesche Workshop, downloadable from http://www.linguistics.ucla.edulgeneral/Conf/LaBretesche/index.htm. Wil liams, Edwin. 2003 . Representation Theory. MIT Press, Cambridge, Massachusetts. .

Appositive and parenthetical relative clauses Tim Stowell

1.

Appositive versus restrictive relatives

Appositive relative clauses differ from restrictive relative clauses in a number of ways. The fundamental distinction is semantically based: an appositive relative like that in ( l a) conveys an independent assertion about the referent of its associated head; the reference of the head is established independently of the appositive relative. In contrast, a restrictive relative like that in (1 b) is interpreted as an intersective predicate modifier, restricting the reference of its head. ( 1 ) a. The prince, who was wounded� withdrew from the battle. b. The prince who was wounded withdrew from the battle .

Correlating with this semantic distinction, several syntactic differences between appositive and restrictive relatives have been attested in English, most of which are discussed by lackendoff ( 1 976), who explains most of them in terms of the semantic distinction noted above. I will discuss j u st a subset of these differences here. First, internal to the relative clause, appositive relatives require overt wh-pronouns to occur in the Spec ,?f CP, as shown by (2a-c), while restrictive relatives do not, as shown by the existence of so-called that­ relatives l ike (2e) and zero-relatives like (2f). (2) a. b. c. d. e.

Max wants to visit Doctor Brown� who his sister works for. *Max wants to visit Doctor Brown, that his sister, Wor.ks for. *Max wants to visit Doctor Brown, his sister \\io�ki for. Max wants to visit the doctor who his sister works for. Max wants to visit the doctor that his sister works for. f. Max wants to visit the doctor his sister works for.

Second, appositive relative clauses naturally allow certain types of pied piping, an example of which is illustrated by this sentence� which are either awkward or ungrammatical with restrictive relative clauses.

Appositive andparenthetical relative clauses 609

Third, though definite and indefinite DPs may function as the head of either type of relative, certain types of QPs (including negative QPs) do not allow appositive relatives (3b). Conversely, appositive relatives may occur with proper names (4c) and, marginally, with reflexive pronouns (4a), whereas restrictive relatives may not (4b,d). (3 ) a. Nobody that George knows is qualified for this position . b. *N obody, who George knows, is qualified for this position. (4) a. b. c. d.

?Mary decided to hire herself, who needed the cash . *Mary decided to hire herself that needed the cash. Mary decided to hire Jim, who needed the cash. *Mary decided to hire Jim that needed the cash.

2. Appositive versus parenthetical relatives

Appositive relatives are also parenthetical. This is reflected in their pronunciation, which is intonationally marked in a way that is typical of parenthetical constituents, and which is conventionally represented in English orthography by means of commas or parentheses. Sentence-medial parentheticals are surrounded by commas, while sentence-final parentheticals are introduced by a comma; both types can instead be enclosed in parentheses. The fact that appositive relatives are interpreted as independent assertions is often attributed to their parenthetical status, since certain o�her types of parenthetical constituents, including parenthetical co-ordinate clauses, have the same kind of interpretation. Indeed, appositive relatives can often be paraphrased by means of other types of parentheticals. Emonds ( 1 976, 1 979), Safir ( 1 986), and Potts (2002) have specifically addressed the parenthetical status of appositives within the context of a general theory of parentheticality. This close association between the semantics of appositive relatives and parentheticality has led to the notions of 'appositive relative' and 'parenthetical relative' often being conflated, so that these terms are often used as virtual synonyms. In this short study I propose to examine the relationship between appositive relatives and parenthetical relatives more closely. While it may be true that all appositive relative clauses are parenthetical, I will show that not all parenthetical relatives are appositive. In particular, I will show that restrictive relatives may occur as parentheticals of a particular type, which I will refer to as 'repair parentheticals'. These have a type of interpretation

6 1 0 Tim Stowell

that is distinct from that of the canonical type of parentheticals. In some cases, the ability of a restrictive relative clause to function parenthetically seems to be contingent on its containing a parenthetical adverb; I will propose a derivational account of this involving pied piping. 3.

Not all parenthetical relatives are appositive

If all parenthetical relative clauses were appositive, as is widely assumed, then that-relatives and zero-relatives should be incapable of functioning as parentheticals. This turns out to be almost correct for zero-relatives, but not for that-relatives: The guy who you sold your car to was arrested today. The guy that you sold your car to was arrested today. The guy you sold your car to was arrested today. The guy next door (who I sold my car to) was arrested today. *The guy next door (I sold my car to) was arrested today. f. The guy next door (that I sold my car to) wa's arrested today.

(5) a. h. c. d. e.

The grammaticality of (Sf) shows either that not all parenthetical relatives are appositive, or that not all appositives require an overt relative wh­ pronoun. The correct answer, I believe, is the first: not all parentheticals are appositive. Three considerations support this conclusion. First, the relative clause in (Sf) has the flavor of a mid-sentence amendment, as though the speaker, having uttered the guy next door, decides that a further restriction is necessary to unambiguously identify the referent, and adds the parenthetical relative to achieve this end. Though the relative clause is parenthetical, its semantic function is to restrict the reference of the head. Second, parenthetical that-relatives similar to that in (Sf) can occur with quantified heads, something that ought to be impossible if they were appositive, especially in the case of (6b), where the QP does not refer:

(6)

a.

All the students (that I have managed to speak to, at least) support the president. b. None of the faculty (that I know of, anyway) have said they will attend.

In both examples, the relative clause clearly functions as a restriction on the quantifier.

Appositive and parenthetical relative clauses 6 1 1

A third reason for assuming that the parenthetical that-relatives in (Sf) and (6) are restrictive relatives, rather than appositive that-relatives, is that they may never have proper names as their heads: (7) a. *Jack Martin (that was a grad student here) likes trees. b. *Hilary (that Bill met in college, reportedly) works hard. Thus, it seems that parenthetical relative clauses can be restrictive. 4.

Restrictions on Parenthetical restrictive relatives

The parenthetical restrictive relative clauses in (Sf) and (6a-b) are all interpreted as mid-sentence corrections - either repairing an insufficiently explicit definite description (Sf) or scaling back the denotation of a QP (6a-b). Interestingly, this type of parenthetical restrictive relative clause seems to be impossible with indefinite heads. Parenthetical restrictive relatives like those in (6a-b) are analogous to parenthetical DPs or QPs (8a) or quantificational adverbs (8b): (8) a. All the students (all the American ones, at least) support the president. b. We are satisfied with your work (somewhat, anyway). These are all examples of what I will refer to as repair-parentheticals; they are used to revise, or amend, the assertion of the main clause, rather th,an supplementing it with an independent . and consistent assertion. Within the domain of relative clauses, restrictive relative clauses have a semantics that enables them to function as repair-parentheticals, whereas appositive relatives have a semantics that enables them to function as the standard type of parenthetical, conveying an independent assertion that is consistent with that of the main clause. Coordinated constituents and certain types of adverbs may also function in this way: (9) a. b. c. d. e. f.

Martha loves her secretary, who loves her too. Martha loves her secretary (and he loves her too) All the students (and many of the faculty) support the president. All the students support the president (and will vote for him too). All the students support (and depend on) the president. All the students support the president, unfortunately.

6 12 Tim Stowell

Henceforth I will refer to these 'standard parentheticals' as additive parentheticals. If parenthetical relatives may in principle be either appositive or restrictive, the question arises why zero-relatives like those in (Se) and ( 1 Oa-b) are disallowed as parentheticals, in contrast to that-relatives like (Sf) and (6a-b): (5) e. *The guy next door (I sold my car to) was arrested today. f. The guy next door (that I sold my car to) was arrested today. ( 1 0) a. * All the students (I admire very much) support the president. b. *Max gave a book (I tried to borrow once) to my sister.

It cannot be because zero-relatives do not contain an overt wh-phrase, since the same is true of the parenthetical that-relatives in (Sf) and (6a-b). Nor can we assume that the complementizer that may behave like a wh-pronoun when it occurs in a relative clause, because this would undermine the traditional account of why that-relatives may not function as (true) appositives in examples like (7a-b). Actually, zero-relatives are not, in fact, completely excluded as parentheticals. Though (Se) and ( l Oa-b) are clearly much worse than (Sf) and (6a-b), examples like those in ( 1 1) sound only marginally less acceptable than (Sf) and (6a,b) to my ear: ( 1 1 ) a. 'IAll the students (1 have managed to speak to, at least) support the president. b. ?None of the faculty (1 know of; anyway) have said they will attend. c. 'IAn old man (Bill met in Kansas, reportedly) has moved in next door. . ? d. "Max gave a book (he later spilled coffee on, unfortunately) to his ,teacher. ' The zero-relatives in ( 1 1 ) and · the that-relative� in (6a�b) differ from the zero relatives in (Se) and ( l Oa-b) in that the former include an additional parenthetical adverb of qualification, evaluation, or evidentiality. The contrast between the two sets of examples seems to hinge at least partly on this. ' One might be tempted to conclude that the parenthetical adverb simply serves a performance function, reinforcing the parenthetical status of the relative, which is otherwise difficult to distinguish from its non.

.

.

.

.

Appositive and parenthetical relative clauses 6 1 3

parenthetical counterpart. This seems dubious, however, given that the parenthetical can be clearly identified by its intonation contour. A second , more interesting, possibility is that the parenthetical adverbs in (6a-b) and ( I I a-d) actually have a grammatical licensing effect on the relative clauses that contain them, enabling them to function as parentheticals. 5.

A Derivation for Repair-Parenthetical relatives

I propose to explain the adverbs' licensing effect in (6a-b) and ( I l a-d) as follows. First, I will assume, as many previous studies of parentheticals have done, that parenthetical constituents ( including adverbs) occupy a structural position distinct from that of their non-parenthetical counterparts. Emonds ( 1 976, 1979) located this position as being right-adjoined to the main clause. Parenthetical adverbs may transparently surface in this position, as in ( 1 2):

( 1 2) a. b. c. d.

All the students support the president, at least. None of the faculty have said they will attend, anyway. An old man has moved in next door, reportedly. Max gave a book to his teacher, unfortunately.

Emonds assumed, with Ross ( 1 967), that parentheticals originate in this sentence-final position; this implies that, when parentheticals occur sentence-medially, as in ( 1 3 ), either the adverbs must have moved leftward, or else some portion of the surrounding clauses must have moved around them, or both. ( 1 3) a. b. c. d.

All the students, at least, support the president. None of the faculty, anyway, have said they will attend. An old man, reportedly, has moved in next door. Max gave a book, unfortunately, to his teacher.

' Whereas Ross assumed leftward (and downward) adverb movement, Emonds provided compel ling evidence that a portion of the surrounding clause moves to the right of the parenthetical. He showed that the portion of the main clause following the parenthetical (henceforth, the coda) must always form a single constituent; thus, examples like ( 1 4) are excluded because the parenthetical relative clause is followed by a non-constituent:

6 1 4 Tim Stowell

( 1 4) a. * He shouldn't have sent that kid, who is so contentious, out that record. h. *The prince of Holland, which is a country of honest businessmen, with the most stock visited NATO headquarters. His analysis captured this by positing a derivation for sentence-medial parentheticals whereby a constituent immediately preceding the parenthet­ ical (the coda) moves to the right of it; thus, ( 1 3a-d) would originate as ( I 2 a-d), with the main clause VP undergoing rightward movement to derive the examples in ( 1 3). (A derivation consistent with Kayne's antisymmetry theory would require a more complex derivation here, involving a left­ peripheral merge position of the parenthetical, with leftward movement of the coda to a position immediately following the parenthetical, followed by leftward movement of the remnant of the main clause to a position preceding the parenthetical.) Emonds's analysis has the disadvantage, however, of being unable to distinguish between cases like ( 1 2) and ( 1 3), where the parenthetical adverb modifies the entire clause, and cases like ( 1 1 ), where it modifies just a subconstituent (in this case, the relative clause). Since, on his account, parentheticals of both types are supposed to originate in the same position (peripheral to the main clause), the distinct interpretations of the adverbs in ( 1 1 ) and ( 1 2/ 1 3) cannot be captured in terms of assigning the adverbs to distinct initial-merge positions. If, however, parenthetical adverbs originate (initially merge) in the same positions as their non-parenthetical counterparts, and undergo movement to Emonds' s parenthetical position peripheral to the main clause, then the interpretative distinction between the main clause adverbs in ( 1 2/ 1 3) and the relative clause adverbs in ( 1 1 ) can be captured in terms of their distinct source (initial merge) positions: in the main clause in ( 1 2/ 1 3) and within in the relative clause in (1 1 ). Although parenthetical adverbs often seem to occupy the same (linear) positions as their non-parenthetical counterparts, casting suspicion on the type of parenthetical movement suggested here, their distribution is not in fact identical. Many classes of adverbs, consisting of those that Cinque ( 1 999) assigns to a relatively high source position in the functional structure of the clause, can surface in clause-final position only when they are parenthetical: ( 1 5) a. John has robbed a bank, allegedly/regrettably. b. *John has robbed a bank allegedly/regrettably.

Appositive and parenthetical relative clauses 6 1 5

This suggests that Emonds's right-peripheral position is available to these classes of adverbs only as a result of parenthetical movement. Cinque's lower adverbs can occur sentence-finally even if they are not parenthetical, but he provides a distinct derivation for these, involving leftward movement of the VP around them (though not to a position to the left of the higher adverbs such as those in ( 1 5). Turning back now to the licensing effect of the adverbs in ( 1 1 ), we can understand this to involve pied piping of the relative clause to the derived parenthetical position of the adverb, foHowed by Emonds's rightward movement of the coda around it (or the antisymmetric variant of this derivation described above). The idea is that the restrictive relative clause cannot get to this position under its own steam, though it may pied pipe along with a parenthetical adverb to this position. This account now leads us to reconsider the status of parenthetical appositive relative clauses, as well as parenthetical restrictive relative clauses like that in (Sf), neither of which requires an adverb (at least an overt adverb) to undergo parenthetical movement. Here I have little to offer in the way of an explanation. Perhaps appositive relatives can undergo parenthetical movement without a licensing adverb because they must be interpreted parenthetically. Put another way, perhaps some semantic feature of an appositive relative triggers this movement, making an adverb unnecessary . This still leaves parenthetical that-relative clauses such as that in (Sf) unaccounted for, however. This type of relative clause differs semantically from the repair-parentheticals in ( I l a-b). First, it lacks a quantificational head; its head is interpreted with specific reference. Second, its function is to clarify the reference of its head, rather than to revise or scale back the scope of the original claim. Actually, the same is true of the relative clauses in ( l I e-d). In both (Sf) and ( l I e-d), it is less clear that the interpretation is unambiguously restrictive than in the case of (I l a-b). Moreover, the relative c lauses in ( 1 1 c,d) cannot occur with quantificational heads like those in ( 1 1 a-b), which would be consistent with their actually being appositives, despite the lack of an overt wh-phrase. ( 1 6) a. * All the students (B ill met in Kansas, reportedly) support the president. b. *None of the faculty (he later spilled coffee on, unfortunately) have said they will attend.

6 1 6 Tim Stowell

On the other hand, these relative clauses may not occur with proper names as their heads, contrary to what we would expect if they were appositive: ( 1 7) a. *Sam, (that) Bill met in Kansas, reportedly, supports the president. h. *George, (that) he later spilled coffee on, unfortunately, may not attend. Curiously, the correct descriptive generalization seems to be that relative clauses lacking both an overt wh-phrase and an adverb of the type found in ( 1 1 a-b) may function parenthetically only if the nature of their head does not serve to identify them unambiguously as either a restrictive or as an appositive relative. I leave the explanation for this generalization to further study. Before closing, I should comment on one important difference between parenthetical adverbs and parenthetical relatives: while main-clause parenthetical adverbs can occur sentence-finally, parenthetical relative clauses must be string-adjacent to their head. This can be accounted for in terms of the theory of cyclic linearization of syntactic structure proposed by Fox and Pesetsky (2004), provided that it is assumed that parenthetical movement is the type of movement that does not operate successive­ cyclically, but rather is subject to the principles of linearization that give rise to Holmberg's generalization in the case of object shift. Provided that the relative clause, like other subconstituents of DP, must be linearly ordered within the clause, if parenthetical movement disrupts the ordering instructions imposed by cyclic linearization, this will trigger repair movement(s) to restore the original order, so that the relative clause will have to surface adjacent to its head in clause medial position. References 1999. Adverbs and Functional Heads, A Cross-Linguistic Perspective, Oxford University Press, Oxford. Emonds, Joe. 1 976. A Transformational Approach to English Syntax, Academic

Cinque, Guglielmo.

Press, New York Emonds, Joe. 1 979. Appositive Relatives Have No Properties, Linguistic Inquiry .

1 0.2, 2 1 1 -43 . Danny, and David. Pesetsky. 2004. Cyclic Linearization of Syntactic Structure, Mansuscript, Massachusetts Institute of Technology, Cambridge. Jackendoff, Ray. 1 976. X' Syntax: A Study of Phrase Structure, MIT Press)

Fox,

Cambridge, Mass.

Appositive and parenthetical relative clauses 6 1 7 C. 2002. The Lexical Semantics of Parenthetical-As and Appositive- Which� Syntax 5 . I J 5 5-88. Ross, J. R. 1 967. Constraints on Variables in Syntax, Ph.D. Dissertation,

PottsJ

Massachusetts Institute of TechnologYJ Cambridge. Safir, Kenneth. 1 986. Relative Clauses in a Theory of Binding Linguistic Inquiry 1 8.4, 663-89.

and Levels,

Overt infinitival subjects (if that's what they are) Anna Szabolcsi

1 . Introduction The interpretive contrast between ( 1 ) and (2)-(3 ) indicates that in the latter examples too associates with the infinitival subject: (I)

Mary wants to be tall. I want to be tall too. 'I, too, want to be tall'

(2)

Trees are tall. 1 want to be tall too. ' I want it to be the case that I, too, am tall'

(3 )

Mary slipped on the wet floor. I don't want to slip too. ' I don't want it to be the case that I, too, slip'

The second sentence in (3) has two near-equivalents in Hungarian, (4) and (5) : (4)

Nem szeretm!k [szintenJ* is elcsuszni] . not would.liketsg likewise/*too sliPinf ' I wouldn't like for it to be the case that I likewise slip'

Nem szeretnek [en is elcsuszni] . not would.liketsg Inom too SliPinf 'I wouldn't like for it to be the case that I, t�q, slip" . . · Krifka ( 1 998) argues that str�ss�d . postposed i dclifive ariic l�s associate · with a clausemate constrastive topic, which need not be overt as long as it satisfies the appropriate contextual role. English too is stressed. Of the two Hungarian particles, szinten is stressed, is is not.

(5)

�

�

Overt infinitival subjects 6 1 9

Although the fact that toolszinten may apparently associate with a covert subject is interesting, this squib wishes to draw attention to the 1 puzzle of (5).

2. The phenomenon The immediately striking property of (5) is that the infinitival complement has an overt and, probably, nominative subject (something as unusual in Hungarian as it would be in English). 2 Thanks to this fact, the construction may contain any operator which, unlike szinten, needs to associate with an overt stressed element: is 'too', as in (5), nem ' not ' , as in (6), csak 'only', as in (7), or the subject may be focussed on its own, as in (8). Consider (6) through (8) in a context where the car available is too small for the whole party and someone must take the bus.

,

(6)

lobb szeretnek [nem en menni busszal]. better would. likelsg not Inom gOinf bus-with 'I'd prefer for it not to be me who takes the bus'

(7)

menni busszal] . Nem akarok [csak en bus-with not wantlsg only Inom gOinf 'I don't want to be the only one to take the bus'

(8)

3 HOLNAP fogok [EN menni busszal]. tomorrow willlsg Inom gOinf bus-with 'Tomorrow is when it will be me who takes the bus'

Not just the interpretations make it clear . that the pronoun belongs within the infinitival complement. Whereas en is, csak en and plain focussed en might occur in postverbal position in uncontroversially monoclausal examples (although with a different interpretation than in the examples I Although the construction is productive and colloquial, the data in this squib has not, to my knowledge, been noted in the literature. I discovered it in 1 988 and have over the years made several attempts to come to grips with it. I thank Zoltan BAnreti, Huba Bartos, Andras Koml6sy, Maria Polinsky, Marta Abrusan, and Chris Collins for discussions. 2 Hungarian has over 20 case suffixes, but the nominative is morphologically unmarked. 3 Small caps indicate focus accent. I mark it only where the phrase is not accompanied by an overt focussing operator like csak . only', etc.

620 Anna Szabolcsi

above), nem en cannot; therefore its position in (6) can only be analyzed as immediately preverbal within the infinitival complement. Moreover, the focussed pronoun at hand can be preceded and scoped over by the same operators within the infinitival complement as the focus of any finite clause or run-of the-mill infinitival clause can (Koopman and Szabolcsi 2000, 1 1 8- 1 2 1 ). Compare (9) with ( 1 0) and ( 1 1 ). (9)

Nem vagyok haj land6 [mindent EN csimllni willing everythin&cc I nom dOinf not am helyetted]. instead.of.you 'I am not willing to do everything for you (i.e. to have a situation where for everything, it is me who does it for you)'

( 1 0)

Mindent EN csinalok helyetted. everythingacc Inom d01sg instead.of.you 'For everything, it is me who is doing it for you'

(1 1)

Nem vagyok haj land6 [mindent INGYEN csinalni not am willing everythingacc for.free dOinf helyetted] . instead.of.you 'I am not willing to do everything for you for free (i.e. to have a situation where for everything, it is for free that I do it for you)'

The construction is only possible with subject control verbs and the future auxiliary Jog-.4 For example, the following are unacceptable, irrespective of the relative order of the infinitival and the matrix clauses, and irrespective of whether the matrix subject is arbitrary PRO or an overt dative: ( 1 2) a. * [Csak en menni busszal] i j6 . lenne. · '' only I gOinf bus-with good would.be , ' It would be good if only I took the bus' b. *J6 lenne (nekemlneked) [csak en menni busszal] . good would.be I-for/you-for only I gOinf bus-with 'It would good (for me/for you) if only I took the bus'

4

Hungarian has a narrower range of infmitival complements than English, so not all environments that might come to the reader's mind are possible to check.

Overt infinitival subjects 62 1

Given subject control, the infinitival subject invariably agrees with the matrix verb. Compare the examples above with ( 13). The construction seems to be subject to two main constraints of its own. First, the infinitival subject has to be a pronoun, although it is not restricted with respect to person and number: ( 1 3)

( 1 4)

Nem akar [csak 0 menni busszal] . not want3sg only 3sg.nom gOinf bus-with 'He/she doesn't want to be the only one to take the bus' *Nem akar [csak Peter menni busszal] . bus-with not want3sg only Peter gOinf 'Peter doesn't want to be the only one to take the bus'

Second, examples with an overt subject in the matrix clause are judged unacceptable or degraded: ( 1 5)

?? Peter nem akar [csak 0 menni busszal] . Peter not want3sg only 3sg.nom gOinf bus-with 'Peter doesn't want to be the only one to take the bus'

The challenge is to account for these constraints, together with the fact that the infinitival subject is capable of being spelled out at all. Two other properties shared by the examples need not be particular to th is construction. One is that the matrix clauses all contain negation, focus, or a verbal modifier. This is simply because otherwise the infinitival material would be broken up by inversion or climbing (Koopman and Szabolcsi 2000); so this certainly does not need a special explanation. The other property is that the pronoun in the infinitival complement is always focussed and/or modified by an operator. The reason may be that it would be pro-dropped otherwise. But it is possible that, in addition, emphasis is important in the construction at hand; see below. 3.

