MIT Working Papers in Linguistics #1 Papers on Syllable Structure, Metrical Structure and Hannony Processes(1979) #2 Issues in Japanese Syntax (1980) #3 Papers on the Grammar of Semitic Languages(1981)
#4 Papers in Fonnal Syntactic Theory(1982)
#5 Papers in Grammatical Theory(1983)
#6 Papers in Theoretical and Applied Linguistics(1985) #7 Papers in Morphology( l984) #8 Papers in Theoretical Linguistics(l986) #9 The 25 th Anniversary of MIT Linguistics(1987) #10 Functional Heads and Clause Structure(1989) # II Papers from the First Student Conference in Linguistics. MIT(1989) # 12 Papers from the Second Student Conference in # 13 Papers on wh-Movement(1990)
Linguistics. MIT (1990)
#14 Papers from the Third Student Conference in Linguistics, MIT(1991) # 15 More Papers on wh-Movement(1991)
#16 Papers from the Fourth Student Conference in Linguistics, Ohio State University (1992) #17 Proceedings of the Kwa Comparative Syntax Workshop. MIT(1992) #18 Papers on Case and Agreement I (1993) #19 Papers on Case and Agreement 11(1993) #20 Papers from the 5th Student Conference in Linguistics. University of Washington(1993) #21
Papers on Phonology and Morphology(1994)
#22 The Morphology Syntax Connection(1994)
#23 Papers from the 6th Student Conference in Linguistics. University of Rochester(1994) #24 Formal Approaches to Japanese Linguistics: Papers from the May 1994 MIT Conference(1994) #25 The Interpretive Tract(May 1998) #26 Papers on Language Processing and Acquisition(1995)
#27 Papers on Minimalist Syntax(1995)
#28 Papers on Language Endangerment and the Maintenance of Linguistic Diversity(1996) #29 Fo nnal Approaches to Japanese Linguistics: Proceedings of FAJL 2 (1996)
#30 PF: Papers at the Interface(1997) #31 Proceedings of the Eighth Student Conference in Linguistics(1997) #32 Papers from the UPennlMIT Roundtable on Argument Structure and Aspect(May 1998)
#33 Papers on Morphology and Syntax. Cycle One #34 Papers on Morphology and Syntax. Cycle Two
#35 Papers from the UPennIMIT Round Table on the Lexicon #36 SCIL 9 Proceedings(200 1)
#37 SCIL 10 Proceedings(2001) #38 SCIL 12 Proceedings(2001) #39 A Few from Building E39 (2001) #40 Proceedings of the HUMIT2000 (200
I)
#41 Proceedings of FAJL 3(2002) #42 Phonological Answers (and Their Corresponding Questions)(2002) #43 Proceedings of HUMIT 2001 (2002) #44 Proceedings of AFLA VIII(2002) #45 SCIL 11 Proceedings(2003)
#46 WAFL
1 Proceedings #47 Romance Op. 47: Collected Papers on Romance Syntax #48 Plato's problem: problems in language acquisition
Perspectives on Phases
MIT Working Papers in Linguistics Volume 49
Edited by
Martha McGinnis and Norvin Richards
February 2005
MITWPL MITWPL is an organization of the graduate students in linguistics at MIT. Department of Linguistics, 32-DS08 MIT 77 Massachusetts Avenue Cambridge, MA 02139
[email protected] http://web.mitedulmitwpl/
NOTICE: The MIT Working Papers in Linguistics is, as its name implies, work in progress. As such, the papers contained herein do not represent final versions of the work and all authors welcome comments. Inclusion in a volume of MITWPL should not preclude publication elsewhere. All rights remain with the authors.
Contents
Preface ............................................................................................... .............. vii Phase Theory account of absolutive extraction in Tagalog Edith A ldridge ................................................................................... Word-level phases: Evidence from Hebrew Maya Arad . . . . .. ............................ ............. ............................. 29 .... . ..... .
.
Split-DPs, generalized EPP and visibility Alastair Butler and Eric Mathieu
..
.. ......... ..................... ...... ........ ..... 49
The phase structure of tense Jonny Butler ............................... ......... ........................................ 69 ... . ..
A phase-geometric approach to multiple marking systems Andrew Carnie .... . . .... ....... ....... ..... .................................... ....... 87 ........ .
.
QR in the Theory of Phases Carlo Cecchetto .. . . .. . .. . ... ....... ..... ................. ................... ....... 103 .. ...
.
. .
...
.
Phases and autonomous features: A case of mixed agreement in European Portuguese JOQO Costa and Sandra Pereira ..... . . . . ............... ........ ......... ..... 1 1 5 . .. ....
.
.
.
Towards a phase-based theory of sentential stress Arsalan Kahnemuyipour . .. .. . .... . .. . . ... ..... . .. ... ....... . ........
..... .
125
.......................................................................................
147
.
. ..
.
. .
. .
.
..
.
. ..
Phases and cyclic agreement Julie Legate Going through a phase Ora Matushansky .
.....
.. .. .. ... . ... .
.
.. .
. .
.
. ... .......... . ....... ...
UTAH at Merge:Evidence from multiple applicatives Martha McGinnis . . . . . .. . . . . .. ..... ..
On theEPP Shigeru Miyagawa
. ...... . . ..
..
.....
.. .
.
..
.. . . .
On phases and cyclicity Juvenal Ndayiragije
....... . .............. . .. .... . .
.
Asymmetries in Prosodic Domain Formation Michael Wagner . . .. . . . ..
. .
. . ..
. ...........
.
. 1 57 .
.
......... .
.......
...
.. . . .
183
..
.
.
....... .. ... .... .... 237
.
.... . .
. ....... .
.
..
..
.
... . .... .. ...... . ...
.... 265 .
. .. ........... . .. . . 283
. .......... .......... ..
Phase edge and extraction: A Tagalog case study Andrea Rackowsld and Norvin Richards . ... .. ...
.
.. .
.. . . .. . ....................... .
Derivations without the activity condition . . ..... Andrew Nevins ........... .. .... . . .. .
.
..... .. . .... .. ......
...... ........... . . .... . . 201
.
.. . .. . .. ........... ............ ..
Transparent parsing: Phases in sentence processing Iris Mulders .. .. ... ..........
.... .....
.
.
.
.
. ..
........ . ... .
..
. ..
..
. ..
. . .
. ..
. . ......... .. ...
.
.. .
307
. . . ... . ......... .. .. . . . 329
.. ............ ....... . ... . .
.. .
.
.. .
...
This volume is a proceedings of the Workshop on Phases and the EPP. which was held at MIT. January 16-17, 2003.