Some analytical possibilities

The fact that the infinitival subject is spelled out and the matrix subject cannot be might suggest that the construction at hand is an instance of backward control (Polinsky and Potsdam 2002). Backward control is a phenomenon brought to light by Hornstein's ( 1 999) proposal that control is to be subsumed under raising. On that analysis, the elements traditionally

622 Anna Szabolcsi

known as the controller and PRO are copies of the same thing. Some languages allow the lower copy to be spelled out: this is backward control. But the rest of the phenomenology does not support the backward control analysis. First, the overt element in backward control does not have to be a pronoun, so the unacceptability of ( 1 4) is not predicted. Second, there tends to be a division of labor between forward and backward control in languages that attest both, with different control verbs participating in each pattern. In contrast, all subject control verbs in Hungarian participate in the overt infinitival subject construction, as long as the result makes sense. The list above can be easily extended: ( 1 6)

Utalok [mindent EN csinalni helyetted] . hate ) sg everythingaee Inom dOinf instead.of.you 'I hate to be the one doing everything for you'

(17)

Context: A postcard was supposed to be signed by all members of the fami ly. Elfelejtettem [en is alairni] . forgot )sg'defobj Inom too signinf ' I forgot to sign it too (i.e. to bring it about that I , too, sign it)'

( 1 8)

Context: The actor's career was in decline. Elkezdte [nem 0 kapni a foszerepeket] . began 3sg.defobj not henom getinf the leading.rolesaee 'He began not to be the one getting the leading roles'

Finally, the pronoun is focussed and ·thus in an operator position. If it formed an A-movement chain with the matrix subject, this would be a case of improper movement. A second analytical possibility is that the pronoun is in fact the matrix subject, and the surface order is a result of a series of leftward movements. This would work roughly as fol lows. I use English glosses to make the derivation easier to decipher: ( l 9) a. b. c. d.

=> prepose [Inom want)sg [cp PRO gOinf bus-with] . . . ] => focus [xp [cp PRO gOinf bus-with]i [Inom want)sg ti ' ' ' ]] [FoeP Inomj [xP [cp PRO gOinf bus-with]i [tj want) sg ti . . . ]]] => prepose [yp [tj want )sg ti . . ' ] k [FoeP Inomj [xp [cp PRO gOin f bus-with]i tk ]]]]

Overt infinitival subjects

623

But while this analysis would explain why the only overt subject appears to be inside the infinitival complement, it would not explain the pronoun restriction. This might not be a problem if the unacceptability of ( 1 4) can be attributed to Condition C. But then the binding theoretic explanation does not make it obvious why an overt pronoun in the same position does not violate Condition B. If the focussed subject pronoun is replaced by a focussed non-subject pronoun, the result is marginal, and a reflexive is 5 highly preferred : (20)

Nem akartam [csak magamat/??engem emliteni]. mentioninf not wanted 1 sg only myselfacclIacc 'I didn't want to mention only myself / ??me'

Moreover, this analysis would rest on a large number of coincidences in creating exactly the same ordering patterns that preverbal operator sequences standardly exhibit, see (9) through ( 1 1), and it is not clear where the negation preceding wantlsg, still absent from ( 1 9), would be merged to able to scope over the infinitival material. A third analytical possibility, suggested to me by Marta Abrus8n, is that the pronoun in the infinitival complement is not a subject at all. Instead, it is an emphatic element. On this analysis, (5) is a variant of (2 1 ): (2 1 )

Nem szeretnek [(en) magam is elcsuszni]. not would.like l sg I myself too sliPinf 'I wouldn't like for it to be the case that I, too, slip '

The advantage of this analysis is that it explains why the infinitival subject cannot be a full non-pronominal DP, and that it provides a more specific reason for its emphatic character than the pro-drop reasoning. On the other hand, there remain some differences between the two constructions. One is that a plain pronoun may serve as an emphatic element here. The matrix subject of a control construction is generally too local for a pronoun - see (20). Two, it remains unexplained why the matrix clause is degraded with an overt subject. The bias against overt matrix subjects is not in effect when an emphatic reflexive is present.

5

A subject control verb in Hungarian exhibits defmiteness agreement with the direct object of the infinitival complement; but the t sg.past fonn is the same in

both conjugations.

624 Anna Szabo/csi (22) a.

Peter szeretne [(8) maga feladni a levelet]. Peter would.like3sg he himself mailinf the letteracc 'Peter would like to mail the letter himself b. ?? Peter szeretne [6 feladni a levelet] . Peter would.like3sg he mailinf the ietteracc 'Peter would like for it to be him who mails the letter'

The relationship between the two constructions is also weakened by the fact that examples like ( 18) are entirely natural with a plain pronoun but not with an emphatic reflexive, as in (23): (23)

4.

Context: The actor's career was in decline. * Elkezdte [nem (0) maga kapni a foszerepeket] . himself getinf the leading.rolesacc began3sg.defobj not he 'He began not to be getting the leading roles himself

Conclusion

At this point the analysis and, therefore, the theoretical significance of the data remains unsettled. The emphatic pronoun analysis, suggested by M. Abrusan, m ight eventually be correct, if the difficulties mentioned above can be resolved in view of a better understanding of the working of the emphatic pronoun+reflexive complex. References Hornstein, Norbert. 1 999. Movement and control. Linguistic Inquiry 30: 69--96. Koopman, Hilda, and Anna Szabolcsi. 2000, Verbal complexes. Cambridge, Mass.: MIT Press. Krifka, Manfred. 1 998. Additive particles under stress. In Semantics and Linguistic Theory 9, · ed. by Devon Strolovitch and Aaron Lawson, 1 1 1-1 30. Ithaca: CLC Publications. Inquiry 33: Polinsky, Maria. and Eric Potsdam. 2002. Backward contro l. Linguistic .

245--282.

Wanna and the prepositional complementizers

of English

Tarald Taraldsen

1.

Introduction

Much of the most exciting work in theoretical linguistics has been focussed on how the different components of grammar interact. For example, Van Riemsdijk and Williams ( 1 98 1 ) (and, more recently, Williams (2003) motivate a particular organization of grammar by showing how it leads to correct predictions regardin g the interaction between different types of syntactic processes as well as the interaction between syntax and the interpretive components. In this squib, I want to suggest that one phenomenon of English grammar that has figured relative ly prominently in discussions about the syntax/phonology interface and the A- vs A'-movement distinction, wanna-contraction, may not be quite what it has been taken to be. In particular, when properly understood, wanna­ contraction and its limitations will provide no argument for saying that wh­ traces ever interact with phonology, and, therefore, no argument for separating A'-movement from A-movement (whose traces should differ from those arising from A'-movement by being invisible to phonology, in the classical accounts). ,�

2. Background The obselVation that the embedded ECM subject of ( 1 ) seems to continue preventing want and to from contracting into wanna even after wh­ movement has removed it has been taken as evidence that (some) wh-traces are visible to some PF processes:

(1) (2) (3)

She wants John to win? Who does she want to win? * Wh o does she wanna win?

Alte rnative ly, and perhaps less artificially, PF processes may be allowed to apply before types of syntactic movement, as in V an Riemsdijk and Williams ( 1 98 1 ) or perhaps in systems with cyclic (phase-wise) (partial) spell-out.

626 Tarald Taraldsen

In the sections to follow, I will outline a rather different approach, one in which the derivation of (4) and other standard cases of wanna­ contraction need not crucially involve contraction at all: (4)

Who does she wanna invite?

By the same token, there will be an account of (3 ) which doesn't invoke contraction either. 3.

Variation across varieties of English

There are regional varieties of English where (3) is in fact allowed, although, of course, wanna continues to be impossible in ( 1 ), e.g. South­ African English. In such varieties, wh-movement does in fact enable wanna-contraction to apply. Within traditional generative approaches, it seems as if one m ight have to account for this by stipulating that some languages allow wh-traces to be invisible to the relevant portion of the phonology, as a parametric option. Within the model proposed by Van Riemsdijk and Williams ( 1 98 1 ), one m ight say that contraction applies at the level of NP-structure in languages rejecting (3), but at a higher level in languages accepting it. The alternative I will adopt, however, is the following: Certain varieties of English allow want to contract with adjacent to within its clausal complement, in a way comparable to the way want can contract with a to give wanna in the pronounciation of I want a glass of water. These varieties will allow contraction of want and infinitival to even when adjacency is brought about as a result of wh-movement. Other varieties will in fact never allow want and infinitival to to contract, although they too allow want a to become wanna in I want a glass of water. Noticing that want to --+ wanna seems ,to involve an extra step in comparison to want a -+ wanna, ,the loss of the iQ-itial t ofto iq addition to ' ' the final t of want assimilating to the preceding' n: we 'nligiit say that this extra step is unavailable in the varieties taken to disallow want to -)0 wanna, perhaps because it is subject to stricter hierarchically defined locality conditions. Thus, in these varieties, (4) doesn't involve contraction, and an alternative derivational path leading to wanna wil l be cut off, for purely syntactic reasons, in (3).

Wanna and the prepositional complementizers 627 4.

Prepositional complementizers in Rom ance

In Romance languages, infinitives may be introduced by two different prepositions, like Italian a 'to' or di 'of, traditionally thought of as a kind of infinitival complementizers. Their relative distribution mirrors in part the grammatical relation between the matrix verb and its infinitival complement:

(5)

ho ordinato di partire subito Gli himda, I-have ordered of leave at once

(6)

L'ho convinto a partire subito himacc.cL I-have convinced to leave immediately

A fact which will eventually become important, is that verbs that select di ' of in control environments still can have no prepositional complementizer introducing their infinitival complements in raising structures, e.g. sembrare 'seem' comes with an obligatory di in (7), an instance of object control, but disallows it in (8), an instance of raising:

(7)

Mi sembra *(di) aver capito me seems of have understood ' I think I have understood. '

(8)

Claudio sembra (*di) aver capito of have understood Claudio seems ' Claudio seems to have understood. '

Likewise, control requires the presence of di 'of in (9), but raising-to­ object (ECM) (obligatorily feeding into wh-movement in Italian) is incompatible with it:

(9)

Credo *(di) aver vinto I-believe of have won ' I believe I have won. '

( l 0)

Chi credi (*di) aver vinto? who you-believe of have won 'Who do you believe to have won?'

628 Tarald Taraldsen 5.

Prepositional complementizers in English

When one compares the inventory of prepositional complementizers available to infinitives in Romance with what English has to offer, it is striking that although English (like German) seems to have a counterpart of Romance a, i.e. to (zu), it apparently has no counterpart of di 'of . The Italian sentence in (5), for example, comes out in English as ( 1 1 ) rather than ( 1 2); ( 1 1) ( 1 2)

I ordered him to leave *1 ordered him a/ leave

Another difference separating English from Romance is that whereas the Romance a, like di, appears to combine with the top layer of the infinitival clause, the English to seems more deeply embedded, since it can follow the negation and some sentence adverbs. (The difference between Romance a and German zu is of course even more dramatic.) Also, the English to never interferes with raising (or ECM), unlike prepositional complementizers in Romance, plausibly reflecting a more deeply embedded position, on standard assumptions. Proceeding from the assumption that the two differences are related, we are led to the conjecture that whenever a language has a more deeply embedded prepositional complementizer, this will always be to, and no additional prepositional complementizer can appear on top of the whole infinitival clause. Speculating on why this conjecture should hold, a still rather impressionistic picture suggests itself. Suppose that prepositional compiementizers are Case-licensers of sorts. Given the partial dependence of the choice of a vs di on grammatical relations, we m ight, for example, say that different argument types (such as 'direct' vs 'indirect' objects) are initially assigned different Case-features, and that attraction by a or di, in the way described by Kayne ( I 999), is keyed to th is difference. To accommodate the English (and German) situation, we would then have to say that a subconstituent of the infinitival clause, an extended VP, is also assigned the Case-feature uniquely associated with ' indirect' objects in Romance, and is therefore likewise attracted by to (followed by subsequent remnant movement across to and the subpart of the infinitival clause previously attracted to it). Identifying the infinitival clause-internal to with an 'oblique' Case-licenser in this fashion, provides an explanation why ( 1 3) is ungrammatical, in contrast to ( 1 4) and the Italian ( I 5), since a subpart of

Wanna and the prepositional complementizers 629

the infinitival clause is now taken to occupy the position the ' indirect' object John would have to access for to .John to be possible: ( 1 3) ( 1 4) ( 1 5)

*1 prom ised to John to leave early 1 promised to John that we would leave early Ho promesso a Gianni di partire presto

The grammatical ( 1 6) would then be taken to involve the same Case­ licensing position for ' indirecf objects that makes double object constructions possible in English: ( 1 6)

1 promised John to leave early

Assuming now that the 'oblique' Case-marking of extended VP somehow exempts the full clausal projection from Case-licensing, we expect that no counterpart of di is called for to license it, i .e. to alone suffices. 6.

Forms of of

What if, exceptionally, that subpart of the infinitival clause which is usually attracted to to, should fai l to have the right triggering Case-feature in some environment? Within the scenario just sketched the prediction should be that whenever the infinitival clause is a 'direct' object it would need to be licensed by an English counterpart of di, e.g. of But of course, one never really sees English of introducing an infinitival complement. On the other hand, Kayne ( 1 997) observes that analyzing forms l ike shoulda as the results of the auxiliary have undergoing phonological reduction in combination with modals, e.g. in ( 1 7), fails to account for the fact that this reduction would only apply in environments where Scandinavian allows the auxiliary to be entirely absent

( 1 7) ( 1 8) ( 1 9) (20 )

We shoulda done it before. *We shalla done it by the time you return Vi sku lIe we should

-

*Vi skal we shall

gjort det fer done it before gjort det innen du done it by the time you

kommer tilbake return PRT.

630 Tarald Taraldsen

So, Kayne suggests that a is not to be viewed as a phonological variant of have, but rather as a variant of oj, a kin to a cup a tea, allowed to surface (in English) just in case there is no auxiliary have. Adopting this view of the a of shoulda, we find that the ungrammaticality of all sentences like (2 1 ) with of occurring as an infinitival complementizer, li ke Italian di, is not quite sufficient to justify the conclusion that (no form of) of ever occurs as a prepositional compIementizer in English: (2 1 )

*We want of win

In particular, one would need to make sure that the variant a doesn't ever occur with infinitives either, and I would now obviously like to suggest that it in fact does. 7.

Wanna without wanna-contraction

So, suppose that (22) is really a licit manifestation of (2 1 ), with of necessarily pronounced as a:

(22)

We wanna win

The illusion that a is a reflex of to (contracting with want) is, on this view, created by the fact that to must be absent for a (= oj) to emerge, for purely syntactic reasons. Why should (3) (repeated below) be ungrammatical with its a l ikewise analyzed as a variant of of!

(3)

*Who does she wanna win?

The answer is straightforward. If a is really o/with the same properties as the Romance prepositional complemeritizer di, we expect it , not to be able to appear in ECM and raising constructions, as seen in (7)-( I 0): (7)

Mi sembra *(di) aver capito me seems of have understood 'I think I have understood .'

(8)

Claudio sembra (*di) aver capito Claudio seems of have understood 'Claudio seems to have understood.'

Wanna and the prepositional

(9)

Credo *(di) aver vinto I-believe of have won ' I believe I have won.'

( l 0)

Chi credi (*di) aver vinto? of have won who you-believe 'Who do you believe to have won?'

compiementizers 63 1

But the source of (3) would have to contain an ECM construction with of (pronounced as a): (23)

*we want who of/a win

By way of contrast, any variety of English which actually allows the phonology to produce wanna from adjacent want and to, like ( by assumption) South-African English, will be correctly expected to find (3) grammatical. 8. A selection of loose ends The many loose ends sti1l flapping in the wind include most prominently the following: (a) What exactly is the relationship between the clause­ internal to of English, and the clause-initial prepositional complementizer claimed to manifest itself as the a (= of) of wanna? (b) Why must of be pronounced a as a prepositional complementizer? As for (a), a more well-worked out account than the one sketched above, should probably also yield an account of the fact that when EngJish must drop of, like Italian drops di, in raising and ECM constructions, clause-internal to must reappear: (24)

*Who do you want win?

In Romance, by contrast, the infinitive is allowed to remain without a complementizer in this situation. Related to this is the question why the clause-internal to is (exceptionally) allowed to be absent in (22), giving rise to clause-initial of/a. (In terms of the proposal put forward in section 5, the extended VP subpart of the infinitival clause usually associated with an 'oblique' Case­ feature must be allowed not to have one in (22).) As for (b), 1 would tend to think that the alof alternation should not really be looked upon as (purely) phonologically conditioned. Conceivably,

632 Tarald Taraldsen

at least, it might be profitable to adopt a decompositional view of the syntax of qf such that, say, the pronounciation a might really indicate that a piece of the syntactic structure otherwise spelling-out as of is missing in certain environments. An account along these lines should also be capable of accounting for the fact that whereas a cannot be pronounced of in (22), it can in environments like ( 1 7), where it is licensed by om ission of the auxiliary have . For now, these issues will have to remain open. Optimistically, I try to think of them as unsolved rather than unsolvable. Referen ces Kayne, Richard

S . 1 997. The English Complementizer of Journal o/Comparative

Germanic Linguistics, 1 , 43-54 S . 1 999. Prepositional Complementizers as Attractors, Probus, 1 1 , 39-73 Riemsd ijk, Henk C. van, and Edwin Williams. 1 98 1 . NP-structure . The Linguistic Review, 1 , 1 7 1-2 1 7 Williams, Edwin. 2003. Representation Theory, MIT Press, Cambridge, MA. Kayne, Richard

A note on asymmetric coordination and subject

gaps

Craig Thiersch

Since Tilman Hohle's pioneering work in asymmetrical coordination and (unexpected) subject gaps in finite sentences in German, there have been many different attempts at finding a solution for the constructions, which also appear in Dutch. ' In this squib I would like to focus on one aspect of the construction, namely the non-occurrance of the construction with object gaps. To review briefly, the problem arises due to the assumption that coordination joins 'like' entities in terms of absolute category, e.g., AP with AP and not AP with NP. That this is not entirely correct has been clear at least since Sag et al. ( 1 985). Rather, what seems to be relevant (among other things) is the functional status of the conjuncts: for example, an argument with an argument, an open predicate with an open predicate, or an assertion with an assertion: ( 1 ) a. They revealed [Fred's true identity] and [that he worked for the Mafia]. b. a diplomat [with great charm] and [whom everybody trusted] ." In the case in question, it appears as if, in 'traditional ' terms, a verb-final IP has been conjoined with a verb-second CP: (2)

Wenn Demand nach Hause kommt] und If someone to house comes and [da steht der Gerichtsvollzieher vor def TOr], ... there stands the bailiff before the door 'If someone comes home and the bailiff is standing there in front of (Hohle 1 990: (6» the door, ... '

I

Thanks to all the colleagues who have discussed these constructions in Dutch and German w ith me over the years, not the least of whom being Henk van Riemsdijk. Also thanks to Henk for providing a lively and inspiring environment over the years, at Tilburg University, in GLOW, and countless workshops and conferences, large and small.

634 Craig Thiersch

More mysterious are sentences like (3), cf. Hohle ( 1 983 ): (3)

Gestem kamen ein paar Studenten und yesterday came a few students and verteilten FlugbUitter distributed gap? pamphlets

Here both clauses are verb-second and hence there appears to be an unexplained gap in the subject position of the second conjunct. The second conjunct in examples like (2) can also contain a subject gap, and Hohle ( 1 990) showed conclusively that both are syntactically within the scope of of the complementizer wenn; for examp le, since they can occur within symmetrically coordinated VPs: (4)

Wenn jemand in die Wilste zieht und lebt dart von If someone in the desert draws and lives there from Heuschrecken oder sich im Wald verirrt hat und grasshoppers or self in-the woods errored has and sich von Wurzeln und Beeren, .. nahrt nourisches self from roots and berries .. ' if one retreats to the desert and lives there from grasshoppers or gets lost in the woods and lives off of roots and berries' (Hahle 1 990:(32a» .

.

Hohle showed in addition that that the two clauses in subject-gap examples like (3 ) do not share the Vorfeld, and indeed that the Vorfeld is asymmetrically extracted from the first conjunct; the second conjunct is an island. A solution which analysed the second conj unct as an adjunct modifier was proposed in Thiersch ( l 993a,b/1 994), henceforth T93, accounting for the V -2 word-order and the islandhood of the second conjunct. BUring and Hartmann ( 1 998), henceforth B&H, independently suggested that the second conjunct is an adjoined CP? While both approaches analyse the second clause as an adjunct, they are somewhat different, and make different predictions.

2 We refer the reader to these papers as well as Lechner (2000) for more discussion

of the data and characteristics of the construction, as well as references to, and discussion of, other approaches.

On asymmetric coordination and subject gaps 635

In particular, B&H analyse the second conjunct as a CP with an empty operator in its Vorfeld. The CP can be adjoined at various points in the tree, but necessarily at least below the real subject in the first conjunct, so as to semantically interpret the operator [their (9); the OPj added for clarity, my gloss] :

!

f

l

!

i

j I

(5) a. Gestem ging ich aus und traf Olaf Thon yesterday went l out and met Oiaf Thon b. [cp gestem ging [IP ich [I' [I' . . . aus . . ] [cp und [ep OPj [C' traf [IP ej O . T. ...]]]] ]]] .

Here the CP [und . .] is adjoined to 1'. The approach in T93 shares most of B&H's assumptions about coordination: 3 .

I

(6) a. Coordinated conjuncts have the same saturation level (shared by Hohle and B&H). Hence categories which count are the functional ones: argument, predicate (modifier), assertion (saturated predicate). b. Coordinators are underspecified heads, and features relevant to (6a) percolate, so coordinating a modifier AP and a modifier CP is a modifier, but neither an AP nor a CP: [PrcdP AP [ and CP]]; cf. Sag et al. ( 1 985).

1

T93 also shares with B&H the generalization in (7):4

I

(7)

a

i

i

I

The verb raises if not C-commanded within its own (extended) projection by a lexical complementizer.

However, this has a different effect when combined with (8): (8)

Travis's hypothesis: subject verb-second sentences are not the same as other verb-second sentences.

i

I

3 For the purposes of this squib we disregard the newer, '3-dimensional ' theories of coordination; cf. De Vries (2005). Hopefully the analy sis presented here can be incorporated in such approaches. 4 However implemented. Th is assumes relative clauses have a lexical but phonetically null head (which is phonetically realized in Bavarian as wo).

636 Craig Thiersch

Travis ( 1 984) assumed they were IP and CP, respectively, but given the plethora of functional categories on the left periphery, the precise categories are not clear, and T93 simply referred to them arbitrarily as Ub and Vt, respectively.s Furthermore, in addition to the usual assumptions about modification, T93 assumes the theory of modification in Kolb and Thiersch ( 1 99 1 ) : (9)

Certain adjuncts (modifiers) are licensed by an empty category in top projection layer which is coindexed to the projection being modified; e.g., [pp ej [PO DP]] (In particular, where the constituent being modified is an argument of the modifier; e.g., the movement of relative pronoun in relative clause turns the assertion into an open predicate).

K&T assumed that coindexing of this empty position assures that the link of the predicate to the structure it modifies satifies selectional restrictions of the modifier. Hence in the case of the asymmetrical coordination structures, the first conjunct could be such an empty category, l icensing the coordination as an adjunct: ( l 0)

[0dand XP ]]

Here the empty category 0i is coindexed to the projection which (10) modifies (say, VP), the coindexing assuring via the coordinate structure a minimal agreement between YP and XP in terms of (6a) above. The coindexing also assures the proper 'semantics, i.e., that the complete structure [yp [YPi ] [0i and XP]] gets a coordinate reading. For e�ample, the second conjunct could be: .••

.

.

. .

.

where the KPj is a CP or I P and the 0j is coindexed � ith (modifies) an assertion. Or it could be:

5 For ' basic

utterance' and 'topic utterance', i.e. assertion. We stick to CP/IP in this squib for simplicity.