The editors would like to thank Lea
. Nash and Alec Marantz for their help in organizing the workshop. MIT's O�partment of Linguistics and Philosophy for funding it, and the audience and
presen ters for their contributions to a lively and informative discussion.
The cover shows a sketch by Galileo Galilei moon.
(1564-1642)
of the phases of the
Preface
Recent work in syntactic theory has suggested that the syntactic derivation proceeds in a series of units known as phases. Phases are taken to be units for phonology, semantics, and syntax. Within the domain of syntax, phases are taken to be relevant for movement in several respects: they (partly) determine the range of syntactic objects that can move, and the domains that constrain movement. These ideas have been developed in various ways in recent work. In this volume, we bring together work on a wide range of topics arising out of this research program. The papers in this collection address a number of issues, including; •
• • •
What are the diagnostics for phasehood? Which categories do these diagnostics identify as phases? How do phases constrain movement? What is the role of EPP-driven movement? What is the empirical scope of phase theory?
These questions are raised against the background of existing work on phases, notably that of Chomsky (2000, 2001a, 2001 b). Chomsky proposes that the derivation proceeds incrementally: the operation Spell-Out sends units of structure- phases- to the phonological component as they are completed. Chomsky 2001a raises the possibility that phases are also the points at which material is sent to the semantic component, and refers to the operation that sends material to both interfaces as Transfer. On this approach, we expect to find independent semantic and phonological evidence for domains that play a role in the syntax. Chomsky proposes that the relevant domains are the ones he calls "propositional", namely CP and active transitive vP. As we will see. several of the papers in this volume further investigate the question of how phases are to be defined. As a consequence of Spell-Out. Chomsky suggests, the material sent to the interfaces becomes inaccessible to the syntax. He proposes the Phase Impenetrability Condition (PIC), which makes a phase inaccessible to operations outside the phase after Spell-Out. Crucially, material at the edge of the phase is exempt from the PIC; the edge consists of any specifiers of the head H of the phase, as well as any adjuncts to HP. For Chomsky (2000), Phase Impenetrability takes effect as soon as the phase has been completed, while Chomsky (2001 b) proposes a looser version of the PIC, according to which it is delayed until the next higher phase is completed. Several of the papers collected here consider evidence bearing on the correct formulation of the PIC along with the relation between the PIC and syntactic locality, and explore the possibility that the PIC subsumes previously posited conditions on movement, such as Shortest Attract. The tight conditions on movement operations imposed by the PIC can be evaded via successive-cyclic movement through the edge of the phase. Chomsky likens such movement to EPP-driven movement to the specifier of TP,
vii
arguing that both movements are motivated by purely syntactic considerations. though they may have intepretive consequences. He proposes that EPP features arc available both on T and on the heads of phases. Once a phrase has been attracted to the edge of a phase, it is exempt from the PIC and may continue to move after the phase has been spelled out. This approach raises a number of questions about the nature of EPP-driven movement, some of which are addressed by papers in this volume. A number of papers in this volume discuss the issue of how phases are to be defined and identified. Julie Legate, for example. draws on her earlier work in which she used existing diagnostics for phasehood to argue that unaccusative and passive vPs are strong phases, and explores the consequences
of this conclusion for the theory of agreement. Other authors aim to sharpen the concept of phase by giving it an explicit semantic characterization. Jonny Butler proposes that phases correspond to domains of existential closure, providing a syntactic representation of situation semantics. Andrew Carnie
develops an account of the complex properties of Case systems in Nez Perce and Dyirbal. based on the hypothesis that each argument is associated with its own phase. Ora Matushansky questions whether a coherent notion of the phase is even possible, arguing that standard diagnostics yield conflicting results in the DP domain. Several of the papers in this volume explore the role of phases and EPP in syntactic movement. One cluster of papers addresses the nature of the Phase Impenetrability C ondition: under what circumstances can the contents of a phase be targeted from outside the phase? Carlo Cecchetto shows that the PIC captures intricate conditions on QR out of non-restructuring infinitives, as long as only the phonological component is fed by mUltiple Spell-out operations.
Juvenal Ndayiragije argues that the PIC should be eliminated in favour of a
general locality condition (Minimal Link Condition I Shortest Attract). Similarly, Andrea Rackowski and Norvin Richards derive the effects of the PIC from Shortest Attract, yielding an account of the conditions on wh extraction in Tagalog. Edith Aldridge provides a different account of Tagalog wh-extraction, arguing that movement to the edge of a vP phase is linked to the
assignment of Absolutive case. Several other papers focus on the nature of EPP driven movement. Shigeru Miyagawa argues that the EPP feature on T works in tandem with agreement in some languages, and with focus in others. Alastair Butler and Eric Mathieu propose two distinct types of EPP features to account
for the optionality of pied-piping in combien (de X) 'how many (X)' questions in French. Andrew Nevins uses quantifier scope interactions in a wide range of languages to argue that EPP plays a crucial role in motivating A-movement, and furthermor� that the role of Case in syntactic movement can be reduced to a ban on multiple Case assignment. The remaining papers in this volume focus on empirical extensions of phase theory. Several of these investigate phonological and morphological implications. Maya Arad proposes that the phase is a special domain for morphology
and
idiosyncratic
meaning,
thereby
predicting
systematic
differences between verbs derived from (phasal) nouns and (non-phasaI) bare roots in Hebrew. Michael Wagner argues that syntactic phases coincide with phonological domains for stress assignment, based on an analysis of stress placement in English and German compounds. Arsalan Kahnemuyipour
viii
provides evidence from English and Persian that sentential stress is assigned to the highest element in the domain of a phase, and that differences in the phrasal phonology of SVO and SOY languages arise from differences in their syntaX.
Iris Mulders invokes a similar notion of "highest element" in a quite different domain of inquiry. She proposes a phase-based theory of sentence processing.. in which a constituent can be reanalyzed to a position outside a phase if it originates as the highest and leftmost element inside the phase. Another pair of of papers extends the syntactic applications of phase theory. Joio Costa and Sandra Pereira argue that phases are domains for syntactic agreement Their proposal accounts for the mismatch between auxiliary and participle agreement with the Portuguese expression a gente ' I pi'. Martha McGinnis uses phase-by phase semantic interpretation to account for apparent constituency conflicts. in Kinyarwanda applicatives. These papers represent an exciting range of approaches to a new set: of questions arising out of recent work in the Minimalist Program. They bring together insights from a variety of subfields of linguistics, offering intriguing answers to some existing questions and raising new ones for future research. -Martha McGinnis and Norvin Richards References Chomsky, Noam (2000). Minimalist inquiries: the framework. In Roger Martin, David Michaels, and Juan Uriagereka (eds.), Step by step: Essays on Minimalist Syntax in Honor of Howard lAsnik, Cambridge, Mass.: MIT Press, pp. 89-155. Chomsky, Noam (200la). Beyond explanatory adequacy.