On asymmetric coordination and subject gaps 637

where the 0j is coindexed with the main predicate (I') forming a complex predicate. The easy case is example (2), in which the second conjunct is a saturated predicate (assertion) CP and the 0j is coindexed to the matrix IP, also a saturated predicate. The second CP being an adjunct, its verb raises (to CO) and it is an island for extraction, as desired. In addition, it is in the scope of the matrix CO (e.g. 'wenn '), as noted by Hohle:

( 1 3)

[cp wenn [IP

[IPd emand nach Hause kommt] [cp 0k und [cp da steht .. , ]] ]]

More difficult at first glance is the subject gap construction (3,4), although it essentially works the same way. The coordinate structure is ( 1 2), an adjunct modifier of the main r and predicated of the subject DP. That is, we have an open predicate modifYing an open predicate, . which are both predicated of the same subject, as in

( 1 4)

John [[ate the meat] [naked]]

How does this proceed in the asymmetric coordination case? We merge the coordinate l' adjunct containing an non-raised empty subject ex (e.g. still in Spec, vP) with the matrix adjunct forming the complex I' in ( 1 5):

( 1 5)

[r. [I'k kamenj [D Pm ein paar Studenten] ej] [ I' 0k und [I' verteiltenj ex Flugblatter e; ]] ]

If the DPm, ein paar Studenten, raises to form an IP,6 it then C-commands the modifier I' as well as the matrix 1', allowing the bound reading of ex, i.e., as ej. (In essence we can regard this as a heretofore unnoticed case of an A-bound parasitic gap.) Given this flexibility, we need to ask why there are no paral lel object gaps: .

6

As noted above, asymmetrical extraction from the (across the board), is unproblematical.

first conjunct, i.e. , not ATB

63 8 Craig Thiersch

( 1 6) a. Gestem hat den Fritz eine Menge wUtender Kunden yesterday has theacc Fritz anom group furiousgen clients erwischt und (dann) haben ihn ihre Anwalte angezeigt caught and (then) have him their lawyers charged 'Yesterday a group of furious customers caugt Fred and their lawyers had him indicted.' b. **Gestern hat [den Fritz] j eine Menge wUtender Kunden elWischt und haben ej [dir.obj.} ihre Anwalte angezeigt. B&H's analysis would actually seem to predict that there should be DO gaps of a similar nature, since [cp OPj ej .. ] involving direct objects is quite natural in German (and much discussed in the l iterature): 7 ••••

( 1 7)

.

(Kennst du Fritz?) OPj kenne ich ej know I know you Fritz?

schon seit langem ! already since a-long-time

Without a theory of how the adjunct is licensed, such a structure could be attached anywhere, including low in the structure (as their subject-gap CP's are), allowing examples like ( l 6b), or even worse examples in which the direct object in the first conjunct has not been scrambled above the subject. On the other hand, suppose we assume the approach of T93, where there has to be some typological agreement as in (6a) between the second conjunct and the modified constituent in the matrix. Then for a direct object gap, the d irect object in the matrix clause would have to have been scrambled above the subject. (If not, the coordinate structure would need to be something less than an IP, say, a VP, in order to be low enough; but then it could not contain a subject, and there would be no possibility of verb­ second.) Suppose the direct object is scrambled as in ( 1 6b). There are two cases to be considered. First, suppose that the second conjunct is a CP with an empty operator: [KP 0j KO [CPi OPace eaec ]], where 0j is coindexed with the index of the matrix clause. As we can see by ( 1 7), structures l ike CPj apparently are saturated predicates which count as independent utterances. But in the asymmetric coordinations where the second conjunct is a saturated predicate as in (2), it must be attached h igher than IP, as one can see by the lack of the binding effects under negation noted in B&H: .•.

...

7 Lechner (200 1 ) also notes this shortcoming, although he claims the T93 approach has the same problem; we hope to show the opposite below.

On asymmetric coordination and subject gaps 639

steht hinter [keinem Lowen]i ( 1 8) a. 1m Zirkus Krone in-the circus Krone stands behind no lion eine Oompteuse und krault ihmi das Fell a trainer and scratches him the fur 'In the Circus Krone a trainer stands behind [no lion]i and scratches itsj fur.' (B&H (3 1 a), subject gap) b. *Im Zirkus Krone steht hinter [keinem LowenJ eine Dompteuse und sie krautt ihmi das Fell (My example, no subject gap) But if the CP is attached above IP, then it is not possible to license the coindexing of the OPace in its specifier, since it is not C-commanded by the scrambled matrix OPace, even assuming a segment-category distinction. Suppose the XP of the second conjunct is an IP with an empty object position. Aside from question of how to license an empty object, the verb has raised to 1°, but subject has not raised to Spec,IP, so this cannot be right either. Assuming the empty object raises to form a structure like [IP 0i [IP . , e; , ]] does nothing to alleviate this problem, since the verb and the subject are still in the wrong order, indicating that the second conjunct must have been a CP, not an IP. Hence there is an inherent contradiction and (direct) object gaps are impossible, as desired. Finally we note an apparent counterexample to the T93 approach: .

..

( 1 9)

Wenn jemandj [nach Hause kommt und da steht if someone to home comes and there stands der Gerichtsvolzieher vor der Tilr], oder (erj) [zum before the door or (he) to-the the bailif Arbeitsplatz fahrt und dort ist das Gebaude umringt workplace drives and there is the building surrounded durch die Polizei], ... by the police

While marginal with the pronoun er, it improves without it, contradicting the assumption that (asymmetrical) coordination should involve predicates of the same saturation. A possible solution is simply that the subjectjemand has ATB scrambled out of a the symmetric coordination (extraction from the first part of the asymmetrical coordination is possible, as noted above) and that the structure is in fact a coordination of IPs:

(20)

OPj-subj [IPk ej ... Vfin [ ek und [cp da Vfin OPsubj ... ]] ] oder [IPj ej . V fin [ ej und [cp da V fin DPsubj .. ]] ] .

.

.

640

Craig Thiersch

I f such an approach proves tenable, it wi ll be interesting to see, as noted in earlier in footnote 3 , if its insights can be adapted to the newer '3dimenstional' theories of coordination. References B Uring, Daniel and Katharina Hartmann.

1 998. Asymmetrische Koordination.

Linguistische Berichte 1 74: 1 72-20 1 . H5hle, Tilman. 1 983 . Subjecktliicken in Koordinationen.

Ms., U . K51n.

Hahle, Tilman. 1 990. Assumptions about asymmetric coordination in German. In

Grammar in Progress; GLOW essays for Henk van Riemsdijk,

ed . by Juan

Mascar6 and Marina Nespor, 22 1 -3 5 . Dordrecht: Foris. Kolb, Hans-Peter and Craig Thiersch. 1 99 1 . Levels and empty categories in a Principles and Param eters approach to parsing.

Derivation in the Theory o/Grammar,

Representation and

In

ed. by Hubert Haider and Klaus Netter,

25 1 -30 l . Dordrecht: Reidel. Lechner, Winfried. 2000. Handout,

Blago

Asymmetric and bivalent coordination in German.

Summer

School.

A vailable

at

http://www2.sfs.nphii.

uni-tuebingen.de/-nnsleO 1 /1 %20Intro.pdf Sag, Ivan, Gerald Gazdar, Thomas Was ow and Steven Weisler. 1 985. Coordination and how to distinguish categories . 3 (2): 1 1 7- 1 72. Th iersch, Craig. 1 993a/ 1 995.

Natural Language and Linguistic Theory

Asymmetrical coordination

as

adjunction. GLOW

workshop, Lund. Revised version at http://let.uvt.nl/generallpeople/thiersch/ pubs.htrn. Thiersch,

Craig.

1 993b.

Some

remarks

Linguistics in the Netherlands,

ed.

on

asymmetrical

by

Frank

coordination.

Drijkoningen

Hengeveld, 1 4 1 -52. Amsterdam: Benjamins. Thiersch, Craig. 1 994. On some fonnal ' properties of coordination. In

issues in mathematical linguistics,

In

and Kees

Current

C. Martfn-Vide, . 1 7 1- 1 80. Amsterdam :

North Holland: Travis, Lisa deMena. 1984 .

Parameters and effects o/word order variation.

diss. Cambridge, MA : MIT Vries, Mark de 2005 . Coordination and syntactic hierarchy. 83- 1 05.

Ph.D.

Studia Linguistica 59: '

The representation of focus and its implications: towards an alternative account of some ' intervention effects' Jean-Roger Vergnaud and Maria-Luisa Zubizarreta

1 . Two types of questions l In the literature on the semantics of questions, there has been much discussion as to whether questions are 'strongly' exhaustive or not; cf. Kartunnen ( 1 977), Groenendijk and Stokhof ( 1 984), Heim ( 1 994), Rullman ( 1 995), Beck and Rullman ( 1 999). 2 He im ( 1 994) argues that strong exhaustivity is not an inherent feature of questions. Beck and Rullman ( 1 999) concurs and provides additional evidence that fronted wh-questions have non-exhaustive readings, at least in English and in German. A fact cited is the compatibility of such questions with the ' marker of nonexhaustivity' for example:

(1)

Who for example was at the party last night?

Interestingly, there is a type of question in French which appears to resist such markers of nonexhaustivity. The facts are described below. Consider the wh-in-situ question in (2):3

(2)

La jeune artiste a danse avec qui? 'The young artist dansed with whom?'

. I We are happy to dedicate this article to our long time colleague and friend Henk van Riemsdijk. During the year 1 985-6. we were both invited to teach at Tilburg University. This period, spent working with Henk's group and amidst the Dutch linguistic community, has remained one of the happiest and most productive in our professional lives. This essay is just a small token of our gratitude to him, and to them. 2 A question is deemed strongly' exhaustive if any answer to it must construed as definitively exhaustive. 3 The term in-situ refers to the surface linear arrangement of items. The analysis in this essay is agnostic as to whether this linear arrangement directly reflects the primordial c-command arrangement or arises as a consequence of remnant movement. •

642 Jean-Roger Vergnaud and Maria-Luisa Zubizarreta

There are two versions of that question form, with two distinct prosodies: (3)

�

�

a.

Lajeune artiste a danse avec qui?

b.

La jeune artiste a danse avec qui?

� �

In the forms in (3), there is a correlation between the intonation of the wh­ phrase and that of the preceding stressed item within the same utterance (artiste in (3a), danse in (3b» : the two phrases have opposite contours, echoing each other (cf. Martin 1 975). The actual intonation of the wh­ phrase, whether falling or rising, is not crucial, however. 4 The critical prosodic difference between the two utterances in (3) is analyzed in (4): (4) a. Each utterance is a sequence of two prosodic domains PI, p] in that order, with PI subordinate to P2• b. The prosodic domains PI and P2.have different contents in (3a) and (3b).

This analysis is displayed in (5), where # is the boundary between PI and p]:

(5) a. La jeune artiste # a danse avec qui? b. La jeune artiste a danse # avec qui?

(corresponding to (3 a» (corresponding to (3b»

In each prosodic domain, the main stress is at the right boundary. The subordination of PI to P2 means that the wh-phrase has main stress within the whole utterance. The pa�ern in (Sa) is that of a: regular assertion, with main stress located at the end of the predicate phrase alld :secondary stress located at the end of the subject ;phrase. The patt�J)l ; in :.($b), is. that of a ' fronted wh-question, except for. the'surface positi6n ).of: the. ,wh-phrase and, ' possibly, its relative prominence. s In fact, the question form in (5b) appears to have all the semantic properties of a fronted wh-question. In particular, it is compatible with the 'nonexhaustivity marker' par exemple 'for example': We are using the fonns with a falling contour on the wh-word to avoid confusion with the use ofwh-in-situ questions as echo questions . 5 The subject in the question i n ( S ii) in the text carries a tertiary stress, with a falling intonation echoing the rising contour on danse. 4

The representation offocus and its implications 643

(6)

Avec qui par exemple (est-ce que) la jeune artiste a danse? 'With whom for exam p le did the young artist dance ? ' b. La jeune artiste a danse # avec qui par exemple? ' The young artist danced # with whom for example?'

a.

In that respect, the question form in (5b), and the fronted wh-questions, contrast with the question form in (Sa). The utterance in (7) is distinctively odd: 6 (7)

*La jeune artiste # a danse avec qui par exem p le? 'The young artist # danced with whom for example?'

This suggests that the question fonn in (Sa) is strongly exhaustive. There is independent evidence to that effect. The question can be naturally answered with a cleft sentence like that in (8), whereas many speakers perceive that answer as providing infonnation beyond what is requested by the fronted wh counterpart to (Sa) : (8)

C�est avec Pierre que la jeune artiste a danse 'It is with Pierre that the young artiste danced.'

We shall call the question fonn in (Sa) (=(3a» a contrastive question (c-question), for reasons to be articulated below.7 There are a number of other differences between the two types of questions just identified. For instance, a wh-word in a c-question may be conjoined with a negative polarity item as illustrated in (9), a surpr isin g property: (9)

La jeune artiste # a clanse avec qui et personne cl' autre? 'The young artist # danced with whom and noone else?'

No such combination is poss ibl e in a wh-fronted question (or its in situ transform): 6

That is, it is odd with par exemple fonning a single intonational group with the

wh-w ord . If par exemple is detached as in (i) (the comma marks an intonational

hiatus), the utterance is felicitous. In such a fonn, par exempJe modifies the whole utterance, not the wh-w ord .

0)

Marie # a danse avec qui, par exemple?

7 The property of ' strong exhaustivity' is then merely an avatar of the exclusive character of the presupposition of the question.

644

Jean-Roger Vergnaud and Maria-Luisa Zubizarreta

( 1 0)

*Avec qui et personne d'autre (est-ce que) la jeune artiste a danse? 'With whom and noone else did the young artist dance?'

(1 1 )

*Lajeune artiste a danse # avec qui et personne d'autre? 'The young artist danced # with whom and noone else?'

Another characteristic property of c-questions is the unavailability of pair­ list readings in structures where the wh-word cooccurs with a universal quantifier. We tum to a brief description of that property. 2. Pair-list readings Consider the wh-fronted question in ( 1 2) and its c-counterpart in ( 1 3):

( 1 2)

Avec qui (est-ce-que) chaque artiste a danse? 'With whom did each artist danse?'

( 1 3)

Chaque artiste # a danse avec qui? 'Each artist # danced with whom?'

The LFs of these questions display an interaction between chaque 'each' and qu 'wh.' There are two logical possibilities: chaque 'each' may have wide scope over qu 'wh' or conversely, the pair-l ist reading being that in which chaque 'each' has wider scope. Both readings are available with the fronted wh-question in ( 1 2). However, the corresponding c-question in ( 13 ) lacks the pair-list reading, as was first 'uncovered by Vergnaud and Zubizarreta (200 1 , 2002, 2003 ). In those works, we showed that the general unavailability of pair-list readings in c-questions follows from a universal principle which bars the cooccurrence in a statement of independent contrasts or negations:

( 1 4)

' A statement may contain at most one contrast or negation. 8

8

The principle is formulated in terms of the notion ' statement' , a notion itself defined in terms of the more primitive notion of point of view. For example, the sentence in (i) includes two points of view and, hence, two statements, that by the speaker, represented as [John said that}, and that attributed to John by the speaker, represented as [Mary was not/air]: (i)

John said that Mary was not fair.

The representation offocus and its implications 645

The demonstration will proceed as follows. First, we will establish the existence of two types of presuppositions and focus for utterances with explicit presuppositions, such as, e.g., the answers to the question in ( I 5):

( 1 5)

Did

.-/ �

John see Mary or Susan?

Adopting standard terminology, the two types of presupposition/focus will be called contrastive and inclusive/informational, respectively. We will then extend this result to utterances with implicit presuppositions, such as, e.g., the answers to the c-question in (Sa). Finally,we will identify the two types of questions exhibited in sections 1 and 2 with the two types of presupposition/focus: 9 ( 1 6)

A strongly exhaustive question is one with a contrastive presupposition. A question with an inclusive/infonnational presupposition is not strongly exhaustive.

In the course of our demonstration, we will generalize the analysis in Chomsky ( 1 976) to both types of focus and we will revise the standard notion of focus, as well as the standard notion of LF. We will also reduce quantifiers to logical connectives, in the spirit of Skolem' s fundamental insight. 3.

Two types of focus: contrastive vs. inclusive/informational focus.

The foundations of the theory of focus were laid out in Chomsky ( 1 97 1 ), Jackendoff ( 1 972), and Chomsky ( 1 976). Chomsky ( 1 976) shows that contrastive focus can be subsumed under quantification. This result can be generalized to the other type of focus, informational focus. We start by

9

What will not be explained is why the particular pairing between form and interpretation described in section 1 in the text obtains, and not the converse. There appears to be no language in which a wh-fronted question would have a strongly exhaustive construal while its in-situ counterpart would have an inclusive presupposition. See Vergnaud and Zubizarreta forthcoming for a proposal . Henceforth, by exhaustivity we mean 'strong' exhaustivity, unless otherwise stated.

646

Jean-Roger Vergnaud and Maria-Lliisa Zubizarreta

discussing the distinction between the two types of focus. Consider the question-answer pairs in ( 1 7)-( 1 8) : ( 1 7)

��

Did John see Mary or Susan? He saw Mary.

( 1 8)

� Did John see Mary or Susan?

He saw Mary. The question in ( 1 7) displays ' contrastive stress', with intonation rising to high pitch on Mary and then fal ling on Susan. The question in ( 1 8) is uttered with a simple rising intonation on the disjunction Mary or Susan and with higher stress on or than on did. 1O The prosodic difference between the two questions correlates with a semantic one, reflected by the answers: the answer in ( 1 7) implies that John did not see Susan, but the answer in ( 1 8) has no such logical implication and, e.g., could be continued by And also Susan. In other words, the question in ( 1 7), but not that in ( 1 8), is ( strongly) exhaustive. The semantic difference between the two question­ answer pairs in ( 1 7)-( 1 8) has obviously to do with the fact that the II connective or is interpreted differently in the two questions: ( 1 9)

The disjunction marked by or is exclusive in the question in ( 1 7), but inclusive in the question in ( 1 8).

The simplest account will assume that: 10

The relative prominence of or vs. did is a critical featu�e. If did in the ques�ion in ( 1 8) in the text has more stress thaD or, then the questiori is jnterpreted as a yes-no question. In both cases, the -question may be analyzed_; � t1ie - reduced fonn of a conju nction as shown in (i)-(ii). The same reduction analysis applies to the question in ( 1 7) in the text. In that fonn, the relative prominence of or and did does not affect the interpretation . (i)

[did he see Mary] or [riiri he see Susan]?

(ii) [did he see Mary] or [he-see Susan]? II

The logical connective or does not have scope over the question, which is not a proposition. It only has scope over the propositional subpart of th e question. See (3 1 ) in the text.

The representation olloeus and its implications 647

(20) a. The semantic difference between the two question-answer pairs in ( 1 7)-( 1 8) is reducible to the logical d istinction between exclusive and inclusive disjunction. b. An answer A(Q) to a question Q reflects the logical properties of Q. We take (20a) to be self-evident. As for the hypothesis in (20b), it has to be qualified: it will only apply to the natural answers to Q in the sense of Chomsky ( 1 97 1 ). To illustrate, consider the assertion in (2 1 ): (2 1 )

John saw Mary and Susan

l2

With neutral intonation, this assertion is a natural answer to the question in ( 1 8). With emphatic stress on and, it can be an answer to the question in ( 1 7), but with the implication that the presupposition of that question -that John could not have seen both Mary and Susan- is erroneous. In some sense, it is then a 'second-order' answer, not a 'natural ' one. The law in (20b) means that the logical structure of a question Q turns up in the Logical Form (LF) of a natural answer to Q. For example, the LF of a natural answer to the question in ( 1 7) or in ( 1 8) must include the disjunction in (22), where OR8 is defined as in (23): (22) (23) a. b.

(John Past see Mary) ORb (John Past see Susan), 0= e, i

ORe =def 'exclusive or' ORi =def 'inclusive or'

Given a question Q and a natural answer. A(Q) to Q, the part of the LF of Q that turns up in the LF of A (Q) is called the presupposition of A (Q). Thus, the LF formula in (22) with 0 = e (resp. i) is the presupposition of the answer in ( 1 7) (resp. ( 1 8» . The law in (20b) is restated in (24) in terms of that notion: (24)

12

Let Q be a question. A natural answer to Q is one whose presupposition is the same as that of Q.

This form is derived from [John saw Mary] and [John saw Susan] by Conjunction Reduction.

648 Jean-Roger Vergnaud and Maria-Luisa Zubizarreta This is the law proposed in Chomsky ( 1 97 1 ) and in lackendoff ( 1 972), where analogous notions of 'presupposition' are assumed. 13 By this account, the nonasserted subpart of the LF of a sentence is unrealized at PF: the LF of an answer in ( 1 7)-( 1 8) is the disjunction in (22), but only the first disjunct in it is pronounced. This is a significant departure from standard assumptions, according to which only copied occurrences of a structure may be elided. This essay may be viewed in part as a defense and illustration of this hypothesis. '4 The subpart of the LF of a natural answer that has a PF representation is called the focus of that answer. For example, the focus of an answer in ( 1 7)-( 1 8) is the LF disjunct in (25):

(25)

[John Past see Mary]

The above description can be generalized, but a notion of 'exclusive disjunction' different from that of standard logic is called for. The required notion, call it contrastive disjunction, is defined in (26), where ORc denotes the corresponding connective: (26) a. ORc is associative b. The formula p/ ORC: P2 ORC: . . . . ORc Pn is true just in case one and only one of the fonnulae Pi, 1 ::; i ::; n, is true. It is false otherwise. l s 13

The assertion in (i) has the same LF as the answer i n (1 7) (resp. in ( 1 8» in the text. The assertion in (i) does not count as a felicitous answer to the question in ( 1 7) (resp ( 1 8» , though, for reasons independent of the grammar (the answer is just uninfonnative). (i) John saw Mary ORc (resp. ORi) John saw Susan .

The fact that the utterance in (2 1) in the text is a natural answer to the question in ( 1 8), but not to that in ( 1 7), can be accommodated by recognizing the connective and as a ' subpart' of ORi. A plausible framework for this treatment of and in relation to OR' is the model theory of sentential reasoning developed by Johnson­ Laird and his collaborators. See Vergnaud and Zubizarreta forthcoming, Johnson­ Laird et a1. 2000. 14 Note that one will then have to distinguish a silent PF from the absence of PF altogether. 15 The standard exclusive or is also associative, but does not have the property in (26b) in the text. For example, the exclusive disjunction in (i) is true if the disjuncts A, B, C are all true: A ORc B ORc C (i)

The representation offocus and its implications 649

Then, an assertion A is defined as in (27), w ith the associated notions of 'presupposition' and 'focus' given in (28): (27) a. A =def ( A., $, 1t) A. is a maximal disjunction of the fonn in (b): ·16 p � 0 b I\. = P I ORo p 2 OR '\ . . . . ORo n, U - c, I $ is the specification of a subpart of A.: c. $ = Ai I ORo A i2 ORo . . . . OR o Aip, I ::; ik ::;n, 1::; k �p (Pik = Aik), 1 ::; k ::;p n is the PF representation of $: c. n = fAil or AiZ or . . . . or AipfPF _

•

(28)

The presupposition and the focus of A in (27) are the fonns A and �, respectively. 1 7

The presupposition/focus (N<1» structure (A, $) in (27) constitutes the LF representation of the assertion (A, �, n), whose PF representation is n. The item '=' in the fonnula (Pik = Aik) in (27c) is the specificational verb be in the sense of Higgins ( 1 973). The relation of Aik to P ik is the same as that of, e.g., [John saw Mary] to [John saw someone] in (29a) or to the formula [John saw x] in (29b):

The formula PI Oft P2 OK . . . . ORc P" in (26b) in the text is defined recursively as follows: (ii) a. Let qJn 'Pn{Pt, . . . Ph, . . . . Pn) stand for P I ORc Pz ORc . . . . ORc Pn' b. 'Pn {qJn-1 AND (NOT Pn» ORi (Pn AND (NOT (P I ORi . . . ORi PlI- 1 » ) = =

=

{qJn- 1 ORc PJ

The fonnula !Pm OR" If{ is then defmed as in (iii): (iii)

qJm ORc qJn

=

qJm+n

Associativity follows: (iv)

{('PI ORc qJm) ORc 'Pn}

=

(qJl ORc ('Pm ORc 'Pn»

We see that the definitions of OK and of standard exclusive or coincide for n=2, but differ in all other cases. 6 6 ORO P" may not include disjunctions of 1 6 A disjunct Pi in the LF PI OR P2 OR the other type. See section 5. 11 This is in essence the definition in Zubizarreta 1 998. .•. .