MIT Occasiorwl
Papers i� Linguistics 20. Cambridge. Mass.: MITWPL.
Chomsky, Noam (200Ib).
Derivation by phase. In Michael Kenstowicz (ed.),
Ken Hale: a Life in lAnguage. Cambridge, Mass.: MIT Press, p.p.
1-52.
Phase Theory account of absolutive extraction in Tagalog Edith Aldridge Stony Brook University
In Tagalog, and other syntactically ergative languages, only absolutive DPs are able to undergo A'-movement.
This
paper proposes a
mechanism which correlates this restriction with absolutive case-cheeking within the theory of Multiple Spell-Out.
I propose that T checks
absolutivc case only in intransitive clauses like antipassives. while v does so in transitive clauses. v is also able to host an EPP feature only when it checks absolutive casco This ensures that an internal argument can move from VP only in
transitive clauses. i.e. when it has absolutive status. An
oblique object in an antipassive will not be able to move out ofVP, since
intransitive v does not check absolutive case and does not host an EPP feature.
1.
Introduction
The purpose of this paper is to propose an analysis of the restriction on A' extraction in Tagalog illustrated in (1). As in many Austronesian languages, Tagalog exhibits a correlation between wh-movement and the type of morpbology on the verb. In (la), a theme is extracted and the transitive inf"IX in- appears on the verb. In ( 1b), an external argument has been moved, and the verb is infixed with the intransitive marker -um-. (1c) shows that the internal argument cannot undergo wh-movement when the intransitive infIX -um appears on the verb. -
(1)
a.
Ano
b-in-asa ang what Comp -Tr.Perf-read 'What did Maria read?'
ni Erg
Maria? M
b.
ang b-um-asa Sino who Comp -Intr.Perf-read 'Who read the book?'
ng ObI
book
b-um-asa ang Comp -Intr.Perf-read what 'Wbat did Maria read?'
si Abs
Maria? M
c.
*Ano
libro?
It is frequently assumed by Austronesian linguists (Georgopoulous 1991, Chung 1998, Rackowski 2002, and others) that this correspondence is the
MIT Working Papers in Linguistics 49, 1-28 n ______ • .:... __ ..... _
nL ____
tr\ "1)11(' C'A•• 1. A'rl�",.. n
Phase theory and absolutive extraction in Tagalog
reflex of an agreement relation in which the verbal morphology indicates the case of the moved DP. (2) -um-in
Case Agreement nominative accusative
This paper takes a different view. I assume an ergative analysis of Tagalog syntax. I exploit the cross-linguistic observation that A' -movement in ergative languages is the privilege of absolutive DPs and propose that the ability of a DP to undergo A'-extraction is correlated with its ability to check absolutive case. As for the relationship between verbal morphology and A' extracted DPs, I propose that -in- and -um- are markers of transitivity and intransitivity, respectively. (3)
-um-
-in
Ergative Analysis intransitive transitive
In transitive clauses in ergative languages, external arguments are marked with ergative case and an internal argument is given absolutive case. In intransitive clauses, including antipassives, it is the highest argument in the clause which checks absolutive case: the external argument in antipassives and the sole argument in simple intransitives. External arguments are able to undergo A'-movement only in antipassives. In transitive clauses, it is the absolutive case-marked internal argument which can be extracted. The issue to be addressed in this paper is how it is guaranteed that the internal argument can move over the external argument precisely in transitive clauses but not in antipassives. The crux of the proposal is that transitive v checks absolutive case with an internal argument and hosts an EPP feature. This EPP feature attracts a VP-internal absolutive DP to the vP phase edge, from where is can undergo further movement to [Spec, CPl. Intransitive v does not have anEPP feature. Therefore, oblique objects in antipassives are be attracted to the vP phase edge and therefore will not be able to move to [Spec, C] without violating the Phase Impenetrability Condition (Chomsky 2000, 2001a, 200tb). In this case, it is only the external argument absolutive, which is merged in the specifier of v, which is eligible to move to [Spec, C]. 2. Absolutive Case-checking Aldridge (2004) proposes an analysis of case-checking in ergative languages in which absolutive case is checked by either v or T. depending on the transitivity of the clause. Unlike accusative languages, in which two structural cases can be checked in a transitive clause, only one structural case is available in ergative languages: absolutive. Ergative case is inherent, assigned to the external argument by v. Which functional head checks absolutive case is detennined by the transitivity of v. Transitive v is merged with the absolutive case feature, just as transitive v in accusative language carries an accusative case feature. However.
2
Edith Aldridge
since only one structural case feature is available in an ergative language, T will not check case in transitive clauses. When v is intransitive, however, it does not carry a case feature, so the absolutive case feature will be merged with T. The rest of this section examines case-checking in different clause types. Tagalog is a VSO language. This word order is derived by moving the verb to T. (4) shows a basic transitive clause. The llninterpretable ahsoilltive case feature is merged on v. This feature probes down into its c-command domain to establish an Agree relation and value the case feature of the closest DP, here the theme. (4)
TP -------
vP
V+v+T
------v' -------
DP[Erg1
VP
tV+v[uCasc:Abs]
-------
t v DP(uCasc:Abs)
In an intransitive clause, v does not carry a case feature. Rather, it is T which checks and values the case feature of the closest DP in its c-command domain. This will be the external argument in an antipassive or unergative clause. (5 )
TP -------
T' -------
V+v+T [uCase:Absl
vP
-------
DP[IICasc:Abs]
VP
In an unaccusative, this will be the internal argument. ] assume with Chomsky (2000, 2001a, 2001b) that unaccusative vP is not a strong phase. Therefore, T and the VP-internal DP can enter into an Agree relation without violating the Phase Impenetrability Condition. (6)
TP
--------
V +V+T[Abs]
vP ------VP
------D P[Abs1 tv
3
Phase theory and absolutive extraction in Tagalog
Tagalog has applicative constructions. These clauses are always transitive, absolutive case being licensed on the applied object. In the following example, the applicative i- licenses a benefactive or instrumental DP as the absolutive.
(7)
Maria ang I-t-in-ali ni Abs M App-Perf-tie Erg 'Maria tied up the box with the tape.'
tape tape
sa Oat
kahon. box
Following Marantz (1993) and Pylkkanen (2002), I propose that the applicative heads a phrase ApplP and selects the applied DP as its argument. ApplP is merged directly below vP. A direct consequence of this is that the applied argument will always be the closest available goal for the probe on v and will always have absolutive status in the clause.
(8)
DP[&g]
v'
.----..-...
V[Abs]
ApplP
.----..-...
Appl'
DP[Abs]
.----..-...
i-
VP
This section has proposed the mechanism for absolutive case-checking, in which this case is checked by either v or T, depending on the transitivity of the . clause. The next section shows how this mechanism relates to the A' extraction facts.
3.
Account of the Extraction Facts
This section discusses how the absolutive case-checking mechanism introduced in the previous section accounts for the asymmetry between internal and external arguments in A' -extraction in Tagalog. As proposed in the previous
section, transitive v is merged with a case feature which values absolutive case on an internal argument. I propose in this section that transitive v also carries an EPP feature, which attracts the closest DP to its outer specifier. It is this EPP feature which allows an internal argument to undergo A' -movement, since only when located in the vP phase edge, can a DP carrying a [wh] feature enter into an Agree realtion with interrogative C and undergo movement to [Spec, C]. (9) shows an example of a basic transitive clause in which the theme wh-word is attracted to [Spec, C).
4
Edith Aldridge
(9)
a.
Ano ang b-in-asa -Tr.Perf-read what Abs What did Maria rea d?
ni Erg
Maria? M
'
b.
CP
� what TP �
V+v+T
vP
� V' twhol � v' lkfaria
�
tv+v(EP
� tv twhat
In an antipassive, v is intransitive. It does not have an absolutive case feature, and therefore also does not have an EPP feature. The result is that an internal argument canno t be attracted to the vP phase edge. The external argument, on the other hand, which is merged in the specifi er of v and therefore located in the vP phase edge in its bas e position, can be attracted to [Spec, C]. (10)
a.
b.
Sino ang b-um-asa who Abs -Intr.Perf-read Who is rea d ing the book?'
libro? book
ng ObI
CP
� TP who ------V+v+T
vP
!who
----V--� ,
t v+v
VP
� book tv
If an i nt erna l argument in an antipassive were to raise directly to [Spec, C], the Phase Impenetrability Condition would be violated accounting for the ungrammaticality of sentences like (11). ,
(11)
a.
an g b-um-asa -Intr.Perf-read what Abs 'What did Maria read?'
*Ano
si Abs
Maria? M
Phase theory and absolutive extraction in Tagalog
b.
*CP
�P
wha
� vP
V+v+T
� v' Maria
� VP
tv+v
�
tv
!what
�
In the case of unaccusatives, where the sole argument of the verb is merged in VP, I follow Chomsky (2000, 2001 a, 200 1 b) in assuming that unaccusative vP is not a strong phase. The internal argument will be able to check case with T and also be attracted by C and undergo wh-movement. (12)
a.
b.
Sino ang Who Abs 'Who arrived?'
d-um-ating? -Intr.Perf-arrive
CP � who TP � vP
V+v+T tv+v
�
P
�
tv !who
The above discussion has focused on interrogative constructions. Transitive v in declarative clauses also carries an EPP feature, meaning that internal argument absolutives must always move out of VP. There is robust evidence that this is the case. Firs� absolutive DPs are always specific. According to Diesing's (1992) Mapping Hypothesis, these DPs would have to . be outside VP at LF in order to escape existential closure. (13)
B-in-iIi ni Maria -Tr.Perf-buy Erg M 'Maria bought the/*a book.'
ang
libro.
Abs
book
Quantifier scope facts also lead to the same conclusion. The following examples indicate that absolutives take scope over other nominals in the clause, indicating that they must be in a higher position at LF than the ergative in (14a) and oblique object i n (14b). Since in the surface word order the absolutive follows the ergative nominal in (14a), this example makes it particularly clear that the absolutive must move in order to take scope over the ergative.
6
Edith Aldridge
(14)
a.
b.
B-in-asa ng [lahat ng Lk all -Tr.Perf.:.read Erg libro] ang [marami-ng book many-Lk Abs 'All the children read many books.' c:> ASS 'many' > ERG 'all'
ang [Iahat Nag-basa all Abs -lntr.Perf-read libro] [marami-ng ng book Obi many-Lk 'All the children read many books.' c:> ABS 'all' > OBL 'many'
bata] ch ild (one set of books) ng Lk
bata] child
(different sets of books)
Semantic properties such as these have long been observed Philippine linguists. Schachter (1976) calls Tagalog absolutives "topics" L_ points out that these nominals are always defmite and referential. Richards (2000) demonstrates that Tagalog absolutives have many of the same characteristics as topics in Icelandic and proposes an analysis in which the absolutive nominal moves obligatorily to an A' topic position. For the pUIposes of this paper, I propose that absolutives in Tagalog obligatorily raise to the vP pbase edge and need not move further in declarative clauses. Empirical evidence that this is the case comes from the fact that quantified absolutives in transitive clauses tend to take scope under negation, indicating that they should be located in a position between Neg and T at LF. (15)
a.
b.
bi-bilh-in ko Hindi l sErg Red.Fut-buy-Tr Neg ang [Iahat ng libro] book Lk all Abs '1 won't buy all the books.' (will buy some, but not all) c:> NEG> ABS 'all'
ng b-in-asa babae Hindi Neg Erg -Tr.Perf-read woman ang [lahat ng libro] Lk all book Abs 'The woman didn't read all the books.' (read some, but not all) r:::> NEG> ABS 'all'
Positing ob ligatory movement of absolutives out of VP also prevents a potential problem for the analysis of A' -movement proposed in this section. Ergative DPs, like oblique DPs, are ineligible for A'-extraction.
(16)
a.
Ano ang b-io-asa what Abs -Tr.Perf-read 'What did Maria read?'
ni Erg
Maria? M
7
Phase theory and absolutive extraction in Tagalog
b.
"'Sino ang
b-in-asa ang who Abs- Tr.Perf-read Abs 'Who read the bookT
libro? book
Obligatory movement of the absolutive DP to the vP phase edge ensures that this nominal is closest to C, blocking attraction of an operator in the position of the external argument. CP
(17)
�P
Crwh
------
V
vP
-----v' book
� v'
wh
�P
V(EP
------
tv
tbook
The issue of linear order must be clarified, however. While wh-words
in
Tagalog appear in clause-initial position, absolutives in declarative clauses
remain in their base positions, following ergative DPs. (I8)
a.
ang b-in-asa Ano Abs -Tr.Perf-read what 'What did Maria read? '
b.
ni Maria B-in-asa -Tr.Perf-read Erg M 'Maria read the book. '
ni Erg
a ng
libro.