650 Jean-Roger Vergnaud and Maria-Luisa Zubizarreta (29) a. [John saw someone] : [John saw Mary] b. [John saw x] : [John saw Mary] To illustrate, the following structure is an assertion:

(30) a. LF Presupposition (John Past see someone) OR i (John Past see a movie) b. Focus (John Past see Mary) c. PF IJohn Past see Mary/pF =

A characteristic feature of such assertions as the answers in ( I 7)-( 1 8) is that its focus is merely a subpart of its presupposition: Aik is not distinct from Pik in such an answer, because the disjunct Pik only includes definite descriptions. This is not the most general situation, though, as we shall see in section 4. On the other hand, the focus of a question in general coincides with its presupposition: it is the whole logical structure of the question . 1 8 T o illustrate, the focus of the question in ( 1 7) is the disjunction in (22) with O=C.19 The LF representation of a question will also include an element which conveys the question force, identifying the utterance as a · question. The LF of a question in ( 1 7)-( 1 8) then is as in (3 1 ), where e is the question force item:20

(3 1 )

[8, «John Past see Mary)

OR�

(Joh n Past see Susan» ] , 0 = c, i

18

For example, it seems that the question utterance in (i) could not have the presupposition in (ii). The question is why it should be so. See Vergnaud and Zubizarreta (forthcoming). (i) ( i i)

Did John see Mary or Susan? . (John Past see Mary) OR� (John Past see S usan) ORB (John Past see Joan)

19 Recall that the definitions of ORc and of standard exclusive or coincide for a fonnula like that in ( 1 7) in the text, which has only two elementary propositions. 20 It is necessary to distinguish between the force of a question and its logical structure because of examples like that in (i ) . The possibility of a pair-list reading in (i) shows that the question quantifier must have narrow scope with respect to every, while the question force obviously has scope over the whole structure.

(i)

What did everyone read?

The representation offocus and its implications 65 1

Since there are two types of disjunctions, contrastive and. inclusive, there are correspondingly two types of presuppositions and two types of focus: (32)

Let A be some assertion. If the presupposition ofA is a contrastive disjunction, the presupposition/focus of A is called contrastive. If the presupposition of A is an inclusive disjunction, the presupposition/focus of A is called inclusive/informational.

The induced categorization appears correct. To illustrate, consider the answers in ( 1 7)-( 1 8) again. Their common focus is the disjunct in (25). However, that focus is contrastive in the case of the answer in ( 1 7), but inclusive/infonnational in the case of the answer in ( 1 8). As indicated in Chomsky ( 1 97 1 ), Jackendoff ( 1 972), and Rooth ( 1 985), for example, the universe of assertions is never Hobsonian:2 1 (33 )

A declarative sentence manifests a choice among explicit or implicit alternatives.

This can be stated more precisely in terms of the notions of 'focus' and 'presupposition' introduced above: (34)

The focus of an assertion is a strict subpart of its presupposition.

This in tum implies that the presupposition of an assertion always contains more than one disjunct: (3 5)

The presupposition of an assertion is never degenerate.

Given that a nondegenerate presupposition can always be expressed as a question, the law (34) can be rephrased as in (36) (see Zubizarreta, 1 998, pA): (36)

Every assertion is an answer to some actual or virtual question.

21

Actually, in Rooth 1 985 and, more generally, in the semantic literature on focus, alternatives are defmed with respect to the statement expressed by the sentence. We depart from that view for beth empirical and conceptual reasons. See Vergnaud and Zubizarreta forthcoming. We note that the notion of used in Rooth

1 985

alternative assertion

to analyze 'contrastive focus' . As argued in Zubizarreta

it can be extended to all types of focus, the approach adopted in this essay.

is

1998,

652

Jean-Roger Vergnaud and Maria-Luisa Zubizarreta

The notions above have been developed on the basis of examples with explicit presuppositions, such as the utterances in ( 1 7)-( I S). The question arises whether these notions can be generalized to utterances with implicit presuppositions, such as in the question-answer pair in (37). In the next section, we tum to a discussion of such systems. (37) 4.

Who did John see? John saw Mary

Utterances with implicit presuppositions

In every disjunction/conjunction, there is a minimal constituent that exhaustively dominates all the material which varies from disjunct/conjunct to disjunct/conjunct. To illustrate, that minimal constituent is TP in the disjunction in (3Sa), vP in the disjunction in (3 8b), and the DP object in the disjunction in (3 8c) (the presupposition of some utterance in ( 1 7)-( 1 8» : (3 8) a. (John Past see Mary) ORli (peter Past call Susan) b. (John Past see Mary) ORs (John Past call Susan) c . (John Past see Mary) ORli (John Past see Susan) In general, Conjunction Reduction can apply to such complex fonnulae, factoring out the material which is constant across all disjuncts/conjuncts so that only the minimal variable constituent in the sense above remains within the scope of the connective. This is illustrated in (39) with the disjunction in (3 8b) (denoted by A): (3 9)

A ==

<(John Past do something), (something (call Mary»}>

=

«see Mary) OR/)

' The pair of linked formula� on the . right-hand side · 9f the equivalence · in (39) is called the reduced form / of A and is denoted . by, ,RED(A)?2 For ' example, the q uestions in ( 1 7)-( 1 8) and the answer in (2 1 ) are reduced utterances in the sense above (cf. notes 1 0 and 1 2). The following notion will be useful: (40)

22

The second component of RED(A) is called the nucleus of A.

The correspondence RED(A)� A is QR.

The representation offocus and its implications

653

To illustrate, consider the answer in ( 1 8). The reduced form of its presupposition is the pair of linked propositions in (4 1 ): (4 1 )

i <(John Past see someone), (someone = (Mary OR Susan» >

The corresponding nucleus is the formula in (42): (42)

(someone = (Mary ORi Susan»

By definition, a degenerate expression -one with a single disj unct/conjunct­ has no reduced form or nucleus. For example, the focus of the answer in ( 1 8) cannot be directly mapped onto a reduced form. But it can be assigned a reduced form and a nucleus derivatively, by the rule in (43): (43)

Let A=(A., $, 1t) be an assertion with presupposition/focus structure (A., .p). The nucleus of t/J is defined as the restriction to � of the nucleus of A.

Since a presupposition is never degenerate (see (33)-(3 5) in section 3), the notions above are always well defined. In particular, the focus of the answer in ( 1 8) has a reduced form and a nucleus, the pair in (44) and the formula in (45), respectively: (44) (45 )

<(John Past see someone), (someone (someone = Mary)

=

Mary»

One notes that the LF fragment in (44) also underlies the sentence in (46): (46)

Someone that John saw was Mary . .

The difference between the above sentence and that in ( 1 8) derives from the different PF manifestations of ' = ' , un pronounced in ( 1 8), surfacing as be in (46). In both cases, though, there are at least two occurrences of someone forming a chain. Consider now the question Who did John see? in (37). It is natural to assume that the presupposition of that question, equal to its focus, is the disjunction in (47), where the value of 0 (c or i) is left open at this point: (47)

(John Past see someone) ORa (John Past see someone else), 0

=

c, i

654 Jean-Roger Vergnaud and Maria-Luisa Zubizarreta

We postulate that Conjunction Reduction applies obligatorily at LF. Then, the LF of the question will also include the reduced form in (48): (48)

« John Past see someone), (someone = (someone ORo someone else»

Crucially, the nucleus in (48) is a recursive fonnula, an instance of the identity in (49): (49)

U

=

U ORo W

The nucleus in (48) then stands for an infinite fam ily of identities of the 2 form in (50), where n can be any number greater than 0: 3 (50)

(someone = (someone (ORO someone elser»

Such unbounded formulae correspond to quantifiers. The general hypothesis is stated in (5 1 ):24 (5 1 )

A Human Language (HL) quantifier is a recursive identity of the form in (i), where . is a logical connective and D a determiner: (i) « D some one) = «D some one) • (D some other» )

The determiner D in (5 1 ) may be empty or not. If it is, the equation in (5 1 ) gives rise to that in (50), with other realized as else. The combination (D=0, . =ORO) in (5 1 ) defines the quantifier any in English, which we assume is a component of every wh-word:2S

23 The identity follows from that in the nucleus in (48) in the text and from the associativity of the logical operations (disjunction and conjunction). Recall that both OR; and ORr: are associative. 24 We are assuming that some is not a quantifier, which goes against much standard treatments. However, it is consistent with the treatement in, e.g., Heim 1 982 or within dynamic semantics. 25 Thus, any is assumed to be ambiguous, having either an ' inclusive' or a ' contrast ive' interpretation. See section 5 and Vergnaud and Zubizarreta forthcoming. The feature wh (qu in French) may just be the overt manifestation of the iterated connective embedded in the quantifier, which must be pronounced at the edge of a phase.

The representation offocus and its implications 655

(52)

LF(any) =def {someone = (someone OR5 someone else»

The assertion John saw Mary, with the presupposition/focus (A/ structure in (53) is as in (54): (54) a. Presupposition <(John Past see someone), {someone = (someone OR!) someone else» b. Focus « John Past see someone), (someone = Mary» Thus, an account of utterances with implicit presuppositions is available within the proposed framework. One question is whether this account can be extended to other types of quantifiers, e.g., to every and to each in English. Consider the sentence in (5 5): (55)

Everyone saw Mary.

We are trying to determine the LF representation of the quantifier every, as separate from the N structure of the utterance. A natural hypothesis is that every is the conjunctive counterpart to the disjunctive any in (52): (56)

LF(every) =def (someone = (someone AND someone else»

Abstracting away from the N structure, the LF representation of the sentence in (5 5) is then analyzed as including the reduced component in (57), as well as its expansion in (58): (57)

<(someone Past see Mary), (someone = (someone AND someone else»

(58)

(someone Past see Mary) AND (someone else Past see Mary)

656 Jean-Roger Vergnaud and Maria-Luisa Zubizarreta That description assumes that [plural] is not a semantic feature, but an LF uninterpretable · morphosyntactic feature. Semantic 'plurality' is taken to arise from the adjunction of a mass classy/er to an inherent count classifier, a process independent of quantification. 6 Similarly, the English quantifier each is analyzed as in (59): (59)

LF(each) =def {the one

=

(the one AND the other one»

Accordingly, and abstracting away again from the AJ(fJ structure, the LF representation of the sentence in (60) is analyzed as including the reduced component in (6 I a), as well as its expansion in (6 I b). (60)

Each one saw Mary.

(6 1 ) a. <(the one Past see Mary), {the one (the one AND the other one» b. (the one Past see Mary) AND (the other one Past see Mary) =

One notes that the trace of the quantifier each in the left member of the reduced form in (6 1 a) is the definite determiner the, not someone as in the case of a quantification by every or by any. The reason is that the definiteness of the arguments of AND in (59) and (6 1 a) is a separate property from the quantificational force carried by AND. If one adopts a Russellian theory of definiteness, it will be of course necessary to compound the each quantification in (6 1 a) with the the quantification; see Vergnaud and Zubizarreta forthcoming. We now return to the formalization of the exhaustivity property and to its relevance for an account of the absence of pair-list readings in the contrastive French wh-in-situ questions. 5.

Questions and pair-list readings revisited

It appears that the salient interpretive properties of exhaustive and nonexhaustive questions can be derived from the identification in (62) (see ( 1 6) in section 2):

26 Following Chomsky ( 1 974), we analyze English all as an intensifier of the (semantically) plural version of the, which may cooccur or not with AND quantification.

The representation offocus and its implications 657

(62) a. exhaustive question <::!> contrastive presupposition b. nonexhaustive question <=> inclusive/infonnational presupposition The identification is immediate in the case of questions with explicit presuppositions (see section 3). It was shown in section 4 that the case of questions with implicit presuppositions reduces to that of questions with explicit presuppositions. Consider the fronted wh-question in (63 ) and its c­ counterpart in (64): (63 )

Avec qui (est-ce-que) Marie a danse? (With whom did Marie dance?'

(64)

Marie # a danse avec qui? ' Marie # danced with whom?'

The presupposition of the question in (63) is the disjunction in (65), while the disjunction in (66) is the presupposition of the question in (64) (modulo translation): 27 (65)

(corresponding to (63» (Marie Past dance with someone) ORi (Marie Past dance with someone else)

(66)

(corresponding to (64» (Marie Past dance with someone) ORc (Marie Past dance with someone else)

Consider now the questions with universal quantification in (67) and in (68): (67 )

Avec qui est-ce-que chaque artiste a danse? ' With whom did each artist danse?'

(68)

Chaque artiste # a danse avec qui? 'Each artist # danced with whom?'

The LF of each question displays an interaction between the quantifications defined by chaque and by the presupposition. Specifically, there are two 27 The quantificational content of the wh word in the forms in (63) and in (64) in the text arises from the application of Conjunction Reducti on to the corresponding presuppositions in (65) and (66), respectively. -

658 Jean-Roger Vergnaud and Maria-Luisa Zubizarreta

possible logical structures, depending on whether AND has wide scope over OR" or conversely. The two possible presuppositions are given in (70) and (7 1 ) (modulo translation), with the notation defined in (69): (69) a. b. c. d. e.

[X, Y] =dcf [X Past dance with Y] 0 def the artist 0' = def the other artist SO =def someone SO' =dcf someone else

(70) (7 1 )

([0, SO] ORo [D, SO']) AND ([D', SO] ORS [0', SO']), 0 = c, i ([D, SO] AND [D', SO]) ORS ([D, SO'] AND [D', SO '] ), 0 = c, i

=

The logical structure in (70) describes the pair-list reading, while that in (7 1 ) describes the single answer reading. Both readings are available in the case of the wh fronted question in (67), but the pair-list reading is absent in the case of the corresponding c-question. In past work, we have argued that this gap was not an intervention effect, but rather the consequence of a general principle barring multiple contrasts in a statement (see ( 1 4) in section 2): (72)

A statement may contain at most one contrast or negation.

The principle in (72) effectively excludes the pair-list construal of the c­ question in (68), since the LF structure of that construal would be a conjunction of independent contrastive disjunctions, as shown in (73) (with the notation defined in (69» : (73)

([0, SO]

c OR

[D, SO']) AND ([D', SO]

c OR

[D', SO'))

The principle in (72) implies that contrastive foci are obligatorily absorbed in a statement. The same principle presumably accounts for the unfelicitousness of double negation. The principle could be formulated as in (7 4):28

28

Clearly, the principle in (72) in the text must be governed by some 'locality' condition within the hierarchy of statements; see note 8. For example, the sentence in (i) is both felicitous and easily interpretable. Similarly, the sentence in (ii) is acceptable with contrastive or (the pair-list reading remaining unavailable).

The representation offocus and its implications 659

(74) 6.

A connective ORc must have widest scope within the minimal statement containing it.

Conclusion

In thi s essay, we have argued that the absence of pair-l ist readings in strongly exhaustive question, which would appear to come under an account in terms of intervention effects, actually has a completely different explanation. When the theory of focus is properly revised, this phenomenon reduces to the universal principle in (72). This principle accounts for the so­ called intervention effect discussed in Boeckx ( 1 999, 2000), Vergnaud and Zubizarreta (200 1 , 2002, 2003). References

Beck, Sigrid, and Hotze Rullmann. 1 999. A flexible approach to exhaustivity in questions. Natural Language Semantics 7 :249-98. Boeckx, Cedric. 1 999. Some in-situ. some in Spec CPo Ms., University of Connecticut, Storrs. Boeckx, Cedric. 2000. Properties of French Interrogatives. Ms., University of · Connecticut, Storrs. Chierchia, Gennaro. 1 993. Questions with Quantifiers. Natural Language Semantics 1 : 1 6 1 -234.

Chomsky, N oam. 1 97 1 . Deep structure, surface structure, and semantic interpretation. In Semantics: An interdisciplinary reader in philosophy, linguistics and psychology, ed. by D. Steinberg and L. Jakobovits. Cambridge: Cambridge University Press. Chomsky, Noam. 1 974. [Amherst Lectures]. Chomsky, Noam. 1 976. Conditions on rules of grammar. Linguistic Analysis 2: 303-35 1 .

Groenendijk, Jeroen, and Martin Stokhof. 1 984. Studies on the Semantics of Questions and the Pragmatics ofAnswers. Doctoral dissertation. University of Amsterdam, Amsterdam. Heim, Irene. 1 982. The Semantics of Definite and Indefinite Noun Phrases. Doctoral dissertation. U . Mass., Amherst. Heim, Irene. 1994. Interrogative semantics and Kartnmen's semantics for 'know'. In Proceedings of the Israeli Association for Theoretical Linguistics 1 , ed. by Rhonna Buchalla and Anita M ittwoch. Jerusalem: Akademon.

(i) (ii)

John did not say that Mary was not fair. Each judge argued that the motion should be rejected or ignored.

660 Jean-Roger Vergnaud and Maria-Luisa Zubizarreta Higgins� Francis R. 1 973 . The pseudo-cleft construction in English. Doctoral dissertation, MIT. Jackendoft: Ray. 1 972. Semantic Interpretation in Generative Grammar. Cambridge, Massachusetts: MIT Press. Johnson-Laird, Philip N., Paolo Legrenzi, Vittorio Girotto, and Maria S. Legrenzi. 2000. Illusions in Reasoning about Consistency. Science 288: 53 1 -532. Kartunnen, Lauri . 1 977. Syntax and Semantics of Questions. Linguistics and

Philosophy 1 : 3-44. Rooth, Matt. 1 985. Association with focus, Doctoral dissertation, Department of Linguistics, University of Massachusetts, Amherst. Rullman, Hotze. 1 995. Maximality in the Semantics of Wh-Constructions, Doctoral dissertation, University of Massachusetts, Amherst. Vergnaud, Jean-Roger, and Maria Luisa Zubizarreta. 200 1 . The Logical Structure of Informational and Constrastive Questions. Paper presented at Universite Paris 8 (June 200 1 ), Instituto Ortega & Gasset (May 2002), Univeristy of Illinois at Urbana-Campagne (October 2002). Abstract published in the 2003 Glow Newsletter. Zubizarreta, Maria Luisa. 1 998. Prosody, Focus, and Word Order. Cambridge Mass: MIT Press. Zubizarreta, Maria Luisa. 200 I . In terven tion effects in the French wh-in-situ constuction: syntax or interpretation? LSRL XXXllUniversity of Illinois at Chicago/April 1 9-22, 200 1 . Published in A Romance Perspective in Language

Knowledge and Use. Selected Papers from the 3 1st Linguistic Symposium on Romance Languages, ed. by Rafael Nunez-Cedeno, Luis L6pez and Richard Camero. 2003 . Amsterdam/Philadelphia: John Benjamins. Zubizarreta, ML. 200 1 . Intervention effects in the French wh-in-situ constuction: syntax or interpretation? LSRL XXXIlUniversity of Ill i nois at Chicago/April 1 9-22, 200 t . Published in A Romance Perspective in Language Know/edge

and Use. Selected Papers from the 31st Linguistic Symposium on Romance Languages, ed. by Rafael Nunez-Cedeno, Luis L6pez and Richard Camero.

2003 . Amsterdam/Philadelphia: John Benjamins.

; '

. " ,

, . . : ", � '. . ' "

Circumstantial evidence for Dative Shift Edwin Williams

Bruening (200 1 ) shows that the scope fixity found in double object constructions is not absolute, but relative: although the scopes of the two objects can go wider than the VP that contains them, they cannot exchange ) places. Example ( 1 b) illustrates scope fixity; among his arguments that it is relative, not absolute, Bruening's gives ( Ic), where the second object scopes over the subject even though in ( 1 b) it cannot scope over the first object: (ambiguous) ( 1 ) a. The teacher gave a different book to every student. b. The teacher gave a different student every book. (*every > a different) c. A different teacher gave me every book. The relative nature of scope fixity would seem to suggest that the relevant limitation was not a locality constraint, but Bruening nevertheless derives it from Shortest Move; in the sense if Richards ( 1 998), in conjunction with an idea about the structure of double object constructions, and of applicative constructions in general, adopted from Marantz ( 1 993), who proposes that constructions like the double object construction have the following structure: (2)

[vp v [VPl NPgoal APPLICATIVE [VP2 give NPtheme]]

( l b), but not ( l a), has an extra layer of structure, headed by the APPLICATIVE morpheme (null in English) which introduces the Goal argument. Because the Theme and Goal are arguments of different heads in ( 1 b), they necessarily do not c-command each other; whereas in the prepositional variant ( 1 a), Bruening proposes that they do c-command each other, by virtue of being dominated by a node marked "?" (p. 264). I *Thanks to Dalina Kalulli, who organized a workshop on Argument Structure in

Vienna, June 2005, at which I first presented this material; and to Henk van Riemsdijk, who in converstation that July opened a new line for the investigation to take, which I will report in subsequent work.

662 Edwin Williams

(3)

I gave [? [a different book] [to each student] ]

From this difference in structure arises the difference in behavior in this way: if v bears a scope-attracting feature P, then in the case where the two arguments do not c-command each other they must be attracted to it under Richards' (200 1 ) 'tucking in' locality regime, which wiJ] preserve their order; but where they c-command each other, as in (3), neither is favored over the other, so that either scope order is possible. I will present two kinds of evidence that extra Applicative structure + locality is not the answer to fixity, and will sketch an account that is based a property of the derivation. The first consideration that it is not the extra structure associated with the applicative morpheme that is making the difference comes from the following:

(4) a. &1 made [a different child [solve every puzzle]] b. &1 persuaded a different child [PRO to solve every puzzle] c. &1 promised a different child [PRO to solve every puzzle] In each of these, we have overt extra structure which separates the two relevant arguments a different child and every puzzle, yet ambiguity holds in all of them. Bruening has a means, which we have not discussed yet, for handling the first case involving the VP internal subject hypothesis­ essentially, an object can get object over the subject of the same verb if it is attracted to the v that that subject is the specifier of--but that account will not extend to (b) or (c). So neither extra structure nor failure of c-command can in general be associated with scope fixity. A more powerful reason for doubting that the extra structure is the answer derives from a consideration of the distribution of the constructions showing scope fixity. Below is a sample: (5) a. double object V NP NPN NP PP b. spray NP with NP / spray NP on NP I *1 sprayed a different truck with each paint (Bruening) c. drape X with YI drape Y over X (examples in Bruening) d. dress X in Y / put Y on X 1 *1 dressed a different baby in each costume 2 I put a different costume on each baby

Circumstantial evidencejor Dative Shift 663

e. transform X into y / make Y from X 1 *1 transformed a different ingot into each statue 2 1 made a different statue from each in got. 3 I transformed the 2 silver ingots into every piece of servingware that we will need. f. steal X from Y / rob Y of X 1 *1 robbed a different person of each thing 2 1 stole a different thing from each person g. drive X out of Y/ rid Y of X 1 *1 rid a different European country of each kind of xenophobia 2 I drove a different kind of xenophobia out of each European country h . grow X into Y / grow Y from X 1 *1 grew a different acorn into each tree 2 1 grew a different tree from each acorn (Jackendoff 1 972) In each of (5a-h), there are two constructions, the first of which shows scope fixity and the second of which does not. The pairings are clearly of related uses of the same predicate or predicate type. In some of the cases, the verb is the same; in some of the other cases, the verbs are different in the two members, hut the second is clearly a paraphrase of the first; for example, steal/rob, dress/put on. The applicative morpheme, applicable only to the first of each pair, is supposed to introduce a layer of structure in introducing its argument. But the pairings shown in (7) raise strong doubts about this: if the applicative introduces new structure and a new argument, then why is there always an alternative without the applicative morpheme and its structure? This is surely not expected. This observation, combined with the earlier demonstration that extra structure does not in general give rise to scope fixity, compels us to consider an account which addresses the pairings directly. Bruening bases a further argument for the extra-structure/locality . account of the scopal behavior of double object constructions. In Bruening's theory, object can scope over subject through interaction of reconstruction and attraction to P: (6)

[v+p tsubj [ V+p . , QDO] ] .