Abs
book
Maria? M
Clearly, movement must be overt in the first case and covert in the ' second. Various analyses (Nunes 1999, Pesetsky 2000, and many others) have been proposed recently to ensure proper deletion of either the head or tail of a chain. To account for the difference in (18a) and (I8b), I will tentatively adopt an approach by Richards (2001) which posits that some instances of covert movement can become overt in the case that further movement takes place. I propose that the relationship between v and a VP-intemal absolutive in Tagalog is typically covert, i.e. the head of the chain in the outer specifier of v must delete. However, if further movement takes place, to [Spec, C), movement will become overt, all but the head of the chain deleting. This section has proposed an account of the absolutive restriction on A'-extraction in Tagalog. v carries an EPP feature only in transitive clauses, thereby ensuring that an internal argument can raise out of VP only when it checks absolutive case with a transitive verb. In an intransitive clause, v will not
8
Edith Aldridge
have an EPP feature. Crucially. in an antipassive, only the external argumell will be located in the vP phase edge and can be attracted to [Spec, C].
4.
Cross-linguistic Analysis
Support
for the
Absolutive Case-checkin:
As presented in section 2, this paper proposes that absolutive case is checked b either v or T, depending on the transitivity of the verb. This is a clear departur from traditional approaches to syntactic ergativity like Murasugi (1992), Bittne and Hale (1996), and Ura (2000), under which absolutive case is equated wit nominative. The position taken by this paper is that absolutive DPs do nc necessarily have the properties of subjects. In section 4.1, I present data from a wide variety of ergative languages and demonstrate that, cross-linguistically, absolutives in transitive clauses, which check their case with v, function more like objects than subjects, while external arguments, regardless of whether they have absolutive or ergative case, behave more like subjects. In section 4.2, I discuss the A'-extraction restriction and show that this is also a general characteristic of syntactically ergative languages, again demonstrating the cross-linguistic applicability of the proposal developed in this paper.
4.1 Surface position and case-checking possibilities of the absolutive DP This subsection presents evidence from a broad range of syntactically ergative languages for the split-level absolutive case-checking system proposed in this paper. Specifically, I show that absolutive case-marking is not necessarily correlated with subject status. It is external arguments, regardless of whether they have absolutive and ergative status, which have the properties of subjects. I also point out difficulties raised by these facts for analyses of the type proposed by Murasugi (1992) and others. In a transitive clause in an ergative language, the ergative DP is typically in a position where it c-commands the absolutive DP. The following examples show that external arguments can bind reflexives. In (19a), the ergative DP binds the absolutive. In (19b), the ergative binds an oblique. (19)
a.
x-O-u-kamsa-j r-iib' Compl.3sAbs-3sErg-kill-Suff 3s-self achih lee man the 'The man killed himself.' (Quiche; Larsen and Norman 1979:349)
b.
Kaali immi-nik Junna-p uqaluttuup-p-a-a. self-Mod Junna-Erg Kaali.Abs tell-Ind-Tr-3sg 'lunna; told Kaali about himselfi.· (W. Greenlandic; Manning 1996:136)
n
Phase theory and absolutive extraction in Tagalog
In Tagalo g t an erga tiv e DP can also bind an absolutivet as in (20a). (20b) shows an a ntipass ive absolut ive as the ante c eden t of an obliqu e reflexive .
(20)
a.
b.
ng lalaki P-in-igi l man Erg -Tr . Perf-c on trol 'The man contro lled him se lf. ' siya
Nag-pigil
3s.Abs Intr.Perf-control 'He/she controlled herself.'
sa Oat
sa ri li self
ang Abs
sariti self
.
niya. 3sPoss
niya .
3sPoss
Exte rnal arguments a lso function as hortative or imperative addressees. (21) shows this for Tagal o g and (22) for Yup i k Esk im o '
(21)
a.
Bigy-an give-App
'Give him b.
(22)
a.
b.
siya mo 3sAbs 2sEr g the coffee .'
K-um-ain na -lntr.Perf-eat now 'Let's eat now!'
ng ObI
.
kape. coffee
tayo.
IPI.Abs
Ner-ci-u! eat-2pl-3sg
'You all eat it ! '
lnar-ci!
lie.down-2pl 'You all lie down!'
(Payne 1982:90)
These facts can be a ccommodated in the current analysis, since the external arguments are merged in the outer spec ifier of v, the highest argument position in the clause. In particular, the reflexive binding facts are accounted for straightforwardly. since the antecedent external argument is lo ca ted in a position c-commanding all other arguments in surface word. order.
(23)
TP
.----..-...
V+v+T
vP
.----..-... v' DP[Erg] .----..-... VP tv+v tv
.----..-...
DP[Abs]
In a no nfinite complement clause, PRO is loca ted in the external argument positi on This position can correspond to either the er gativ e or intra nsitive absolutive case-checking position. (24) shows examples of .
10
Edith Aldridge
transitive complement clauses, where PRO is in the ergative position, and an absolutive internal argument appears overtly. (25) shows intransitive examples, where PRO corresponds to the absolutive. (24)
(25)
a.
ikiu-ssa-llu-gu] [PRO Juuna Miiqqat help-Fut-Inf-3sg] children.Abs [(Erg) Juuna.Abs niriursui..pp..u-t. promise-Ind-Intr-3PI 'The children promised to help Junna. ' (W. Greenlandic; Manning 1996:124)
h.
Maria-ng si Nag-ba..balak Maria-Lk Abs Intr.Perf-Red-plan si Pedr01 [PRO tulung-an Abs Pedro help-App (Erg) 'Maria is planning to help Pedro.'
(Tagalog)
a.
Miiqqat [PRO qiti-ssa-llu..tik] children.Abs (Abs) dance-Fut-Inf-4PI niriursui..pp..u-t. promise-Ind..lntr-3PI 'The children promised to dance.' (W. Greenlandic; Manning 1996:124)
b.
ni Gusto Maria-ng want Erg Maria-Lk ng [PRO b-um-ili Obi (Abs) ..Intr-buy Maria wants to buy a book. ' •
libro] book (Tagalog)
Such data pose a challenge for standard Generative analyses of syntactic ergativity, such as Murasugi (1992), Bittner and Hale (1996), and Ura (2000). These analyses equate absolutive case with nominative. Murasugi (1992). for instance, proposes that ergative DPs move overtly to [Spec, AgrO], typically the direct object case-checking position, to check case. Absolutive DPs, on the other hand, check case covertly with AgrS, the subject case checking position. Since absolutive case-checking is always associated with AgrS, it is predicted that PRO always be associated with the absolutive position and that this case not be available for checking in a nonfmite clause. This prediction is contradicted by the data in (24).