664 Edwin Williams

The quantified direct object is attracted to Spec v, where it scopes over tsubj; if the subject reconstructs we get 0>8, if it does not, S>O. This explanation is flawed in that the relation between attraction to P and reconstruction must be regulated in such a way as to prevent reconstructed subjects from being attracted to P, for if they are, they wi)) always have scope over objects. The regulation can take the fonn of ordering (covert attraction precedes reconstruction) or some other means, but it does not follow from anything othelWise present. Furthermore, even with that regulation in place, a problem arises with double object scope and subjects: the attraction-reconstruction theory predicts that either both objects will have scope over the subject (P is present, and reconstruction takes place), or neither object has scope over the subject (either P is absent, or reconstruction does not take place: (7) a. [vp !subj [ v+p Q l Q2]] b. [vp Q I Q2 [vp !subject t l t2 ] C. Q 1 Q2 Subject d. Subject Q 1 Q2 e. Q 1 Subject Q2 ...

•••

This al10ws for readings (c-d), but not (e); but examples that interpretation:

(8)

like (8)

do have

A different person gave every girl every assignment (All girls [ a different person [ every assignment [ ]]]) . . . .

Furthennore, there are cases of frozen S-O scope obviously akin to the double object cases; but the attraction/reconstruction mechanism would unfreeze them : (9) a. &A different tree grew from every acorn .. b. -&A different tree grew into every flower If (9b) were to have the structure in (6), with the aforementioned provisions governing it, it should allow every to scope over a different, which is impossible. Bruening's further argument for the extra-structure/locality theory assumes the reconstruction/attraction theory, and is based on the finding that the following is ambiguous:

Circumstantial evidencefor Dative Shift

( 1 0)

665

&A different girl was given every book.

The argument is that passive enables wide scope of every, which is not possible in the corresponding active. The reconstruction/attraction theory predicts that it is ambiguous, but does so with the same problem already mentioned: the interaction of reconstruction and attraction must be stipulated. Further, the account depends on the passivized subject moving through Spec v, certainly a possibility worth considering, but not standard. ( 1 0) is indeed better than the corresponding active, but this is a narrow fact; if we look at the other cases, we find uniform ungrammaticaIity: ( 1 1 ) a. * A different metal ingot was transformed into every statue b. *A different country was rid of every kind of xenophobia Furthermore, in the progressive, the wide scope reading of ( 1 0) degrades essentially to ungrammaticality: ( 1 2)

I I

t

I1

I

I

I

*On that day, a different girl was being given/handed/assigned every book.

I conclude then that there is something special about ( 1 0) itself; perhaps it is that it can be read as "a different girl got every book", which is itself ambiguous. So it would appear that scope fixity does not reduce to a locality condition operating in the context of trees with an extra layer of Applicative structure, but then where is the answer to fixity? I think we may construct another kind of answer if begin with the peculiar fact shown in (5)--every freezing construction is paired with a nonfreezing one. The obvious conclusion is that the pairs are related to each other, and that there is something about the relation which is responsible for scope fixity. A scheme in which one of the pair is derived from the other promises to solve both problems at once--the pairing is directly accounted for, and the fixity could be attributed to the derivation. That is fact is the traditional 'Dative Shift' analysis of the double object construction. But there are obstacles to extending literal Dative Shift to all of the examples in (5). Dative Shift is feasible for the cases for which it was originally proposed because the preposition present in one of the two related constructions but absent in the other can be thought of as somehow absorbed by the relation. However, for other constructions in (5), there are prepositions for both the fixed and unfixed construction, and they are

666 Edwin Williams

different. And worse, for some of the cases (d-g) there is not a single verb that occurs in both the fixed and the unfixed construction, so the notion of deriving one from the other becomes problematic. The desirable features of the Dative Shift analysis are first the pairing it establishes, and second, the asymmetry it introduces. We can get these benefits from a slightly different scheme. Rather than deriving one from the other, we derive them both from a common source; thus giving the pairing. If it turns out that one, but not the other of the pair has a more direct relation to the common source, we derive the asymmetry on which we would hope to build an account of the difference in scope fixity. What the two members of each pair have in common is a couple of theta roles--Theme and Goal in the double object case, and some other roles in some of the other cases. The members of each pair differ in how those theta roles are realized; in particular, how the theta roles are associated with case positions. Representation Theory as developed in Williams ( 1 998, 2004) supposes that grammatical representation is a series of representations beginning with a representation of Thematic relations, followed by a representation of Case relations, and so forth. In the simplest case, the representation of Case relations is isomorphic to the representation of thematic relations, and the map of to one to the other is transparent:

( 1 3)

[v Theme Goal]

-+

I

1

[v NPacc [to NPdat]]

t

Here, the different parts of the two structllres align in an obvious way. But in the double-object construction, the relation is 'twisted', or 'shifted ' :

!

( 1 4) [V Theme Go al] -+

I

[v NPd�t NPacJ .

j

The representations of the particular sentences in ( 1 ) m ight look like: ( 1 5) a. give [book]theme [girl] goal ] -+ [give [a different book] acc [to [every girl]] give b. [boo k]theme [girJ ] goaI] -+ [give [every girI]acc [a different book]acc

Circumstantial evidencefor Dative Shift 667

The second one is obviously 'twisted' or 'sh ifted' compared to the first. We may treat all of the pairs in (5) this way. So, for example, rob and steal will have the same Thematic Structure, but they will be mapped onto different Case Structures; steal will be straight, but rob will be twisted. Verbs clearly can control what Case Structures their Thematic Structures map to. The applicative morpheme, in languages where it is present, is not adding an argument, it is marking the misalignment between one level and another. Now we have a sketch of a scheme which pairs the constructions correctly, and establishes an asymmetry across them: one member of each pair will be a 'straight' representation of the Thematic Structure, and the other will he 'twisted' . So the task that remains is to build an account of fixity on top of this. An attractive possibility would be to say that once two NPs have exchanged place, as they have in the twisted cases, they cannot exchange places again; in other words, *AB-4BA-4AB. This will certainly work for the cases at hand: the first part of the map would be the map from Theta Structure to Case Structure as illustrated in ( 1 3- 1 5); the second part of the map would be QR itself, and the constraint would bar QR reordering two NPs that had already been reordered by our version of Dative shift. But such a constraint would be too broad. It would for example bar the WH movement of the second NP in any of the shifted cases, and this is wrong: ( 1 6) a. With what did you cover the chairs? b. What did you transform the ingots into? c. What did you rob John of? Rather, the scope of the constraint must be sensitive to the kinds of mappings involved. But why should the Theta-Case map interact in this way with Quantifier Scope Assignment? Widespread work on the interpretive content of the direct object relation has established that it is implicated in the 'quantity' interpretation of the argument it expresses. So, for example, Anderson's ( 1 970) examples in ( 1 7) show how the relation of the "amounts of apple" eaten to the event described is regulated by the assignment of accusative case, or lack of it. ( 1 7) a. John ate the apple b. John ate at the apple

668 Edwin Williams In subsequent work on this topic I will explore the possibility that that is the reason that the map to Case Structure interacts with QR in the observed way. References Anderson, S. 1 970. A Little Light on the Role of Deep Structure in Semantic Interpretation. National Science Foundation Report 26: 11. 1-11. 1 3 . Bruening, Benjamin. 200 I . QR Obeys Superiority: Frozen Scope and ACD.

Linguistic Inquiry 32, 233-3 73.

Jackendoff, Ray. 1 972 . Semantic interpretation in generative grammar. Cambridge (Mass.)/London: MIT Press. Marantz, Alec. 1 993 . Implications of asymmetries in double object constructions. Theoretical aspects of Bantu grammar, ed. Sam A. Mchombo, 1 13-1 50. Stanford, Calif : CSLI Publications. [Distributed by Cambridge University Press.] Richards. Norval. 1 997. What moves where when in which language? PhD dissertation, MIT, Cambridge, Mass. Williams, Edwin. 1 998. Economy as Shape Conservation. MIT Press 's N. Chomsky 's 70th Birthday Celebration Web Page, Jay Keyser and Kenneth Hale, eds. Williams, Edwin. 2004. Representation Theory. Cambridge (Mass.): MIT Press. Williams Edwin. 2005. Double-Object Scope Fixity-Locality or Entanglement? Presentation at Argument Structure Workshop, D. Kallulli organizer, University of Vienna, June 2005. Williams Edwin. In preparation. Scope Entanglement.

Why should diminutives count? Martina Wiltschko

1. Introduction

Years ago, when I had the pleasure to be a student of Henk' s in Vienna, he made us sqUib. We were terrified but Henk commented with a smile: "Every linguist should always have at least 1 5 squibs at hand - so this should be easy". He was right about the should part - but it wasn't easy. I'm grateful to the editors of this volume for giving me the opportunity to squib for Henk again. I n Dutch, diminutive suffixes tum mass nouns into count nouns, an observation attributed to Henk van Riemsdijk in Borer (2004, Ch .4, Fn.6). Table J veel zout much salt 'much salt' veel brood much bread 'much bread'

*veel zout-je much saltDIM * veeI brood-je much breadDIM

veel zouten many salts 'many kinds of salts' veel brood-en many bread-PL 'many kinds of breads'

veel zoutjes many saltDIM_PL 'many salt crackers' veel brood-je-s many breadDIMpl 'many rolls'

The same is true in German where diminutive marking on mass nouns is very productive. Al l dim inutive suffixes found in German tum mass nouns into count nouns (some are lexicalized with a special meaning, but some are ful ly compositional). Here I wil l discuss -chen, and -lein from Standard German and -er//-/ from Austrian German: ( 1 ) a. v iel Brot m uch bread 'much bread' b. vie I Wein much wine ' much wine'

a'. b' .

viele Brot-chen/BrOt-leinlBrot-erl manYpl bread oIM 'many little rolls/sandwiches' viele Wein-chenlWein-leinlWeind-eri manYpl weinolM 'many l ittle (good) wines'

670 Martina Wiltschko

Interestingly, the individuating function of the diminutive is not an idiosyncratic property of Dutch and German but occurs in a number of diverse languages (Jurafsky 1 996:555). Table 2

YIDDISH

OJIBWA

EWE

BAULE

CANTONESE

der zamd goon sukli ajwe toni]

'sand' 'snow' 'sugar' 'rice' 'sugar'

dos zemdl goonens sukli-vi ajweba tong35

' grain of sand' 'snowflake' 'piece of sugar' 'rice kernel' ' piece of candy'

I n light of this generalization, we are lead to conclude that diminutive affixes function as classifiers. However, as reported in the same footnote of Borer (2004), Van Riemsdijk points out that we would expect diminutives to be in complementary distribution with classifiers (on the assumption that classifiers and number marking occupy the same functional head; cf. Borer 2004, among others). However, this is not the case as indicated by the data in Table 1 . The purpose of this squib is to defend the classifier analysis of diminutives. I start by briefly considering two approaches which do not work. S uppose that diminutive affixes are associated with a selectional restriction that requires them to combine with count nouns. When combined with a mass noun, the diminutive will treat the mass noun as if it were a count noun. In other words, the diminutive suffix coerces a mass noun into a count noun just l ike plural marking which triggers a subkind interpretation of mass-nouns as in Table 1 (the universal sorter effect; Gillon 1 992). There are two problems with such an approach. First, there are certain mass nouns which cannot be coerced by plural marking but nevertheless they allow for diminutive marking and the diminutivized noun still behaves like a count noun: Table 3 viel Geschirr much dishes viet Schlaf much sleep viel Luft much air

*viele Geschirr-e many dishespl *viele SchUife many sleeps *vie le Liift-e many airpl

viele Geschirr-chen many dishesolM viele SchUif-chen manYpl sleeps 'naps' viele Liift-chen many airDlM ' breezes'

Why should diminutives count? 67 1

If diminutive suffixes would coerce mass nouns into count nouns we would expect the diminutivized nouns in Table 1 to be ruled out since their base nouns resist coercion otherwise. Furthermore, we observe that selection for a given feature does not usually trigger coercion. For example the verb wonder selects for a [+Q] complement: (2)

I wonder jf/*that that's what' s going on

If selection would trigger coercion, then the embedded clause containing that should be interpreted as an embedded question Gust like diminutivized mass nouns are interpreted as count nouns). I conclude that coercion via selection is not a successful approach towards our problem. There is a second possible approach that comes to mind. Suppose that the semantics of diminutives (roughly small) is such that it requires individuation. Again there are a number of problems with this approach. First, we observe that several nouns which freely allow for diminutive marking resist modification by klein an adjective which refers to smallness in size: kleiner (3) a. *ein D ETindef smallmasc.sg. kleine h. *eine DETindef smallfem.sg

Schlaf a' . sleep Luft b' air

.

ein DETindef ein D ETindef

SchUif-chen kleines sm allneutsg sleep-chen kleines LUftchen smallneut.sg airDlM

Second, smallness does not necessarily refer to the size of an individuated object, it can also refer to the amount of an unindividuated substance as for example the modifier hisser! (see section 4): (4) a. ein . biss-erl Bier DETindef bitDlM beer 'a little bit of beer'

b.

ein biss-erl Wasser DETindcf bitoIM water 'a little bit of water'

Third, in Halkomelem Salish, diminutive marking (Ci-reduplication) never turns a mass noun into a count noun: combined with an individual-denoting noun the diminutive marker is translated as 'small'; combined with a substance-denoting noun it gets translated as ' little bit of' :

672 Martina Wiltschko (5)

steqiw stitiqiw spath spipath

'horse' 'small horse' 'bear' ' little bear'

Ihemex lhilhemex spehals spipehtils

'rain' 'I ittle bit of rain 'wind' 'little bit of wind/breeze'

We conclude that there is nothing inherent about the semantics of ' smallness' and thus diminutives which would force an individuated interpretation of the noun it combines with. This means that we should look for a formal account which can be subject to language variation. Thus, it is time to go back to the initial analysis according to which these diminutives are classifiers and I will show that such an analysis does indeed capture the properties of diminutives. In particular, I will show that German diminutive markers behave like classifiers and that the possibility of plural marking diminutive nouns does not render this analysis untenable. I propose that the German diminutive marker is akin to numeral classifiers of the type found in (6): (6)

a.

c.

b.

12 Stiic k Vieh 12 piece cattle 4 Kopf Salat 4 head lettuce

d.

3 Blatt Papier 3 sheet paper 1 2 Mann Besatzung 12 men crew

I tentatively propose that such classifiers (including the diminutive) are best analyzed as light nouns (semi-lexical nouns) which take full nouns as their complements:

[n [n {StUcki-chen} ] [N Vieh]]

(7)

' The lightness of these nouns is consistent with their bleached meaning. The classifiers used in (6) convey: ' piece' , 'flat piece', ' round piece', and 'human ' , respectively. S im ilarly, the dim inutive marker simply conveys ' small piece'. 1 I According to the analysis in (7) the diminutive suffix is a light noun. This implies that diminutivization is a process akin to compounding. Evidence that this is indeed on the right track stems from the patterns of final devoicing. A noun shows final devoicing before a diminutive suffix, but not before a plural suffix or a derivational suffix like (-isch):

(i)

a.

pfsad-a horsepl

a'.

pfsad-iS hors-ish

a". pfsat-�an horsedim

Why should diminutives count? 673

In what follows, I show that diminutives pattern just like numeral classifiers in a number of crucial ways. First, we observe that all of the numeral c lassifiers in (6) turn mass nouns into count phrases. That is, these classified phrases contain numerals, which require countability. On their own these nouns do not allow for numerals. (8) a. * I 2 Vieh 1 2 cattle c. *4 Salat 4 lettuce

b. *3 Papier 3 paper d. * 1 2 Besatzung 1 2 crew

Consequently, the generalization about the function of diminutives has to be slightly revised: it is not that diminutives tum mass nouns into count nouns. Rather, diminutives are light nouns, which take full nouns as their complements and create count phrases. Another way in which diminutives and classifiers behave alike has to do with gender. Given that the l ight noun is the head of the phrase, we correctly predict it to determine gender: It is always the classifier (including the diminutive) and never the full noun which determines the gender of the entire phrase: Gelatine DETfcm gelatine Kuchen h. der DETmasc cake

a'. ein klein-es DETneut smallneut b/. ein gross-es DETneut bigneut

Schnaps DETmasc schnaps b. die Luft DETfem air

a'. das Schnaps-erl DETneut schnapsDlM b'. das LUft-chen DETneut airDlM

(9) a. die

( l O) a. der

·

Blatt Gelatin sheet gelatin StUck Kuchen piece cake

Next we tum to the interaction between classifiers (including diminutives) and plural marking. Recall from above that this seemed to create a problem for the analysis of diminutives as classifiers. Note first, that the possibility for plural marking on diminutives (as in Table 1) only creates a problem if we assume that classifiers are generated in the same functional head as plural marking. But according to the present proposal numeral classifiers are analyzed as light nouns .and as such they should be compatible with

674 Martina Wiltschko

2 plural marking. I will now show that the pattern of plural marking in the context of classifiers (including diminutives) is more complicated. Crucially though, the parallelism between classifiers and diminutives holds. First, we observe that the languages under consideration differ with respect to the possibility of plural marking diminutive nouns. As already mentioned, in Dutch diminutive nouns can be pluralized. In fact, as indicated in Table 1 ), plural marking is obligatory if a plural interpretation is intended. Next we turn to Austrian German, where plural marking of diminutive nouns is possible but it is not required even if a plural interpretation is intended: ( 1 1 ) a. viele many a" . viele many b. viele many b". viele many

Zuck-erl-(n) sugarDlMpl Wasser-I-(n). waterDlMpl B lum-efl-(n) flowerDlMPI Haus-erl-(n) houseDlMpl

a' . viele many

Papier-I-(n) paperDIMpl

b'. viele many

Zahnd-erl-( n) toothDlMpl

Moreover, in Standard German, diminutive nouns are never overtly marked for plural.) Pferd-chenlPferd-lein horseoIM/horseDlM a'. die Pferd-chenlPferd-lein DETpi hOrSeDIM/horseDlM Schnaps-chen/Schnaps-lein b. das DETNEUT.SG' schnapsDIM/schnapsDIM b' . die Schnaps-chen/Schnaps-lein DETpI schnapsDIM/schnap�DIM :

( 1 2) a. das

DET

NEUT.SG

.

:.

. ' . .

2

See Aikhenvald 2003 for several exceptions to the generalizations that classifiers and plural marking are in complementary distribution. Obviously, we must allow for the possibility that some classifiers do not occupy the same functional head as plural markers (see also Wiltschko 2005). :3 In principle, this pattern could receive two different analyses: we are either dealing with the absence of plural marking or with a 0 plural allomorph (which is independently attested in Standard German).

Why should diminutives count? 675

Interestingly, the different patterns associated with the diminutive markers across these languages is reminiscent of the behavior of different types of numeral classifiers supporting the present analysis. First, we observe that there are classifiers which obligatorily require plural marking> in the context of a plural interpretation:

( 1 3 ) a.

eine Prise Salz a bit salt 'a bit of salt' a" *2 Prise Salz 2 bit salt

a' . 2 Prise-n Salz 2 bitpl salt 'two bits of salt'

.

Thus the behavior of the German classifier Prise 'small amount of powdery substance' is reminiscent of the behavior of Dutch diminutives in that plural marking is obligatory. We further note that there are classifiers which allow but do not require plural marking without any obvious difference in meaning: ( l 4) a. 2 StiicklSiicke 2 piece/piecepl c. 2 B latt/BUitt-er 2 sheetlsheetpJ

Holz wood Papier paper

b.

2 Glas/GUiser 2 glass/glasspJ

Bier beer

The optionality of plural marking on these types of classifiers is reminiscent of the behavior of the diminutive marker in Austrian German (­ erl) which shows the same pattern. Finally, there is also a numeral classifier which does not allow for plural marking just like the diminutive suffixes in Standard German are not compatible with >overt plural marking:

( 1 5)

a. 1 2 Mann 1 2 man

Besatzung crew

a'. * 12 Mann-er Besatzung 1 2 manpl crew

Whatever is responsible for determining the possibility of plural marking on numeral classifiers - the same type of restriction can be used to determ ine the possibility of pluralizing diminutivized nouns: it depends on the kind of diminutive suffix used. Note further that the parallel behavior of diminutive suffixes and numeral classifiers manifests itself in the type of plural marker used: classifiers just like diminutive suffixes each select a particular kind of plural marker which is independent of the plural marker of the base noun:

676 Martina Wiltschko

( 1 6) a. HOIz-er woodpJ b. Wein-e winepl c. Papier-e paperpl

a'. 2 Stiick-e Holz 2 piecepl wood h'. 2 Gliis-er Wein 2 glasspl wme c'. 2 Blatt-er Papier 2 sheetpl paper

a". Holz-erl-n woodDIMpl h" . Weind-erl-n wineDlMpl cIt. Papier- I-n paperDrMpl

As shown in ( 1 6) a dim inutivized noun in Austrian German is always pluralized with -n independent of the plural marker the base noun would select.4 This is of course expected under an analysis according to which the dim inutive marker is the head of the complex phrase. We have now seen empirical evidence to the effect that dim inutive suffixes in German behave like numeral classifiers of the type Blatt. Stuck. etc. We have tentatively analyzed these classifiers (including the diminutive suffix) as light nouns. Since (some of) these classifiers can be pluralized it does not come as a surprise that (some) diminutive suffixes allow for pluralization as wel 1. This suggests that the complementarity between classifiers and plural marking affects only certain types of classifiers. Consequently, we need to be careful not to jump to conclusions when we see certain morphemes glossed as diminutives or classifiers: such morphemes can have a different syntax despite their common gloss. In a similar vein, we expect so-called diminutives to display different syntactic behavior cross-linguistically. This is indeed the case. We have already seen that in Halkomelem Salish, dim inutives do not have an individuating function. I claim (without further discussion) that in this language diminutives are not heads but adjoined modifiers and as such they cannot determine the behavior of the entire phrase. 5 According to the present analysis, diminutives (in German) have the same syntax as numeral classifiers (in German), namely that of a light noun. Accordingly, we predict diminutives to be in complementary distribution with such classifiers (in German). This prediction is indeed borne out: a diminutivized noun cannot be classified by a numeral classifier:

4

Sim ilarly, Dutch diminutive nouns are always pluralized with-z independent of the plural marker the noun would select. 5 Th i s is probably also the proper analysis of Spanish and Portugese diminutives which do not change the gender of the noun (cf. Bachrach and Wagner 2005 for such a proposal).

Why should diminutives count? 677

( 1 7) a. 2 2

Glas Schnaps glass schnaps

a'.

*2 Glas Schnaps-erl 2 glass schnapsDlM

Crucially, the classifier itself can be diminutivized (and note that diminutivization of the full noun is still excluded in this context indicating that we are not dealing with some form of diminutive agreement): ( J 8) a. 2 2

Glas-erl Schnops glassDIM schnaps

a'.

*2 G las-erl Schnaps-erJ 2 glassDlM schnapsDIM

tentatively analyze this pattern as follows: suppose that the suffix requirement of the diminutive marker in n can be satisfied either by moving (copying) the main noun (N) to n (as in ( 1 9a» or else by inserting another light noun into n (as in ( 1 9b). Furthermore, if the suffixal requirement of the diminutive is neither satisfied by copying nor by inserting another classifier, we might expect (as a last resort strategy) that it can be filled by an expletive n. I suggest that German hiss- can be analyzed along these lines (as in ( l 9c» . ( 1 9) a. [n [n Schnaps-erlJ [NSeRHafls]] b. [n [n Glas-erlJ [NSchnaps]] c. [n [n hiss-erlJ [NSchnaps]] This analysis is consistent with the fact that hiss is not independently attested (i.e., it is a bound root). If so, we predict that hisserl cannot precede a diminutivized noun. This is indeed the case as shown in (20). For a diminutivized noun to be grammatical in this context it must be embedded within another layer of functional structure, as indicated by the obligatoriness of the indefinite determiner a in this context - a phenomenon I leave for a different occasion. (20)

biss-erl {Schnops/* Schnaps-erll a Schnaps-erl} . Trink a schnapsl schnapsDlM/0ETindef schnapsDlM drink OETindef bitDIM Prost Henk! Cheers Henk! ' Have a little bit of Schnaps! Cheers Henk! '

678 Martina Wiltschko

References

Aikhenvald, A .Y. 2003 . Classifiers. A typology of Noun Categorization Devices. Oxford University Press. Bachrach, Asaf, and Michael Wagner. 2005. The Prosody of Adjunction vs. Complementation. The case of diminutives in Brazilian Portugese. Paper presented at MOT, McGill University. Borer, Hagit. 2004. Structuring Sense: Oxford University Press. Doetjes, Jenny. 1997. Quantifiers and Selection: On the Distribution of Quantifying Expressions in French, Dutch and English. Doctoral dissertation, Leiden University. Gillon, Brendan. 1 992. Towards a common semantics for English count and mass nouns. Linguistics and Philosophy 1 5 :597-693 . Jurafsky, Daniel. 1996. Universal Tendencies in the Semantics of the Diminutive. Language 72:533-578. Wiltschko, Martina. 2005. Many things are not [PLURAL). Ms., UBC.

Adjacency, PF, and extraposition Susi Wurmbrand and Jonathan David BobaJjik

1.

Dutch)

In the OV Germanic languages, certain verbs selecting infinitival complements (roughly, the restructuring predicates) appear to form a tight cluster with the heads of their complements. This is particularly striking in Dutch, where clustering is overtly signaled in some contexts by an inversion of the order of the two verbal heads (assuming a head-final base order). This inversion motivated the original movement (verb raising) analysis in Evers ( 1 975) whereby the embedded verb adjoins to the selecting head, as indicated in ( 1 b). An alternative analysis without syntactic head movement, offered by Haegeman and Van Riemsdijk 2 ( 1 986), takes the inversion to be a PF phenomenon, as sketched in ( l c).

l

Our thinking about syntax, and about verb clusters/verb raising in particular, has greatly benefited from many discussions wjth Henk, and from his many writings on the topic. We are very pleased to be able to offer this paper to the Festschrift. For additional discussion of the ideas presented here, we would also like to thank Hubert Truckenbrodt and the audience of the workshop on Infinitives in Konstanz (September 2004). 2 For Haegeman and Van Riemsdijk, inversion was preceded by a process of rebracketing or reanalysis, which we have indicated in ( I e). For most cases, it is also possible to treat inversion as linearizing syntactic V, VP projections, without any prior rebracketing operation, on the assumption that material from the lower VP has first moved out, leftwards (see Wurmbrand to appear). We will proceed on the assumption that inversion is a PF process, and take no stand on whether or not it is fed by a distinct reanalysis operation. See Wurmbrand (to appear) for a review of the various theories that have been proposed, and the papers collected in E. Kiss and Van Riemsdijk (2004) for recent proposals. There is extremely widespread (but not unbounded) variation among speech varieties and across verb classes in terms of the orders that are permitted (see Wurmbrand 2004); this variation is orthogonal to the point to be made here.

680 Susi Wurmbrand and Jonathan David Bobaljik

( 1 ) a. dat hij een verhaal belooft te vertellen that he a story promised to tell 'that he promised to tell a story ' b. dat hij [ een verhaal ti [ belooft [ te vertellen]i ] c. dat hij [[ een verhaal te vertellen ] belooft ] [[ een verhaal [ te vertellen belooft ] reanalysis) ( [[ een verhaal [ belooft te vertellen ] inversion (PF) As Van Riemsdijk ( 1 998) discusses, inversion in Dutch interacts with extraposition in an interesting way. As (2a) shows, the infinitival complement clause as a whole may undergo extraposition (we treat the infinitival clause as a VP, though nothing h inges here on this choice). When this happens, a further instance of extraposition within the infinitival clause is possible; this is illustrated by movement of the PP to the right edge of VP (2b). (2) a. dat hij ti probeert [ vp de emmer met eeo lepel leeg that he tried the bucket with a spoon empty te scheppen]i to scoop 'that he tried to scoop the bucket empty with a spoon' b. dat hij probeert [[ Vp de emmer ti leeg te scheppen ] the bucket empty to scoop that he tried met eeo lepeli ] with a spoon 'that he tried to scoop the bucket empty with a spoon ' Note that extraposition of the entire infinitival VP as in (2a) is not the only option. As (3a) shows, it is also possible for the infinitival clause to remain in the position preceding the selecting verb proberen, 'try', but when this happens, the inversion process identified above (however it is to be formally described) is obligatory. 3 Exactly in this configuration, though, 3

There is a continuing debate about whether the inversion in (3a) involves reordering just among the heads (,verb raising') or is a special case of reordering of a maximal VP projection where the remaining material from with in the VP complement has undergone leftwards movement, such as scrambling. Examples such as (3a) are perhaps the best cases for the view that at least sometimes, reordering targets something smaller than a VP, since resultative predicates such as /eeg are typically unavailable for scrambling. On these grounds, we assume that at least some projection of the infinitival VP remains in complement position in (3),

Adjacency, PF. and extraposition 68 1

PP-extraposition within the (now headless) infinitival clause is prohibited. (For ease of exposition, we have assigned a trace position to the infinitival verb in (3 b), though if inversion is a PF process, there should be no trace here.) Note that extraposition from a base such as (3a) is in principle possible, however, it must take the PP all the way to the edge of the main clause (3c). (3 ) a

dat hij de emmer met eeo lepel leeg probeert that he the bucket with a spoon empty tried te scheppen to scoop 'that he tried to scoop the bucket empty with a spoon' b. *dat hij [ [ de emmer tpp leeg ti ] met een lepel ] probeert that he the bucket empty with a spoon tried te scheppeni to scoop 'that he tried to scoop the bucket empty with a spoon' c . dat [ hij de emmer leeg probeert te scheppen ] that he the bucket empty tried to scoop met eeo lepel with a spoon 'that he tried to scoop the bucket empty with a spoon'

Van Riemsdijk ( 1 998) describes this interaction as an adjacency constraint on cluster formation/inversion - (3b) is blocked since the PP intervenes between the base position of the infinitive and the higher verb. In other words, extraposition blocks inversion. Since extraposition is in general · optional, this phrasing might suggest that extraposition within an unmoved infinitival complement is free to occur, and simply blocks inversion when it does. But this is not correct, as (4) shows. (4)

*dat hij [[ de emmer tpp leeg probeert ] met eeo lepel ] that he the bucket empty tried with a spoon te scheppeni to scoop 'that he tried to scoop the bucket empty with a spoon'

as indicated, though remaining agnostic about the X' -level of the inverted infinitival verb.

682 Susi Wurmbrand and Jonathan David Bobaljik

The example in (4) is excluded because, as already noted, inversion is obligatory in Dutch in this context. The complement of a verb like proberen 'try' must either extrapose or it must undergo inversion. Thus, one might perspicuously restate the problem the other way around, namely, by saying that inversion (an obligatory process) blocks extraposition (an otherwise optional process) . Now, if inversion is a PF operation, then the interaction just described would appear to be a case of ' look-ahead' - the PF acting as a filter on syntactic derivations. Extraposition (syntax) is blocked if it will have a deleterious consequence for a subsequent component (PF). On some interpretations, this would constitute a rejection of the position that PF is solely an interpretive component (Chomsky 1 965 et seq.), and thus such look-ahead should not be countenanced. We suggest that it is possible to maintain both the PF character of inversion and the syntactic nature of extraposition, while successfully navigating around the look-ahead problem. The key to the proposed solution is a version of the 'copy theory of movement' under which movement occurs in the syntax, creating a chain of copies (or occurrences) of the moved element, but where the choice of copy to pronounce is made at PF, and is thus subject to PF considerations, including string adjacency (as suggested in Bobaljik 1 995, 2002 and Boskovic 200 1 ). Viewed from this perspective, we may maintain that extraposition is always optional, as far as the syntax is concerned. As an instance of syntactic movement, extraposition creates a chain of copies or occurrences of the moved item. It is the role of PF, not syntax proper, to determine which member of the chain (which copy) �eceives phonetic instantiation. In line with the proposals of Bobalj ik (1 995, 2002) and Boskovic (200 1 ), we suggest that the default is to pronounce the higher copy, unless doing so would interfere with some other phonological constraint.' We assume that Van Riemsdijk' s adjacency constraint on PF-inversior{ i s ': one such other phonological constraint (but see below . for a suggestion lo"n ;how it may be ' derived). Thus (5) is an econorhy condition, at PF�4 " . '.' :,. \' : , . '

(5)

Pronounce the highest copy (unless doing so would violate another PF condition).

4 This is called Speak Up in Bobalj ik ( 1 995), building on work of David Pesetsky.

In Bobaljik (2002), this is replaced by a slightly different proposal (Minimize

LF:PF Mismatch), but the difference is of no real consequence here.

Acfjacency. PF, and exIraposition 683

Our proposal then is sketched in part in (6). The first line of the example shows syntactic extraposition targeting the right edge · of the infinitival complement, creating two copies/occurrences of the PP. Since inversion is obligatory, the PF adjacency requirement overrides the PF economy condition in (5) and forces the except ional pronunciation of the lower copy and thus deletion (indicated by strikethrough, of the higher copy). Inversion then applies without hinderance. Importantly, the only non-trivial rule­ interaction here is within PF, there is no P F constraint on syntactic movement. (6)

te scheppen syntax [de emmer met een lep el leeg empty to scoop the bucket with a spoon met eeo lepel] probeert ] with a spoon tried 'that he tried to scoop the bucket empty with a spoon.' PF [[de emmer met eeo lepel leeg te scheppen met eeB lepel] probeert ] (Copy deletion) PF [[ de emmer met een lepel l eeg probeert te scheppen] (Inversion)

2. German

Converging evidence for this account, and reason to believe that inversion is in fact somewhat of a red herring in the fonnulation of the problem, comes from what might be called Haider 's Puzzle in German, as it has been raised in a series of works by Hubert Haider (most recently Haider 2003). In German, for the class of VP-complementation structures (modals, auxiliaries and other restructuring predicates) extraposition to the right edge of an in situ infinitival is blocked. We i llustrate here with relative clause extraposition - compare (7a-b) to PP-extraposition in Dutch in (3b). gegeben] (7) a. *daB er [vp [vp [jenen tRELl etwas that he those something give [die ihn darum gebeten haben] hat] who him for.it asked have has 'that he gave something to those who asked him for it' b. *daB er [vp [vp [jenen tREd etwas ] [die ihn darum that he those something who him for.it gebeten haben] gegeben hat] asked have given has

684 Susi Wurmbrand and Jonathan David Bobaljik

c.

daB er [vp [vp Denen tREd etwas gegeben] hat] that he those something given has [die ihn darum gebeten haben] have who him for.it asked

Just as in Dutch, when the infin itival clause is itself moved (here by VP­ topicalization), the relative clause extraposition in question is licit, as (8) shows. (8) a.

[[vp jenen etwas gegeben] [die ihn darum gebeten those something given who him for.it asked haben]] hat er noch nie have has he yet never ' He has never yet given something to those who asked him for it.' b. [[ Vp jene zu verstehen] [die ihn kritisieren]] hat er noch those to understand who him criticize] ] has he yet nie versucht never tried ' He has never tried to understand those who criticize him. '

While the German facts look strikingly similar to the Dutch puzzle addressed above, one point of difference is that there is no clear evidence for cluster formation in the German cases at issue (see Wurmbrand 2005). In particular, there is no inversion, as there is in Dutch - the word order in German is exactly what is expected from the (right headed, VP­ complementation) structure in the syntax. Truckenbrodt ( 1 995) has addressed the German facts by proposing that extraposition is itself a PF operation, and that it is both driven by, and blocked by, considerations of phonological phrasing. Truckenbrodt's argument for treating extraposition as a PF process comes from a careful study of intonational phrasing in German. Specifically, he suggests the constraint on extraposition in (9). (9)

Let XP be a syntactic category that is canonincally mapped into the prosodic category 1C upon extraposition (where 1C is either the phonological phrase or the intonational phrase in the following). The extraposition from NP will take XP as far as out of a prosodic constituent of the same cateogry 'It. (Truckenbrodt 1 995: 503) ( . . . XP . . . )n => ( . . . tj . . . )n (XPj)1t

Adjacency, PF. and extraposition 685

According to (9), an extraposed phrase of a particular prosodic category has to be placed immediately outside of the phrase (of the same prosodic category) it originates in (unlike syntax, phono logical phrasing is not recursive). The constraint is violated if movement is either too short (i .e., if the extraposed phrase does not leave the prosodic phrase it originates in) or too long (i.e., if the extraposed phrase moves out of more than one prosodic phrases of the same category). The relevant part for our purpose here is the effects (9) has on movement that is too short. Th is constraint together with the properties of prosodic phrasing in German then derives the facts considered above. As shown in ( 1 0), a series of clause final verbs/auxiliaries which belong to one clausal domain is mapped into one single prosodic phrase 1[, whereas a VP in topicalized position, even if not the topmost VP in its original position, constitutes a separate prosodic phrase. ( I O) a. Clause final VPs -----p y

�

�

xp.

�� '"

y

-

p

V2

� � . . .. .. .. ...

Txp

V3

1t

*

b. Topicalized VP CP

� VP

*

c'

686 Susi Wurmbrand and Jonathan David Bobaljik

The evidence for this prosodic phrasing comes from various prosodic properties, including phrasal stress and the distribution of boundary tones (the reader is referred to Truckenbrodt 1 995 for details). These facts provide clear evidence that all clause-final verbs/auxi liaries in a mono­ clausal construction are part of one single prosodic phrase, whereas topicalized phrases are prosodic phrases on their own. Returning now to Haider's Puzzle, the contrasts fol low from Truckenbrodt's constraint in (9): extraposition cannot target any of the intermediate VPs in ( 1 O)a, since this would d isrupt the prosodic phrasing, whereas extraposition to the VP in topicalized position in ( 1 O)b is possible. Since extraposition has to leave the prosodic phrase it originates in, movement to any of the intermediate VPs in ( 1 0)a would be too short. Since the topicalized VP in ( l O)b, on the other hand, is a separate prosodic phrase, extraposition can attach to that VP. We now have an understanding of the prosodic factors that constrain extraposition in German. Truckenbrodt stated the constraint as if extraposition itself was constrained by prosodic phrasing. Yet this leaves us with the same potential for a look-ahead problem that the Dutch data presented. By adopting the copy theory of movement, as above, we may instead assume that extraposition, qua syntactic movement, applies freely, but that the prosodic factors constrain the choice of copy to pronounce at PF. We might then restate Truckenbrodt's proposal as ( 1 1 ), maintaining the remainder of Truckenbrodt's theory unchanged. (1 1)

Choice of copy in an extraposition chain: Pronounce the higher copy ... unless doing so interrupts the maximal parsing of the remaining material into a prosodic phrase.

This, of course, is nothing other than the PF economy condition in (5), but where the interaction is with independently detectable properties of prosodic phrasing, rather than the PF-adjacency condition on inversion. Among the antecedent proposals for PF-regulation of copy choice mentioned above, the proposal here can in particular be seen as the complement to BoskoviC's (200 1 ) proposal for Serbo-Croatian second position clitics. Boskovic argues that clitics obey a non-peripherality constraint at PF, whereby they cannot occur at the edge of an intonational phrase. The copy-pronunciation algorithm he proposes will choose the highest copy (of a clitic) if that does not conflict with the anti-peripherality constraint, but where the highest copy is phrase-peripheral, a lower copy is instead pronounced. Our interpretation of Truckenbrodt's proposal, given

Adjacency, PF, and extraposition 687

in ( I I ) is the flip-side of this coin. Extraposed elements, constituting independent prosodic phrases, obey a peripherality constraint: they cannot be embedded inside another prosodic phrase. Thus, where the highest copy is not peripheral, a lower copy is instead pronounced. 3.

Summary

We have offered a more or less unified account of curious restrictions on extraposition in Dutch and German. In both languages, regardless of whether or not inversion applies (as it does in certain constructions in Dutch), the sequence of clause final verbs cannot be interrupted by extraposed elements, even though the intermediate VP nodes that must be . posited in these structures are in principle targets for extraposition . The constraining factor in both cases is a PF condition, as recognized by Van Riemsdijk (for Dutch) and Truckenbrodt (for German), but in each case, this appears to raise an issue of PF actively constraining syntactic derivations, rather than simply interpreting them. We have suggested that the copy theory of movement, under which the choice of copy to pronounce is ultimately a PF decision, provides a solution for both puzzles, by keeping the interesting rule interaction to a single component, namely PF. In this, we have extended previous proposals along the same lines. We may ask in passing to what degree the two solutions proposed here may be even further unified. If Dutch intonational phrasing turns out to be similar enough in relevant respects to Gennan intonational phrasing, then the inversion phenomenon in Dutch may truly be a red herring, as we hinted at above. Rather than an inversion rule constrai.ned to adjacency, the relevant PF condition that blocks pronunciation of the higher copy of the moved element could be the same for both languages, namely, the prosodic phrasing requirements discussed by Truckenbrodt. We do not know enough about Dutch intonational phrasing to commit to this position at this time. As a final note, we observe that the piece de resistance of this proposal, of course, would be clear syntactic (or semantic) evidence that extraposition has (or can have) applied in a structure like (6), even though in surface fonn, the sentence would be indistinguishable from one in which no extraposition applied (compare the Lower Right Corner effect in Bobalj ik 2002 and the interaction of extraposition and NPI licensing in English in Fox and Nissenbaum 1 999). However, a host of factors conspire with the result that we cannot see how to construct decisive examples for Dutch and Gennan. And thus, we offer this squib as merely an appetizer, hopefully sufficiently tantalizing to warrant further work.

688 Sus; Wurmbrand and Jonathan David Bobaljik

References Bobaljik, Jonathan David. 1 995 . Morphosyntax: The syntax of verbal inflection, Ph.D. Dissertation, MIT, Cambridge, Mass. [Distributed by: MIT Working Papers in Linguistics.} Bobaljik, Jonathan David. 2002. A-chains at the PF-interface: Copies and 'Covert' Movement. Natural Language and Linguistic Theory 20.2 : 197-267. Bo§kovic, Zeljko. 200 I . On the nature of the syntax-phonology interface: Cliticization and relatedphenomena. Oxford: North-Holland Elsevier. Chomsky, Noam. 1 965. Aspects of the theory of syntax. Cambridge, Mass: MIT Press. E. Kiss, Katalin and Henk van Riemsdijk, eds. 2004. Verb clusters: A study of Hungarian, German, and Dutch. AmsterdamlPhiiadelphia: John Benjamins. Evers, Arnold. 1 975. The transformational cycle of Dutch and German, Ph .D. Dissertation, University of Utrecht. Fox, Danny and Jon Nissenbaum. 1 999. Extraposition and Scope: a case for overt QR. In Proceedings of the 18th West Coast Conference on Formal Linguistics (WCCFL 1 8), ed. by Sonya Bird, Andrew Carnie, Jason D. Haugen, and Peter Norquest, 1 32- 1 44. Somerville, Mass: Cascadilla Press. Haegeman, Liliane and Henk van Riemsdijk. 1 986. Verb projection raising, scope, and the typology of rules affecting verbs. Linguistic Inquiry 1 7:4 1 7-466. Haider, Hubert. 2003. V-clustering and clause union: Causes and effects. In Verb constructions in German and Dutch, ed. by Pieter Seuren and Gerard Kempen, 9 1 - 1 26. AmsterdamlPhiiadelphia: John Benjamins. Riemsdij k, Henk van. 1998. Head movement and adjacency. Natural Limguage and Linguistic Theory 1 6.3 :633-678 . Truckenbrodt, Hubert. 1 995. Extraposition from NP and prosodic structure. In

Proceedings of the North Eastern Linguistics Society Annual Meeting 25

(NELS 25), ed. by Jill N. Beckman, . 503-5 1 7. Amherst: University of Massachusetts, GLSA. Wurmbrand, Susi. 2004. West Germanic verb clusters: The empirical domain. In Verb clusters: A study of Hungarian, German and Dutch, ed. by Katalin E. Kiss and Henk van Riemsdijk, 43-85. Amsterdam: John Benjamins. Wurmbrand, Susi. 2005. How complex are complex predicates? Ms., University of Connecticut, Storrs. Wunnbrand, Susi. to appear. Verb clusters, verb raising, and restructuring. In The Syntax Companion (SynCom), ed. by Martin Everaert and Henk van Riemsdijk. Oxford: B lackwell.

A note on functional adpositions Jan-Wouter Zwart

1.

Functional ad positions

A case can be made, as in Zwarts ( 1 997), that all adpositions (pre- and postpositions) are non lexical elements: adpositions form a closed class, they are often grammaticalized nouns or verbs, and they generally serve to link constituents (turning a noun phrase into a complement or adjunct). But unlike true functional elements, adpositions are generally not without referential content - albeit that the concept referred to is not an entity but an element of spatio-temporal organization. Discussion of the lexical­ functional opposition with adpositions, then, often centers on the literal vs. metaphorical use of the adposition, as in at home vs. (good) at math. But as Tseng (200 1 ) shows, the relevant distinction is not clear-cut and solid criteria for classifYing adpositions along those lines do not exist. Still, it would seem that referential content is a useful criterium for distinguishing lexical and functional adpositions. I take functional elements (with Abney 1 987) to be devoid of descriptive content. A grammaticalized noun, such as Dutch richting 'direction, in the direction or is not devoid of content and will therefore still count as a lexical adposition. The same holds for the more peripheral use of at in (good) at math. But of in a hell of a story or the city of Boston or out of the blue seems to function as a mere linker, without a trace of the core meaning of a lexical adposition of (see Zwarts 1 997 and Van Eynde 2004 for similar considerations). My tentative definition of a functional adposition, then, is in ( I ): (1)

A functional adposition is a non inflectional dependency marking particle.

The definition in ( 1 ) implies a distinction between functional adpositions and case affixes, the latter being inflectional dependency marking elements. It may well be that this distinction is artificial, but for now I will assume that adpositions and case-markers are two distinct devices potentially serving the same purpose. It is to be expected that languages will in general have a highly limited number of functional adpositions, since the lexical content of the functional

690 Jan-Wouter Zwart

adposition is by definition arbitrary. If we are interested in finding functional adpositions, then, we should be on the lookout for general, all­ purpose adpositions. 2. Some case studies 2 J The functional preposition in Mende . .

Mende (a Niger-Congo language of the Mande group) is a head-tinal language in the sense that the verb follows the direct object and adpositions are postpositions (Migeod 1 908). Sti ll, Mende has a single preposition a which is presented as meaning 'with' (Migeod 1 908: I l l ), but in fact appears to have a range of uses. In contrast, the postpositions, a number of which are relational nouns, have well-defined meanings (including 'with', conveyed by the referential noun ngeya 'hand'). Some examples of the use of a are given in (2) (Migeod 1 908: J 1 2f): gili a mi (2) a. ye, bit say 2S0.FUT.NEG think P DEM 'He said, do not worry about it. ' b. ndondo a sore ! cease P noise 'Stop that noise! ' c. 1 gu a li-Ia 3SG.NEG can P go-INF ' He cannot go.' d. i ti dewe a fe-fele 3 S0 3PL send P DISTR-two 'He sent them two by two.' I would like to suggest that the .Mende element ! .Cf, ; isJ ;�n : all-purpose dependency marking particle, used ; for direct obj�cts (2a;t,bX·/embedded ' infinitives (2c), secondary predicates (2d), among other uses (e.g. that of conjoining pronouns). If we then hypothesize that Mende a is a functional adposition, the question may arise whether it is an accident that the only preposition in this head-final language is functional. My impression is that no head-initial languages exist where the functional adposition is a postposition. This suggest that the following generalization may hold:

A note onfunctional adpositions 69 1

(3)

If a l�nguage has both pre- and postpositions� and it has a general, all-purpose adposition p, p is invariably a preposition.