Phase theory and absolutive extraction in Tagalog
AgrSP
(26)
-----AgrS' PRO -----AgrS TP -----... T AgrOP -----AgrO' DP[Erg] -----VP AgrO ------
V'
-----OP[Absl v
�
-------
(24) does not pose a prob lem for the current a nalysi s however. Absolutive case in a transitive clause is checked by v and not T, so it is still availab Ie in a nonfinit e clause. ,
TP
(27)
PR
�P
-----V· tpRO -----...
�v
V
VP
�l-J
------
This subsect ion has given evidence from c-command relations and control structures to indicate that transitive absolutiv es are lo cated in a position below the ergative OP and check case with a functional head lower than T. These facts can easily be accommodated in the case-checking system proposed in this paper: transitive v checks absolutive case with an internal argument, while T checks case when the clause is intransitive. 4.2. Absolutive extraction restriction in ergative languages The primary goal of this paper is to relate possibilities for A'-movement to absolutive case-checking. Specifically, this paper proposes that internal arguments can be A' -extracted when they check absolutive case with v and are attracted to the vP phase edge by the EPP feature on v. Internal arguments are prevented from moving in antipassives, because v is intransitive and therefore does check absolutive case or have an EPP feature to attract a VP-internal OP. In an antipassives, then, only the external argument is eligible for extraction.
12
Edith Aldridge
This pattern holds not only for Tagalog but for ergative languages in general. In Dyirbal, for example, relative clauses can be fonned only on absolutives. The sole argument of an intransitive verb can be relativized in (28a). In order to relativize a transitive agent, the clause has to antipassivize, as in (28b). (28)
a.
ngumaj [ej father.Abs (ABS) yabu-nggu
b.
bural-nga-ngu nguma-gu] yabuj [ej mother.Abs (ABS) see-AP-Rel.Abs father-Oat banaga-nyu retum-Nonfut 'Mother, who saw father, was returning.' (Dyirbal; Dixon 1994: 170)
banaga-ngu] retum-Rel.Abs bura-n see-Nonfut mother-Erg 'Mother saw father, who was returning.' (Dyirbal; Dixon 1994:169)
Manning (1996:84) gives examples from West Greenlandic. In the transitive clauses below, only the absolutive internal argument can be relativized, not the external argument. (29)
a.
b.
nanuq Piita-p tuqu-ta-a P -Ergkill-Tr.Part-3sg polar.bear.Abs 'a polar bear killed by Piita' *angut aallaat tigu-sima-sa-a take-Perf-Rel.Tr-3sg man.Abs gun.Abs 'the man who took the gun' (W. Greenlandic; Manning 1996:84)
Manning (1996:24-5) shows the same characteristic is also found in Mayan languages. Transitive patients, as in (30a), but not transitive agents, as in (30b), can be extracted in constituent questions. In order to extract a transitive agent, the clause must be antipassivized, as in (30c). (30)
a.
tzaj chi rna-a7 Rec-Emph 3pAbs Dir Cheep kab' xiinaq man two J 'Jose grabbed the men.'
b.
alkyee-qa x-hi tzaj Rec.Dep-3pAbs Dir who-PI 'Wbom did Jose grab?'
t-tzyu-7n 3sErg-grab-Ds
t-tzyu-7n 3sErg-grab-Ds
Cheep J
13
Phase theory
c.
d.
and
absolutive extraction in Tagalog
*alkyee saj Rec.Dep.3sAbs.Dir who xiinaq kab' man two 'Who grabbed the men?'
t-tzyu-7n 3sErg-grab-Ds
alkyee saj tzyuu-n grab-AP who Rec.Dep.3sAhs.Dir xiinaq ky-e kab' man 3p-Rn two 'Who grabbed the men?' (Mam; Manning 1996:24-5)
The preceding examples indicate that the absolutive restriction on A' movement is a general characteristic of syntactically ergative languages. This fact lends support to the analysis developed in this paper that correlates extraction from VP with the ability of v to check absolutive case.
5.
Alternative Approach to Asymmetry - Case Agreement
the
Austronesian
Extraction
As introduced in the beginning of this paper, other accounts of the extraction asymmetry in Austronesian languages assume an agreement relation between the verb and the case of the moved constituent. In this section, I introduce the case agreement analysis proposed by Rackowski (2002) for Tagalog. Rackowski (2002) assumes Tagalog to be an accusative language. Nominals check or are assigned case in their base positions. Nominative case is checked by T and accusative by v. Inherent oblique case is assigned by applicative projections to DPs in their specifiers. The nominal I call absolutive is assumed by Rackowski to be the subject of the clause. In her analysis, the subj�ct is the nominal which is located closest to T in the structure and enters into an Agree relation with T. As the highest DP in TP, it is also the DP which will be attracted to [Spec, C] in A'-movement contexts, thereby deriving the A' extraction restriction. Under Rackowski's proposal, detennination of the subject of a clause is not the result of checking a specific case, for instance nominative. Subjects may bear nominative, accusative, or oblique case.. Subject choice is determined rather by semantic properties of the OP. By exploiting the generalization that absolutives are always specific and antipassive direct objects generally nonspecific, Rackowski proposes (following Diesing 1992 and others) that specific VP-internal DPs undergo object shift and move to the outer specifier of v. The specific internal argument is then located in the position closest to T and will be the nominal to agree with T. When Agree obtains between T and the subject, the case feature of the subject is copied to T and is spelled out as an affIX on the verb. what I have identified as transitive, intransitive, or applicative morphology. This means that the verbal morphology is not directly responsible for checking case in Rackowski's analysis but is rather the reflex of case agreement. For example, (31) shows a transitive clause, where the theme has absolutive status. For Rackowski, this is an example of accusative agreement.
14
Edith Aldridge Under her analysis, accusative case is checked by v with the internal argument
in its base position.
(3 1 )
Lu-Iutu-in Asp-cook-Acc
a.
'The man b.
will
lalaki ng Case man cook the adobo.'
ang Ang
adobo. adobo
VoiceP man
�
voice
vP
/'-....
�
v[CV]
VP
O[Acq
The direct object then shifts to the phase edge (because it is specific)�jl When T is merged into the structure, it probes into its domain for a DP to chec�jl its uninterpretable features. The closest DP is the theme argument, located initl
the outer specifier of v. Agree takes place between this DP and T, and the;;' accusative case feature of the theme is copied to the verb and spelled-out as the
affix -in.
TP /'-.... VoiceP cook+v+T[ucase]
(32)
"-- adobo�voice'
/"'.... Voice' /'....
man
Voice
vP tv+v�P ........-..
tcook
1irdobo
(33) gives the derivation of the antipassive version of (3 1). In The internal Rackowski's tenns, this is a case of nominative agreement. argument is nonspecific and will not undergo object shift. When T probes for a DP, it will now agree with the external argument. The nominative case feature on this DP is then spelled out on the verb as a reflex of -um-.
15
Phase theory and absolutive extraction in Tagalog (3 3)
TP
� � (?
cook+v+T(cv.
oiceP
man[Nom
oice'
.......
vP
Voice
� tv+v
VP /'-... adobo t.:ook
6. Problems with the Case Agreement Approach By exploiting the different interpretive properties of absolutive and non absolutive internal arguments, Rackowski's analysis is able to straightforwardly identify the absolutive nominal and ensure that it is the DP eligible to undergo A' -movement in Tagalog. However, as I will discuss in this section, this proposal makes unusual typological statements about Tagalog and has doubtful applicability to parallel syntactic phenomena in Austronesian languages outside the Phil ippine subgroup. I will also point out a technical difficulty faced by the case.checking mechanism.
6.1
Function of subject
Rackowski claims that Tagalog patterns with accusative languages in its case marking system. Case is checked with or assigned to nominals in their base positions, as follows. (34 )
Type of Argument Complement of verb External argument
Type of Case Accusative (from v) Nominative (from T)
High applicative Low applicative
Dative (from Appl)
Dative/Oblique (from Appl)
However, this case assignment system is not made to correlate with the grammatical function of subject as observed in other accusative languages. Rackowski states that the subject of a clause is structurally the highest nominal and the one that agrees with T in case features. This nominal does not. however, necessarily function as subject As discussed in section 4.1, it is the external argument and not necessarily the absolutive (or subject) DP, which has the subject properties of binding reflexives, serving as an imperative addressee, and functioning as contro lled PRO in a nonfmite clause. The sole property of absolutive DPs which might be understood to identify them as subjects is the A' -extraction privilege. Keenan and Comrie (1 977) report that if a language allows only one grammatical relation to relativize, then this grammatical relation is the subject of the clause. It should be noted, however, that nine of the eleven languages cited by Keenan and Comrie as having the extraction restriction are Austronesian, many, if not most,
16
Edith Aldridge
of which are ergative. Additionally, Keenan and Comrie choose to accounl the extraction restriction in Dyirbal by identifying the absolutive as a sub:, Bearing in mind the discussion in section 4.2, in which 1 showed that tn� extraction restriction is a general property of syntactically ergative languages, Keenan and Comrie's ( 1 977) relativization hierarchy can then be understood tc imply that if only one nominal can relativize in a given language, then this must be a subject or an absolutive. The extraction restriction in Tagalog therefor� cannot necessarily be assumed to identify the absolutive as a subject. _ _ __
6.2 Case-checking in applicative constructions In addition to the conceptual problem pointed out above, Rackowski's (2002) case agreement analysis also faces technical difficulties. Under Rackowski': analysis, monotransitive verbs check accusative case with the internal argument A specific direct object raises to the outer specifier of Voice where it agrees with T and copies its case feature to the verb, in this case -in. (35)
a.
b.
Bi-bilh-in ko ang lsErg Abs Red.Fut-buy-Tr 'I will buy the book from Maria.'
Ubro
book
kay Dat
Maria. M
TP /"...... buy+T[uCase] VoiceP
boO�M<� Voice
vP /"...... VP v .......
In applicative constructions, Rackowski follows Pylkkanen (2002) in merging an applied object directly into the specifier of an applicative projectiol. where it receives inherent oblique case. The applied DP then raises to the outer specifier of Voice to agree with T, and the case feature is spelled out as tn� applicative affix on the verb, either i- or -an. However, since applicatives appear on transitive verbs which should otherw�se check accusative case, there remains a question as to the fate of tn... accusative case feature of this verb. The presence of the applicative projection should block agreement between v and the theme DP. Because of thi.. Rackowski claims that the theme is assigned inherent case in this situation. But the verb, which is the same verb as in the monotransitive example in (35), should still have an unchecked case feature, which should cause the derivation to crash.
Phase theory and absolutive extraction in Tagalog
( 36)
a.
libro book
ko Bi-bilh-an ng Red.Fut-buy-App 1sErg Obi ' 1 will buy a book from Maria.'
si Abs
Maria.
M
vP
b.
1� vlCVJ
VP ........-.. ApplP buy
Maria[Obl]
� Appl
book{obl]
Under the current proposal, applicative constructions do not pose this type of problem. As discussed in section 2, the applied argument is merged directly in the specifier of the applicative projection, following Marantz (1993) and Pylkkanen (2002). In this position, the applied DP checks absolutive case with v, while oblique objects are assigned inherent case, just as they are in antipassives. (37)
TP
buy+� vP 1� ApplP
v Maria(Abs)
� VP
Appl
18
........-.. tv book[ObI]
Edith Aldridge
6.3 Applied objects as objects or subjects? Under Rackowski's (2002) case agreement analysis, the applied object in applicative constructions is merged directly in the specifier of Appl, where it is assigned inherent case.
(38)
a.
lalaki man
ng Erg
I-t-in-awa App-Perf.laugh
aS8W8.
kanyang
ang
wife his Abs 'The man laughed for his wife.' VoiceP
b.
man
�
ApplP
Voice
wife(o� Appl vP � VP
I laugh
When this DP raises in specificity shift, applicative morphology will C: spelled out on T, and the applied object will be identified as the subject of tn':" clause.
(39)
TP
.......-. laugh+T(uCasc) VoiceP
w{fe(Obl)� man "
�
ApplP
Voice
t wife
� vP Appl v
.......-.
VP
There is some doubt, however, as to whether the applied DP should ��. considered a subject. Cross-linguistically, applicatives are thought to license or create direct objects with accusative structural case (Marantz 1984, Baker 198t. and others), but Rackowski claims they have inherent oblique case. In many languages with applicative constructions, the applied obtect takes on the properties of direct object (Chichewa; Baker 1988:247-8). In (40[. the benefactive applied argument controls object agreement. In (40b) , this argument becomes the subject of a passive.
1�
Phase theory and (40) . a.
absolutive extraction in Tagalog
mtsuko mwana. a-ku-mu-umb-ir-a Amayi SP-Pres-OP-mold-for-Asp waterpot child woman 'The woman is molding the waterpot for the child. 7
b.
zi-na-gul-ir-idw-ansapato Mbidzi SP-Past-buy-for-Pass-Asp shoes zebras 'The zebras were bought shoes by the hare.'
(ndi by
kalulu). hare
These functions of accusative DPs in these constructions unavailable to the theme argument, i.e. the underlying direct object. (4 1 )
a.
b.