2.2. Complex adpositions in Tikar

Tikar (a Broad Bantu language spoken in Cameroon) is a head-initial language featuring both prepositions and postpositions, each class also containing a number of relational nouns (Stanley 1 99 1 ). One of the prepositions, ke, appears to have a generic locative meaning. Apparently all

postpositions and relational nouns may be combined with ke to form complex adpositions. This is illustrated with a true postposition in (4a) and with a relational noun in (4b) (Stanley 1 99 1 :345f): (4) a. mim pyi aye

ke

ndon zwem

1 SG fall sleep LaC bed on ' I fell asleep on the bed. ' b. a k e n I E mw5 ' ke nlim au' 3SG leave with child LaC heart river 'He left with the child on the river.' The complex adpositions express a range of spatio-temporal relations, the particular character of which is a function of the lexical content of the postposition/relational noun. It follows that ke in this particular use may

well be characterized as a functional adposition. This raises the question of whether in complex adpositions the functional element is ever a postposition. Van Riemsdijk ( 1 990) argues that in Continental Westgermanic circumpositions of the type of German allf mich zu 'towards me' [lit. onLoe meACC tODIR] the fmal adposition is functional, but this assumes a different definition of 'functional ' since the final adposition is not semantically vacuous (Van Riemsdijk 1 990:239). Perhaps . significantly, in head-final languages featuring circumpositions, the final element always appears to bring in descriptive content, and the first element may be an all-purpose preposition, as in � Southsemitic Zay b C-o o.dcr 'on', be-oo. 2af8f 'beside , b c-... 2anCi '(temporal) after'� b c-.. y cn eq e 'based on', etc. (Meyer 2005 :274f). This suggests the following generalization: .

(5)

If a language has circumpositions consisting of a lexical and a functional adposition, the functional element is a preposition.

692 Jan-Wouter Zwart

2. 3. Relational noun constructions

Many languages express spatio-temporal relations in a possessive construction where the relational concept is expressed by a (grammaticalized) noun� such that for example in the house is rendered as (the) inside (oj) the house. The relational noun may either precede or follow its complement, depending on the organization of possessive constructions. It seems to be generally the case that relational nouns may be supported by a second adposition� as in Eastern Kayah dy t0 8 khu [lit. 'on � table' s top'] 'on the table (Solnit 1 997:209). In these situations the preposition may easily become a generic relation marker, as in the Tikar examples above, or in the West-Atlantic (Niger-Congo) Fulani language Toucouleur (Sylla 1 993 :94): (6) a. mi naat-ii (e) nder suudu I SO enter-ASP P inside hut ' I entered the hut. ' b. mi yan-ii (e) dow joowre pataas 1 SO fall-ASP P top heap potato 'I fell on a heap of potatoes.' As Sylla ( 1 993 :92) argues, e no longer functions as a productively used directional preposition, suggesting that it has become a functional element. In view of Greenberg's ( 1 966:78) Universal 2 (stating that postpositional languages have preposed genitives and prepositional languages postposed genitives) we expect P-initial languages to show the order in (7a) and P-final languages the order in (7b): (7) a. P - [relational noun - NP(GEN)l b. [NP(GEN) - relational noun] - P '

;

'

,

"

. ,' ' "

' I have seen a number of lan guages: where the ' P el enlerif precedes an NP- relational noun complex, such as verb-initial Eastern Kayah (Solnit 1 997:209) and verb-final Zay (Meyer 200S :274f). But I have not seen any languages where the P-element fonows a relational noun- NP complex, in violation of (7a). This leads to a hypothesis stating that functional adpositions occupying unexpected positions are always prepositions. It would be interesting to see if that hypothesis could be strengthened to (8):

A note onfunctional adpositions 693

(8)

If a lariguage has a relational noun construction which requires or allows support from a functional adposition, the latter is invariably prepositional.

The data at my disposal suggest a number of counterexamples to (8), namely all those languages instantiating the type of (7b), e.g. Supyire (Carlson 1 994: 1 8 1 ), Kham (Watters 2002: 1 37), possibly Zaghawa (Jakobi and Crass 2004: 1 62), and presumably quite a number of others. Clearly, it would have to be established in each case whether the P-element is an adposition or something else (such as a case marker or some other type of suffix). But until that work has been done, there is reason to believe that (8) may be too strong. 3.

Complex adpositions

The observations in section 2 suggest that functional adpositions are at least preferably prepositions. From this perspective, it would appear that the analysis of Gennanic complex adpositions of the type auf mich zu (see above) in Van Riemsdijk ( 1 990) and Rooryck ( 1 996), where the postposition is taken to be a functional adposition, compels us to think of the relevant languages as typologically marked. In view of this, a reconsideration of the analysis of the Germanic type as involving stacked lexical prepositions, comparable to English onto me, might be called for (cf. Zwart 1 993 :359f). Slightly different is the type of out of the house where it is reasonable to consider of as a functional element. However, whereas Rooryck ( 1 996:234) takes of to be a postposition taking out the house as its complement (leading to the anomaly of a functional postposition in a head­ initial language), I would like to suggest that of functions as a linker between out and the house, i.e. as a minor adposition in the sense of Van Eynde (2004) or a functional adposition in the sense ofZwarts ( 1 997). This approach has more in common with that of Koopman ( 1 997) and Den Dikken (2003), where complex adpositions are analyzed as spell-outs of lexical adpositions in combination with some adpositional functional material reflecting aspects of spatial organization. However, the circumstance that these functional elements are associated with descriptive content would again lead me to apply a different tenninology to them. Clearly, I was unable in this squib to do justice to the full array of data and analyses, but if this small contribution could somehow be grafted onto

694 Jan-Wouter Zwart

Henk van Rie m sdij k ' s majestic adpositional oeuvre, the effort will not have been in vain. Refe rences Abney, Steven P. 1 987. The English noun phrase in its sentential aspect. MIT dissertation. Carlson, Robert. 1 994. A grammar of Supyire. BerlinlNew York: Mouton de Gruyter. Dikken, Marcel den. 2003 . On the syntax of locative and directional adpositional phrases. Ms., CUNY Graduate Center. Greenberg, Joseph. 1 966. S ome universals of grammar with particular reference to the order of meaningful elements. In Joseph Greenberg, ed., Universals of language. Cambridge: MIT Press. Jakobi, Angelika and Joachim Crass. 2004. Grammaire du beria (langue saharienne). Cologne: RUdiger Koppe Verlag. Koopman, Hilda. 1 997. Prepositions, postpositions, circumpositions and particles: the structure of Dutch PPs. Ms., UCLA. Meyer, Ronny. 2005. Das Zay: deskriptive Grammatik einer Ostguragesprache (Athiosemitisch). Cologne: Rudiger Koppe Verlag. Migeod, F.W.H. 1 908. The Mende language. London: Kegan Paul, Trench, Triibner and Co. Riemsdijk, Henk van. 1 990. Functional prepositions. In Harm Pinkster and Inge Genee, eds., Unity in diversity: papers presented to Simon C. Dik on his 50th birthday. Dordrecht: Foris Publications. Rooryck, Johan. 1 996. Prepositions and minimalist case-marking. In Hoskulour Thrainsson, Samuel D. Epstein and Steve Peter, eds., Studies in comparative Germanic syntax 11, 226-256. Dordrecht: Kluwer Academic Publishers. Solnit, David. 1 997. Eastern Kayah Li: grammar, texts, glossary. Honolulu: The University of Hawai'i Press. Stanley, Carol. 199 1 . Description morpho-syntaxique de la langue tikar (parlee au Cameroun). Epinay: SIL. Sylla, Yero. 1 993 . Syntaxe peule: contribution a la recherche sur [es universaux du [angage. Dakar: Les Nouvelles Editions Africaines du Senegal. Tseng, Jesse. 200 I . Rethinking lexical and functional prepositions� In Ljiljana S aric and Donald F. Reindl, eds., On prepositions, 283-327. Oldenburg: University of Oldenburg. Van Eynde, Frank. 2004. Minor adpositions in Dutch. Journal of Comparative

Germanic Linguistics 7,

I -58.

Watters, David E. 2002. A grammar of Kham. Cambridge: Cambridge University Press.

A note on functional adpositions 695

Zwart� C. lan-Wouter. 1 993 . Dutch syntax: a minimalist approach. University of Groningen dissertation. Zwarts, Joost. 1997. Lexical and functional prepositions. In Dagmar Haumann and Stefan 1. Schierholz, eds., Lexikalische und grammatische Eigenschaften prtipositionaler Elemente, 1 - 1 8. Tiibingen: Max Niemeyer Verlag.

Bibliography of Henk C. van Riemsdijk

1 973 Riemsdijk, Henk van, and Norval Smith. Zur InstabiliUit komplexer phonologischer Segmente. In Linguistische Perspektiven, ed . by Peter lordens and Abraham ten Cate, 293-304. Tiibingen: Niemeyer. Besten, Hans den, Henk van Riemsdijk, and Catherine Snow. Ambiguous sentences: Perceptual strategies? Spektator 2:4 1 5-432. 1 974 A propos de I'extension du principe A sur A aux syntagmes prepositionnels. In Actes du colloque franco-allemand de grammaire transformationnelle Vol. l , ed. by Christian Rohrer and Nicolas Ruwet, 206-2 1 5. TUbingen: Niemeyer. De relatie tussen postposities en partikels. Spektator 3 : 447-462 . 1 975 A case for a trace: preposition stranding in Ziiritiiiitsch. In Linguistics in the Netherlands 1972-1973, ed. by A. Kraak, 1-94-200. Assen: Van Gorcum . 1 976 The phonology and syntax of the preposition 'met' (with) in Dutch. In : Henk van Riemsdijk (ed.), Green ideas blown up: Papers from the Amsterdam colloquium on Trace Theory, Publications of the Linguistics Department 1 3, University of Amsterdam. 1 977 Wat is de Uitgebreide Standaardtheorie? In Wetenschap en {aal, ed. Bernard Tervoort, 94- 1 1 5. Muiderberg: Coutinho.

Bibliography ofHenk C. van Riemsdijk 697

1 978

A case study in syntactic markedness: The binding nature ofprepositional phrases. Dordrecht: Foris. (Dissertation). On the diagnosis of Wh-movement. In Recent transformational studies in European languages, ed. by Samuel Jay Keyser, 1 89-206. Cambridge, MA: MIT Press. Vat, Jan (as coiauthor). On footnote 2: evidence for the pronominal status of l>rer in Old English. Linguistic Inquiry 9:695-7 1 6. 1 979

Extrapositie van vrije relatieve zinnen in het Duits. Spektator 8:360-37 1 . 1 98 1

On 'adjacency' in phonology and syntax. I n Proceedings of the J lth annual meeting of the North Eastern Linguistic Society, ed. by Victoria Burke and James Pustejovsky, 399-4 1 3. University of Massachusetts, Amherst. Syntaktische Konfiguration und Merkmalsprojektion. In Sprache: Formen und Strukturen. Akten des J5. Linguistischen Kolloquiums, Band 1 , ed. by Manfred Kohrt & Jiirgen Lenerz . Tiibingen: Max Niemeyer. Bouwen aan een nieuwe conceptie voor Letteren. Rede bij de opening van het academisch jaar 1 98 1 - 1 982. Kat� ol ieke Hogeschool Tilburg. Developing a new concept for Humanities. Higher Education and Research in the Netherlands 25 :27-3 5. Henk van Riemsdijk and Anneke Groos. Matching effects in free relatives: a parameter of core grammar. In Theory of markedness in generative grammar. Proceedings of the 1 979 GLOW Conference, Annali della Seuola Normale Superiore, ed. by Adriana Belletti et ai., 1 7 1 -2 1 6. Pisa. Henk van Riemsdijk and Edwin Williams. 1 98 1 . NP-structure. The Linguistic Review 1 : 1 7 1 -2 1 7.

698 Bibliography ofHenk C. van Riemsdijk

1 982

A note on case absorption. Wiener Linguistische Gazette 27/28:7 1 -82. Derivational grammar vs. representational grammar. In Linguistics in the Morning Calm. Proceedings of the 1 98 1 Seoul International Congress of L in gu istics , ed. by In-Seok Yang, 2 1 1 -23 1 . Seoul: Hanshin.

Locality principles in syntax and phonology. I n Linguistics in the Morning Calm. Proceedings of the 1 98 1 Seoul International Congress of Lin guistics, ed. by In-Seok Yang, 693-708. Seoul: Hanshin. Over de verscheidenheid van de menselijke taalwetenschap. Inaugural Lecture. Tilburg Studies in Language and Literature 3 . Henk van Riemsdijk and Riny Huybregts. The Generative Enterprise - A Discussion with Noam Chomsky. Dordrecht: Foris. 1 983

Generatieve Grammatica. Tijdschrift voor Taal- en Tekstwetenschap 3 : 1 93- 1 97. Correspondence effects and the empty category principle. In Studies in generative grammar and language acquisition, ed. by Yukio Otsu, Henk van Riemsdijk, Kazuko Inoue, Akio Kamio, Noriko Kawasaki, 5- 1 6. Tokyo: ICU. The case of German adjectives. In Grammatical categories: Auxiliaries and related puzzles, Vol . l ., ed. by Frank Heny and B. Richards, 223-252. Dordrecht: Reidel. Summary of the plenary session on syntax. In Proceedings of the Xlllth international congress of linguists, ed. by Shiro Hattori and Kazuko Inoue, Tokyo. Henk van Riemsdijk (ed ) Speciaal Generatieve Grammatica, Tijdschrift voor Taal- en Tekstwetenschap 3 . .

.

Henk van Riemsdijk and Konrad Ehlich (eds.). Connectedness in sentence, discourse and text. Tilburg Studies in Language and Literature 4, Tilburg University . Yukio Otsu, Henk van Riemsdijk, Kazuko Inoue, Akio Kamio and Noriko Kawasaki (eds.). Studies in generative grammar and language acquisition. Tokyo:International Christian University.

Bibliography 0/Henk C. van Riemsdijk 699

1 984 Some thoughts on generative grammar and its components. In Proceedings 0/the 1983 Harbin (PRC) conference on generative grammar. Introductory Remarks. In Sentential complementation, ed. by Wim de Geest and Yvan Putseys. Dordrecht: Foris. Derivations- versus Reprasentationsgrammatik. In Syntaktische Struktur und Kasusrelation, ed. by Hartmut Czepluch & Hero Janssen. TUbingen: Gunter Narr.

1 985 On pied piped infinitives in German relative clauses. In Issues in the grammar o/German, ed. by lindrich Toman. 1 65- 1 92. Dordrecht:Foris. Zum Rattenfangereffekt bei Infinitiven in Deutschen Relativsatzen. In Erklarende Syntax des Deutschen, ed. by Werner Abraham. TUbingen: Gunter Narr. Why long reciprocals don't exist. Theoretical Linguistic Research 2:37- 45 . Henk van Riemsdijk and Riny Huybregts. 1 985 . Parasitic gaps and ATB . In Proceedings 0/the NELS XV Conference, 1 6 8- 1 87. Providence, RI 1 984. 1 986 NV-skeletten. Proeven van Taalwetenschap. TABU 1 6: 14 1 - 1 49. Crossover between Acquisition Research and Government & Binding Theory. In Studies in the Acquisition ofAnaphora, Vol. 1 , ed. by Barbara Lust, 3 I 1 -3 1 8. Dordrecht: Kluwer. Henk van Riemsdijk and Liliane Haegeman. Verb Projection Raising, scope, and the typology of verb movement rules. Linguistic Inquiry 1 7:4 1 7466. Henk van Riemsdijk and Edwin Williams. Introduction to the theory of grammar. Cambridge, MA: MIT-Press. Henk van Riemsdijk and Pieter Muysken. Projecting features and featuring projections. In: Henk van Riemsdijk and Pieter Muysken (eds.). Features and projections. Dordrecht: Foris.

700 Bibliography ofHenk C. van Riemsdijk

-

1 987 25 years after 'Current Issues' . In Proceedings of the XIVth International Congress �f Linguists, ed. by Werner Bahner, Joachim Schildt and Dieter Viehweger. Berlin: Akademie-Verlag. Riemsdijk, Henk van, Hubert Haider and Martin Prinzhorn. Syntaxreport Generative Grammatik.Europa. Folia Linguistica XXI/2-4:47 1 -487.

1 988 Against adjunction to the head. McGill Working Papers in Linguistics. 239259 (translated as Contra a adjuncao ao nueleo. Ensaios de Linguistica 1 3 ). Enkele stellingen over interdisciplinair onderwijs. In Interdisciplinariteit, Wetenschap en Beleid, ed. by F. van Doorne, 1. Vromen, and J. van Wezel. Leuven: Acco.

1 989 Over de relatie tussen vorm en betekenis. GLOT 1 1 : 1 07- 1 1 4. The representation of syntactic categories. In Proceedings of the Conference on the Basque Language, Basque World Congress. 1 04 1 1 6. -

Movement and regeneration. In Dialect variation and the theory of grammar, ed. by Paola Beninca, 1 05 - 1 3 6. Proceedings of the GLOW workshop on linguistic theory and dialect variation. Dordrecht: Foris. Swiss relatives. In Sentential complementation and the lexicon. Studies in honor of Wim de Geest, ed. by Dany Jaspers, Wim Klooster, Yvan Putseys and Pieter Seuren, 343-354. Dordrecht: Foris.

1 990 Generative syntax in the Netherlands. In Proceedings of the parasession on linguistics in the Netherlands, Georgetown University Round Table Meeting, spring 1 988. Georgetown University Press Functional prefositions. In Unity in diversity. Papers presented to Simon C. Dik on his 6(j birthday, ed. by Harm Pinkster and Inge Genee, 229-24 1 . Dordrecht: Foris Henk van Riemsdijk and Edwin Williams. 1 990. Introducciim a la teoria gramatical. Madrid: Catedra. (Translation of the 1 986 boc;>k).

Bibliography ofHenk C. van Riemsdijk 70 I

1 99 1 Terug naar ERVANDAAN. In De binnenbouw van het Nederlands, Een bundel artikelen voor Piet Paardekooper, ed. by Hans Bennis & Jan de Vries, 3 1 1 -3 1 7. Dordrecht: leG Publications. Indefinite article reduction in Dutch. Rivista di Grammatica Generativa 1 5 . Henk van Riemsdijk and Louis DesTombe. 1 99 1 . Eurogrammar: Towards a new standard for grammatical descriptions. In Linguistica Computazionale Vol. VII, Special Issue dedicated to Bernard Quemada. Model, Jan (as co-author). Grammatische analyse. Dordrecht: ICG Publications. (chapters 2, 4, and 5 written by Henk van Riemsdijk) Model, Jan (as co-author). Incorporatie in het Nederlands. Gramma 1 5 :5788. Henk van Riemsdijk and Luigi Rizzi (eds.) 199 1 . Clitics and their hosts. ESF-EUROTYP Papers Vol. 8, 1 . Henk van Riemsdijk and Edwin Williams (eds.) 1 99 1 . Introdu�ao a teoria da gramatica. Sao Paolo: Martins Fontes. (Translation of the 1 986 book) 1 992 Kan het zonder grammatica? In De geletterde mens, ed. by Cees van Rees & Ludo Verhoeven . Tilburg: TUP. 1993 Complements, adjuncts and adjacency in phrase structure. In De fa musique a la linguistique. Festschrift for Nicolas Ruwet, ed. by Liliane Tasmowsky and Anne Zribi-Hertz. Special edition of Communication & Cognition, 498-5 1 2. 1 994 Review article of: Ken-Ichi Takam i, Preposition Stranding -- From Syntactic to Functional Analyses" . BerIinINew York: Mouton de Gruyter, 1 992. Linguistics 3 I .

702 Bibliography ofHenk C. van Riemsdijk

Norbert C orver and Henk van Riemsdijk. 1 994b. Introduction: approaches to and properties of scrambling. In: Norbert Corver and Henk van Riemsdijk (eds.). Studies on Scrambling. BerlinlNew York: Mouton de Gruyter Henk van Riemsdijk and Lars Hellan (cds.) 1 994a. Clitic doubling and clitic groups. ESF-EUROTYP Papers, Vol. 8,5. Henk van Riemsdijk and Lars Hellan (eds.) 1 994b. CUtics: their origin, status and position. ESF-EUROTYP Papers, Vol. 8,6. 1 995 Generative grammar. In Handbook of Pragmatics, ed. by Jef Verschueren, Jan-Ola Osunan, Jan Blommaert and Chris Bulcaen. Antwerpen: IPRA. Complementi, aggiunti e adiacenza nella struttura sintagmatica. In Studi di grammatica tedesca e comparative, ed. by Roberto Dolci & G iuliana Giusti, 225-240. Venezia: Centro Linguistico Interfacolta. Another note on clausal pied-piping. In Paths towards Universal Grammar: Studies in honor of Richard S. Kayne, ed. by Guglielmo Cinque, Jan Koster, Jean-Yves Pollock, Luigi Rizzi and RaffaeIa Zanuttini, 33 ] 342. Washington D .C.: Georgetown University Press. 1 996 Adverbia en bepaaldheid. TABU 25: 1 90- 1 93 . Henk van Riemsdijk and Edwin Williams. 1 9 96. wi qo - ruF thR anS qjQ .fhS. Seoul: Hanshin Publishing Co. (Korean translation of 1 986 book.) 1 997 Left Dislocation. In Materials on Left Dislocation, ed. by Anagnostopoulou, Elena, Henk van Riemsdijk and Frans Zwarts, 1 - 1 0. AmsterdamlPhiladelphia: John Benjamins. Push chains and drag chains: Complex predicate split in Dutch . In Scrambling, ed. by Shigeo Tonoike, 7-33 . Tokyo: Kurosio Publishers.

Bibliography ofHenk C. van Riemsdijk 703

Henk van Riemsdijk and Frans Zwarts. 1 997. Left Disclocation in Dutch and the status of copying rules. In Materials on Left Dislocation, ed. by Anagnostopoulou, Elena, Henk van Riemsdijk and Frans Zwarts, 13-29. AmsterdamlPhiiadelphia: John Benjamins. Norbert Corver and Henk van Riemsdijk. The position of the head and the domain of scrambling. In Typology: Prototypes, item orderings and universals. Proceedings oj the LP'96 conference on word order, ed. by Bohumil Patek, 57-90. Prague 1 996. Prague: Charles University Press. Vat, Jan (as co-author). 1 997. Left Dislocation, connectedness and reconstruction. In Materials on Left Dislocation, ed. by Anagnostopoulou, 67-92. Elena, Henk van Riemsdijk and Frans Zwarts, AmsterdamlPhiladelphia: John Benjam ins. Elena Anagnostopoulou, Henk van Riemsdijk and Frans Zwarts (eds.). 1 997. Materials on Left Dislocation. Amsterdam : John Benjamins. Dorothee B eermann David LeBlanc and Henk van Riemsdijk (eds.) 1 997. Rightward movement. Amsterdam: John Benjamins. 1 998

Trees and scions - Science and trees, Fest-Web-Page for Noam Chomsky (32pp). URL: http://mitpress.m it.edu/ chomskydisc/ riemsdyk.html. Head movement and adjacency. Natural Language and Linguistic Theory 1 6 :633-678. Categorial feature magnetism: The endocentricity and distribution of projections. Journal ojComparative Germanic Linguistics 2: 1 -48. Syntax driven (crazy) by morphology. In Mengelwerk voor Muysken, ed. by Adrienne Bruyn and Jacques Arends, 67-74. Publicaties van het Instituut voor AIgemene Taalwetenschap 72 . University of Amsterdam . Hans Broekhuis and Henk van Riemsdijk. 1 998. Een grammatic a voor taalkundigen: A Modem Grammar of Dutch. Nederlandse Taalkunde 3 ;282-290.

704 Bibliography ofHenk C. van Riemsdijk

-

1 999 Clitics: a state of the art report. In Clitics in the languages of Europe, ed. by Henk van Riemsdijk. Vol. VIn of Language typology; 1 -30. BerlinlNew York: Mouton de Gruyter. Henk van Riemsdijk (ed.) Clitics in the languages of Europe. Vol. VIII of Language typology. BerlinINew York: Mouton de Gruyter.

2000 Adjunction and adjacency. In La variation linguistique: des faits aux theories, ed. by Moha Ennaji, 64-8 1 . Publication de la Faculte des Lettres et des Sciences Humaines, Dhar EI Mehraz, FES (Maroc). Free relatives inside out: Transparent free relatives as grafts. In PASE Papers in language studies -- Proceedings of the 8th annual conference of the Polish association for the study of English, ed . by Bozena Rozwadowska, 223-233 . University of Wroclaw. Binnenstebuitenbouw: Enkele observaties over de transparante vrije relatiefzin. In Met taal om de tuin geleid Opstellen voor Georges De Schutter, ed. by Steven Gillis, Jan Nuyts & Johan Taeldeman, 43 7-448 . Universitaire Instelling Antwerpen. Wh-prefixes: The case of waisch in Swiss German. In Naturally! Linguistic studies in honour of Wolfgang Ulrich Dressler, presented on the occasion of his 60th birthday, ed. by Chris Schaner-Walles, John Rennison and Friedrich Neubarth, 423-43 1 . Torino: Rosenberg & SeIlier.

200 1 A far from simple matter: Syntactic reflexes of syntax-pragmatics misalignments. In Perspectives on semantics, pragmatics and discourse. A Festschrift for Ferenc Kiefer, ed. by Robert Hamish and IstvAn Kenesei, 2 1 -4 1 . Amsterdam: John Benjamins. Henk van Riemsdijk and Riny Huijbregts. 200 1 . Location and locality. In Progress in grammar, ed. by Elena Anagnostopoulou and Marc van Oostendorp, 1-23 . Amsterdam: Meertens, KNAW. Hans Broekhuis and Henk van Riemsdijk. 200 1 . Naschrift bij het besprekingsartikel van Sjef Barbiers. Nederlandse taalkunde, 6: 1 49- 1 53 .

Bibliography ofHenk C. van Riemsdijk 705

Norbert Corver and Henk van Riemsdijk. 200 I b. Semi-Lexical categories. In: Norbert Corver and Henk van Riemsdijk (eds.). 200 1 a. Semi-Lexical categories: The content of function words and the function of content words. Berlin: Mouton -de Gruyter.

2002 The unbearable lightness of GOing. The projection parameter as a pure parameter governing the distribution of elliptic motion verbs in Germanic. The Journal of Comparative Germanic Linguistics 5: 1 43- 1 96. Contour templates in syntax? A note on the spreading of (in-)definiteness. In A new century ofphonology and phonological theory. A festschrift for professor Shosuke Haraguchi on the occasion of his sixtieth birthday, ed. by Takeru Honma, Masao Okazaki, Toshiyuki Tabata, and Shin-ichi Tanaka. Tokyo: Kaitakusha. East meets West: aboutness relatives in Swiss German. In Germania et alia. A linguistic webschrift for Hans den Besten, ed. by Jan Koster and Henk van Riemsdijk, 1-20. TilburglGroningen: University of Groningen. Some thoughts on specified ellipsis. In Festschrift for Christer Platzack (Grammar in Focus, Volume II), ed. by Lars-Olof Delsing, Cecilia Falk, GunlOg Josefsson, and Halld6r Sigurdsson (eds.), 257-263 . Lund, Sweden: Dept. of Scandinavian Languages.

2003 Jan Koster and Henk van Riemsdijk (eds). Germania et alia. A linguistic . webschrift for Hans den Besten. TilburglGroningen: University of Groningen . . Noam Chomsky, Riny Huybregts, and Henk van Riemsdijk. Seisei Bumpo no Kuwadate (Japanese translation and revised version of 'The Generative Enterprise ', 1982, Dordrecht: Foris Publications). Tokyo: Iwanami Shoten.

2004 Anne Breitbarth and Henk van Riemsdijk. The role of triggers in linguistic theory: Some introductory remarks. In:

706 Bibliography ofHenk C. van Riemsdijk

Anne Breitbarth and Henk van Riemsdijk (eds.). BerlinINew York: Mouton de Gruyter. Noam Chomsky, Riny Huybregts, Henk van Riemsdijk, Naoki Fukui and Mihoko Zushi. The Generative Enterprise Revisited: Discussions with Noam Chomsky. BerlinINew York: Mouton de Gruyter. Katalin E .Kiss and Henk van Riemsdijk. 2004b. Verb clusters: Some basic notions. In: Katalin E..Kiss and Henk van Riemsdijk. (eds.). Verb clusters: A study of Hungarian, German and Dutch. AmsterdamlPhiiadelphia: John Benjamins Publishing Company. Henk van Riemsdijk and Susi Wurrnbrand (eds.). The Journal of Comparative Germanic Linguistics 7. 2005 Graft is the logically missing case of Merge. Visnyk of the Kiev National Linguistic University 7.2. Silent nouns and the spurious indefinite article in Dutch. [n Grammar and Beyond, ed. By Mila Vulchanova and Tor Afarli, 1 63- 1 78. Oslo: Novus. Positive polarity and evaluation. This volume. [Adaptation by Norbert Corver and Jenny Doetjes of Positivite et evaluation. Tenn paper Universite de Paris VII, Vincennes, 1 970] . Henk van Riemsdijk and Susi Wurrnbrand (eds.). The Journal of Comparative Germanic Linguistics 8. To appear Free Relatives. In The Blackwell companion to syntax, ed . . by Martin Everaert and Henk van Riemsdijk. Volume II, 3 36-378. Oxford: Blackwell. Martin Everaert and Henk van Riemsdijk. The Blackwell companion to syntax. Volumes I-V. Oxford: Blackwell.

1

Index A.-abstraction, 74 "be, 226 IIenough, 29 1 #gwneud 'do', 554 "kasta 'throw', 492 #much, 294 "OJZO 'or so ', 78 "one, 303 "schijnenllijken 'seem ', 1 58 #Sembler 'seem', 1 56 IIsembrare 'seem', 1 52 I#there. See expletive construction #tout 'all', 1 5 6 #wanna, 625 #way, 439 Hwel (affinnative marker/modifier), 5 1 4 /r/, Intrusive -, 2 1 7 Accent, 459 ACD. See Antecedent-contained deletion Acquisition, Language -, 466, 469, 572 Adequacy, Beyond explanatory -, 505 . Adjacency, 49, 390, 626, 679 Adjective, 230, 23 1 , 3 1 1 , 429, 515 Adjunction, 322 Adposition, 1 50, 1 97, 424, 434, 689, 69 1 , 693 Adverb, 3 1 , 47, 1 40, 5 1 5, 565, 610 Agreement, 223 Anti- -, 347 Case -, 263

C lassifier -, 228 Complementizer -, 347, 579 Determiner -, 228 Inverse -, 1 08 Object -, 1 08 Subject-verb -, 548 Alabama Lowering, 2 1 7 Alignment, 45 1 Allomorph, Null -, 1 1 6 Analogy, Structural -, 259 Analytical expression, 146 Anaphor, 304, 506 Animacy, 1 34 Answer, Natural -, 647 Antecedent-contained deletion, 5 85, 606 Argument, 63 5 - chain, 267 Aristotle, 352 Articulatedness of words, 149 Aspect, 24 1 , 557 - shift, 24 1 Modal - Hypothesis, 1 . Assertion, 635, 649 Attribute, 3 1 4 Barrier, . 59 Binding - domain, 540 - theory, 1 27, 506, 600 Long distance -, 1 27 Biolinguistics, 27 1 , 350 Biology, Evolutionary developmental -, 260 Bootstrapping, Phonological -, 470

708 Index

Case - agreement, 263 - licenser, 628 - suspension, 89 - theory, 374 - valuation, 263 Ablative, 434 Benefactive, 434 Dative, 434, 486, 493 Exceptional - marking, 625 Genitive, 25 Inherent -, 375 Locative, 434 Morphological -, 374, 483 Oblique, 434 Structural -, 375 Categorial Identity Thesis, 427 Category, 322 Empty -. 455 Lexical -, 434 Semantic -, 222 Syntactic -, 434 Causativity, Stative -, 568 Chain formation, 507 Checking, 1 64 C ircumposition, 69 1 Classifier, 222, 293, 670 Clause (Non-)canonical - type, J72 Embedded -, 55 Relative -, 3 1 5, 495 Root -, 529 Small -, 430 C litic, 1 22, 4 1 5, 576 - Cl imbing, 47, 1 52 - Cluster, 4 1 6 - LeftlRight Dislocation, 359 Cliticization, 390, 476 Co-indexing, 506

Communication theory , 3 8 5 Comparative, 434, 527 Complement, 1 64 infinitival -, 344 VP- -, 683 Complementizer, 529 - agreement, 347 - system, 53 1 - trace violations, 263, 544 Doubly-filled -, 27 1 , 5 72 Infinitival -, 627 Prepositional, 290 Prepositional -, 627 Complexity, 148, 1 5 1 , 207 Component, 45 8 Compositionality, 439 Computation, Phase-based -, 266 Concord with determiner, 3 03 Condition Alignment -, 45 1 Binding -s, 600 . Elsewhere -, 453 Inclusiveness - See Inclusiveness condition Interface , 269 Phase impenetrability -. See Phase impenetrability condition : '/ Subject�isliiti.d : , 60 l · ct v Conjugation, .Obje i e ·. -, 1 1 4 Conjunctor� Governing -, 426 Constituency test, 202 Constraint, 458 - ranking, 478 Construction Afinite -, 54 Copula - See copula construction Dative-subject -, 365 I

.

-

'

'

.

Index 709

Copula - See copula construction Dative-subject -, j65 Double object -, 66 1 Expletive -, 1 86 F-/I- -, 365 Partitive-genitive -, 235 Possessive -, 692 Way - -construction, 439 Constructional approach, 439 Containment, 323 Proper -, 449 Contiguity of Reference, 206 Contraction, 625 Contrast, 5 1 8 Control, 344, 620, 62 1 , 627 Coordination, 427, 633 , 648 Coordinator, 78 Copula, 1 50 - construction, 1 86 Copy, 602 - theory, 1 02, 682 Dative Shift, 665 Definiteness, 1 93, 4 1 0 Degree word, 294, 526 Deletion, 1 24, 474. See also Drop and Ellipsis Dependency - relation, 257 Formal -, 509 Determiner, 227 Concord with -, 303 Demonstrative -, 3 00 Devoicing. See also Whispering Final -, 476 Diminutive, 669 Directionality, 1 76 Discourse, 1 4 1 Distributivity, 1 3 7 D-linking, 5 74 .

Dominance, 320 Doubly-filled complementizer. See Complementizer Drop Finite ha- - (Swedish), 53 Na- - (Bulgarian), 3 59 Topic -, 533 Dualism, 355 Economy, 509 - of derivation, 1 2 1 - of representation, 427 Global -, 1 68 Element theory, 208 Ellipsis, 54 VP- -, 296, 5 54 Empiricism, 3 5 3 Enchainement, 480 ENFOLD, 276 EPP, 396 Ergativity, 376 Evaluation, 5 1 4 Evaluation metric, 408 Event - identifier, 5 70 - structure, 563 - variable, 74 initial-boundary -, 563 Temporal ly (in)dependent , 566 Evolutionary history, 409 Expletive, 1 84 Extraposition, 25, 679 Feature - checking/valuing, 1 64, 484 - driven movement, 483 - inheritance, 59 (Un)interpretable , 87 Agree- -, 5 9, 280 Case -, 628 -

7 1 0 Index
Modal aspect -, 1 Strict Layer -, 450 Identifiability, 3 04 Idiom, Constructional -, 439 Imperfectivity, 7, 24 1 Inclusiveness condition, 1 67, 508 Incorporation, 8 1 Infinitivus Pro Participio, 1 59 Inflection, 1 1 6, 254 Information - presentation preferences, 3 87 - structure, 1 3 7 Innatism, 384 I-Nominal, 365 Intensifier, 306 Interface condition, 269 Intersubjectification, 1 Intervention, 266 Inversion, 679 Free subject -, 544 Predicate -, 432 IPP. See Infinitivus Pro Participio Language - change, 53, 1 47, 1 49 - contact, 1 46, 480 - diversity, 434 Head-finaVinitial -, 1 75, 690 I-/E- -, 350 Languages Abkhaz, 5 43 Basque, 374 Baule, 670 Bulgarian, 359, 4 1 5 Burushaski, 375 Cantonese, 670 Chinese (Mandarin), 69 Chinese (Shiaoxing), 2 1 7

I

J

'Ii 1 !

Index 7 1 1

Bulgarian, 3 59, 4 1 5 Burushaski, 375 Cantonese, 670 Chinese (Mandarin), 69 Chinese (Shiaoxing), 2 1 7 Chukchee, 1 08 Danish, 547 Dutch, 3 1 , 53, 62, 78, 87, 97, 1 1 8, 1 27, 1 5 8, 235, 241 , 426, 458, 582, 669, 679 - varieties, 209, 230, 474, 477 English. Passim Ewe, 670 Finnish, 543 Frankish dialects, 2 1 0 French, 1 3, 88, 1 23, 1 56, 29 1 , 3 1 3 , 3 1 7, 529, 64 1 Gennan, 3, 1 3 , 22, 32, 1 24, 1 59, 1 8 1 , 1 98, 209, 230, 340, 495, 533, 63 3, 683, 69 1 - varieties, 54, 69, 572, 669 Greek, 390, 543 Greenlandic (West -), 375 Hindi, 1 32, 543 Hungarian, 1 08, 1 37, 1 46, 1 97, 1 99, 3 1 2, 543, 6 1 8 Icelandic, 483, 5 1 2 Italian, 87, 1 23, 1 3 1 , 1 52, 290, 532, 627 Japanese, 1 33, 1 78, 222, 279, 53 1 Kamchadal, 1 08 Kannada, 1 27 Kayah, (Eastem), 692 Kham, 693 Koryak, 1 08 Lezgian, 543

Marathi, 132 Mende, 690 Mongolian, 543 Norwegian, 5 1 2, 547 ojibwa, 670 Picardian, 479 Polish, 1 3 1 , 563 Quechua, 434 Romanian, 3 1 4 Russian, 365 Salish, 67 1 San Dionicio Octotepec Zapotec, 33 1 Scandinavian -, 483 Slavic -, 4 1 5 Spanish, 43 , 532, 543 Supyire, 693 Swedish, 54, 547 Tikar, 69 1 Toucouleur, 692 Turkish, 341 Welsh, 447, 543, 554 Yiddish, 670 Yokuts, 592 Zaghawa, 693 Zay, 69 1 Learning - mechanism, 467 - path, 546 Level - of description, 458 relation between -s, 465 Lexical - entry, 1 1 8 - Integrity hypothesis, 13 5 - item, 1 1 8 Lexicon, 253 Licensing of empty verb, 53 Linear Correspondence Axiom, 25

7 1 2 Index

Locality, 396, 506 Logical Form, 1 3 0 Mapping, 458 Marker, TMA -, 543 Match ing, 5 83 Mathematics, 357 Mental state, 563 Merge, 1 63 , 265 Merger Late -, 1 03 Order of-, 333 Midpositions. See Semi-lexical adposition Minimal ist Program, 275 Missing lexical item, 293 Modal Aspect Hypothesis, I Modality, 8 Modification, 430, 526 Modifier, 3 1 4 Modularity, 458 Module, 253, 407 Mood, Evidential -, 1 53 Morpheme Applicative -, 66 1 Null -, 1 1 6 Movement - as an imperfection, 1 62 - from subject, 59, 65 A- -, 59, 267 A'- -, 59 Copy theory of -, 682 Head -, 297, 3 9 1 Leftward -, 622 Parallel -, 265 Remnant -, 26, 82, 3 1 6, 391 , 614 Repair-driven -, 27 Wh-, 540 Naturalism, Linguistic -, 3 50 Negation, 4 1 5, 5 1 4

Nominalization, 568 Noun Common -, 3 1 0 Deverbal -, 23 8 Light -, 672 Mass/Count -, 292, 669 Relational -, 69 1 Semantic c lasses of-, 225 NP - structure, 625 - Structure, 599 Propositional interpretation of-, 9 1 Split- - Topicalization, 230 Number marking, 670 Object 1 st and 2nd person -, 1 08 Cognate -, 488 Omission. See drop Operator, A.-abstraction -, 74 Optimality Theory, 258, 477 Pair-list readings, 644 Palatization, 479 Parameter, 290, 384, 5 3 8 - schemata, 4 1 0 - setting, 3 84 Directionality, 466 Directionality -, 539 Matching -, 5 83 Stabilization of -s, 272 Very Early - Setting, 546 Parametric linguistics, 407 Parametrization, 407 Parenthetical, 609 Parser - friendliness, 1 75 UG-friendly -, 1 77 Participant, 567 Participle, 1 5 Adjectival -, 1 6

Index

Parser - friendliness, 1 75 UG-friendly -, 1 77 Participant, 567 Participle, 1 5 Adjectival -, 1 6 1- -, 42 1 Past -, 490 Resultative -, 1 9 Stative -, 1 8 Particle Focus -, 23 8 Locative -, 20 1 pseudo-copula temporal -, 366 Particulars, 352 Partitive of, 425 Passive Copular -, 8 8 Existential -, 490 Get- -, 1 4 PATH, 1 1 9 Perceptual mechanism , 47 1 Perfectivity, 7, 24 1 Periphery, Left -, 3 62 Person, 227 PF, 679 Phase, 280, 295, 529, 604 - head, 59 - Impenetrability Condition, 560 Phonetic Fonn, 679 Phonetics, 257, 283 Phonological - epistemological principle, 283 - phrasing, 27, 684 - process, 625 Phonology Declarative -, 257

713

Government -, 207, 259, 283 Lexical -, 252 Post-lexical -, 252 Prosodic -, 458 Pied Piping, 33 1 Platonism, 350 Polarity Negative -, 98 Positive -, 5 1 4 Possessor, 398 Postposition, 1 99, 434, 689 Predicate, 1 40, 635 - relation, 300 Eventive -, 557 Individual/Stage level -, 31 1 Object Experiencer/Psych­ -, 563 Stative -, 5 57 Preposition, 390, 434, 689, 69 1 Missing -, 583 Preposition Stranding, 3 1 Prepositional - complementizer, 290 - phrase, 3 3 1 locational/directional -, 35 Present time interpretation, 556 Presupposition, 5 1 7, 649 Preverbal position, 201 Primitive, 253 Principle, 3 84 Invisible Category -, 1 2 1 Last Resort -, 276 Least Effort -, 275 Phonological epistemological -, 283

7 1 4 Index

Projection - Principle, 1 67 LeftJRight-branching -, 1 77 Matching -, 276 Pronom inal ization Sentence -, 89 Pronoun, 506, 62 1 Demonstrative -, 300 Indefinite -, 3 1 0 Null -, 87 R-, 3 1 R- -, 5 85 Reciprocal --, 1 27 Reflexive --, 1 1 4, 1 27, 623 Relative -, 495 Resumptive -, 347, 579 Weak -, 342 Pronunciation, 289 Proper names, 83 Proposition, 95 Prosodic -- Criterion :, 1 40 - organization, 474 -- structure, 474 Bare -- structure, 1 67 Prosody, 4 1 5, 458, 5 72 Pseudo-gapping, 297 Psychologism, 353 Psychology, Evolutionary -, 35 1 Quantifier, 3 1 6, 64 5 -- with distributive reading, 1 37 Floating -, 1 3 1 Question - test, 1 42 Embedded -, 53 1 Non-exhaustive -, 641 Pair-list readings, 644 Strongly exhaustive --, 64 1

Wh- -, 64 1 Split --, 35 Raising, 627 Raising, Head -, 348 Rationalism, 353 Realization Alternative/Canon ical -, 121 Reconstruction, 1 0 1 , 599 Recoverability, 298 Reduction, 208 Referential index, 508 Reflexivity, 507 Relation Geometric -- on nodes, 322 Integrated spatial -, 34 Linguistic -- on categories, 322 Probe-goal -, 509 Relative clause, 1 50 - with/without that, 60 8 Appositive --, 609 Free -, 97 , 340, 348, 582 Transparent --, 97 Gapless -, 69 Headed -, 340 Infinitival -, 343 Irrealis -, 345 Light headed -, 3 4 1 Matching effect, 340 Parenthetical --, 609 Restrictive -, 608 Universally quantified -, 348 Reordering, 343 Restructuring, 43 R-expressions, 506 Rhythm, 458 Romanian, 3 1 4 Rule

Index 7 1 5

Universally quantified -, 3 48 Reordering, 343 Restructuring, 43 R-expressions, 506 Rhythm, 458 Roman ian, 3 14 Rule Adjustment -, 254 Allophony -, 254 Combination -, 253 Global -, 458 Language-particular -, 538 P- -, 254 Phonotactic -, 253 Phrase structure -, 253 Prosodic -, 253 Transformational -, 254 Schwa, 48 1 Scope, 1 78 Scope fixity, 66 1 Scrambling, 1 79 Secondary triggering, 1 04 Segment, 3 22 Segmentation, 469 Selection Categorial -, 87 Semantic -, 95 Semantic - component, 44 1 - decomposition, 44 1 Semantics, 255 Sentence, Thetic/Categorical -, 1 39 Sociobiology, 3 5 1 Soft mutation, 447 Sonorant, 592 Sonority scale, 2 1 0 Specifier, 1 64 Multiple -, 276

Speech - stream, 468 - wave, 284 Spell-out, 295, 529 - rules, 453 Sprachbund, 479 Spreading, 269 State, Initial/Steady -, 409 Stranding, 332 Stress, 42 1 , 458 - attraction, 208 Main -, 458 Structure - preservation, 1 68 Bare prosodic -, 1 67 Head ascendant -, 277 NP- -. See NP-structure Three-dimensional -. See Three-dimensional structure Stylistic fronting, 547 Subcategorization, 1 1 9 Subextraction, 59, 398 Subject, 1 30 - island, 59 Infinitival -, 6 1 8 Logical -, 1 3 8 Non-nominative -, 365, 372, 547 Null -, 540, 547 Transformed -, 372 Subjectification, 1 Subset Principle, 54 1 Substitution, Gwneud- -, 554 Syllabification , 474 Syllable boundary, 474 Syntactic - diagnostics, 77 - influences, 480 Syntacticon, 1 1 7

7 1 6 Index

Binding - See Binding theory Case -, 374 Communication -, 3 85 Copy. See Copy theory Element - See Element theory Representation -, 666 Three-dimensional structure, 97 Title, 223 Tobler-Mussafia effects, 362 Tone, 2 1 1 Topic, 1 3 7 Topicalization, Split -, 84 TopP, 534 Transfer, 1 74 Transition probability, 468 Tree, 3 1 9 - -order, 324 Truncation, 533 Typology, 540 - of reduction, 208 - of stress attraction, 208 Underspecification, 542 UNFOLD, 276 Uniformity of theta-assignment hypothesis, 484 Universal, 3 5 1 - Grammar, 1 73, 538 Macro- -, 506 Unvoicing. See Whispering UTAH. See Hypothesis, Uniformity of theta-assignment .

.

Variability of linguistic units, 253 Variation, Grammatical -, 408 Velarization of nasal, 209 Verb - Raising, 8 1 - Second, 80, 533 - with dative object, 486 Copular -, 1 1 6 Ergative -, 3 80 Irregular -, 56 1 Modal -, 2, 53 Motion -, 1 1 8 phonologically null - GO, 53 Restructuring -, 43, 1 52, 679 -s like help, 483 Subject control -, 620 Unaccusative -, 380 Unergative -, 3 80, 444 Verbal complex, 43 Virtual conceptual necessity, 4 1 3 Visibility, 2 1 6 Vocalization of Irl, 209 Vowel, Breaking of 5, 479 VP shell, 1 76 Wat voor 'whatfor ', 63 Wat voor-split, 63 Wh-in-situ, -278, 64 1 Whispering, 592 Word - formation, 253 , 333 - length, 465 - order, 1 73 -