* Amayi a-na-u-umb-ir-a mwana Woman SP-Past-OP-mold-for-Asp child 'The woman is molding the waterpot for the child.' *Nsapato zi-na-gul-ir-idw-ambidzi shoes SP·Past-buy-for-Pass-Asp zebras 'Shoes were bought for the zebras by the hare.'
(ndi by
are
mtsuko.
waterpot
kalulu). hare
Rackowski's treatment of applied objects as subjects in Tagalog would be difficult to reconcile with the clear object properties of applied DPs in oilier languages. The current proposal does not, however, suffer from this lack of cross-linguistic applicability. Applied objects check absolutive case with v and therefore should have the grammatical properties of absolutives, the canonical case of direct objects in ergative languages. The object properties of applied objects are therefore accounted for automatically. (42)
TP
� laugh+v+T vP
�
man tV+V[Abs]
ApplP
�VP
Wife(AbS
Appl
6.4 Extension to other Austronesian languages Rackowski's case-agreement analysis may have limited applicability to Tagalog syntax. However, it would face difficulty accounting for the same phenomenon in certain other Austronesian languages. Indonesian is such a language. Unlike Tagalog, Indonesian is essentially an accusative language. Morphologically, Indonesian verbs are inflected as active meng- or passive di-. Subjects move to preverbal position to check nominative case. (43a) and (43b) show examples of an ac tive agent and a passive goal in a ditransitive.
20
Edith Aldridge (43)
mem-beli buku. book A Act-buy 'Ali bought a book.'
a.
Ali
b.
Ali. Buku·nya di-beli A book-Def Pass-buy 'This book was bought by Ali.'
Like other accusative languages with applicative constructL applicative morphology can appear on both active and passive verbs. In Tagalog, only transitive verbs can host applicatives, since the applied object will have the status of absolutive. In the active clause in (44a), the applied object remains in the VP, while the external argument is licensed as the subject. In th: passive clause in (44b), the applied object raises to subject position to check nominative case. It should be noted that this is the same behavior just observed for applied objects in Cbichewa. __ .
(44)
a.
Ali mem-beli-kan Nuri N A Act-buy-App 'Ali bought Nuri a book.
buku. book
•
b.
buku di-beli-kan Nuri N Pass-buy-App book 'Nuri was bought a book by Ali.'
oleh by
Ali. A
Indonesian does share an important syntactic feature with Tagalog. In the case of A' -extraction, only the subject is eligible to move. This is essentially parallel to the absolutive restriction on A' -movement in Tagalog. In an activ.. clause, only the external argument subject can be extracted. Not even thoe applied object is eligible. (45)
a.
b.
c.
Siapa yang mem-beli Comp Act-buy who 'Wbo bought the book?' * Apayang Ali Comp what 'Wbat did Ali buy?'
buku-nya? book-Def
mem-beli? Act-buy
A
*Siapa yang Ali Comp A who 'Who did Ali buy a book?'
mem-beli-kan Act-buy-App
buku? book
Internal arguments can only b e extracted from passive clauses. The applied argument will be extracted in the case of a passive applicative construction.
21
Phase theory and absolutive extraction in Tagalog
(46)
a.
b.
c.
yang Apa what Comp ' What did A li buy?'
di-beli Pass-buy
Siapa yang di-beli-kan who Comp Pass-buy-App 'Who was bought a book by Ali?' *Siapa yang di-beli who Comp Pass-buy 'Who bought a book?'
Ali? A
buku book
oleh
by
Ali? A
buku? book
The similarities and differences between Tagalog and Indonesian can be accounted for straightforwardly with one parameter added to the analysis so far developed in this paper. Case in Indonesian is checked as in other typical nominative-accusative languages: nominative by T and accusative by v.
(47)
Indonesian Case-checking T = Nominative v = Accusative For comparison, I repeat the Tagalog case-checking system below.
(48)
Tagalog Case-checking T :: Absolutive (intransitive verb) v = Absolutive (transitive verb)
Though the case-checking mechanisms differ slightly, the extraction restriction can be analyzed in essentially the same way in the two languages. I propose that in active clauses in Indonesian, like intransitives in Tagalog, v Only passive v in Indonesian (analogous to cannot host an EPP feature. transitive in Tagalog) will carry an EPP feature and allow extraction of an internal argument. The extraction facts are accounted for in the following way. In an active clause, the external argument wh-phrase in [Spec, v] is attracted by C. v checks accusative case with the highest VP-intemal DP. But no DP can raise out ofVP, since there is no EPP feature on v .
22
Edith Aldridge
(49)
buku-nya? book-Def
Siapa yang mem-beli Comp Act�buy who ' Wh o bought the book?'
a
CP
b.
wh
�P
..------.... vP ..------....
v' ..------.... VP m eN-
twho
�ook
an
Internal arguments can be extracted only in passive clauses, when v has EPP feature.
(50)
a.
b.
yang Apa what Comp 'What did Ali buy?'
Ali? A
di-beli Pass�buy
CP ..------.... TP what ..------.... vP di-V+v
..------.... V
twhal
Ali
,
..------.... v' ..------.... VP tV+v[EPP1 � tv
1Whal
As in Tagalog, applied objects are merged in the specifier of ApplP between vP and VP. Applied objects check accusative case with v in active clauses. Again, since there is no EPP feature on v, they are not eligible to undergo A' -movement.
23
Phase theory and absolutive extraction
(5 1 )
a.
*S iapa who 'Who did
yang Ali Camp A Ali buy a book?'
in Tagalog
mem-beli-kan
buku?
Act-buy-App
b ook
* CP
b.
------""P
who
....-----..
vP
Ali
....-----.. v'
....-----.. AppP
meN-
�PPl'
tWh
....-----..
V-lean
VP
....-----.. tv book
Internal arguments can be extracted only in passive clauses, when v has an EPP feature. The applied object will be the one to move in a passive applicative construction.
(52)
a.
b.
Siapa yang di-beli-kan who Comp Pass-buy-App 'Who was bought a book for by Ali?'
buku book
oleh by
Ali? A
CP
....-----.. who TP di-V.kan�vP ....-----.. V !who ....-----.. v· Ali ....-----.. ,
IV-kon+v(EPPl
AppIP
�pl'
twh
....-----.. VP tV-klIn tv
....-----.. book
To summarize, I have shown how the extraction restnctlOn in Indonesian can be accounted for in parallel fashion to the related language
24
Edith
Aldridge
Tagalog. I The passive prefix di- has an EPP feature, which allows an intern&: argument to move out of VP. Internal arguments cannot be A'-extracted in active clauses preflXed with meng-. The Indonesian extraction facts cited above pose a challenge for a case agreement analysis. It might be possible to claim that meng- is the reflex of nominative case agreement and di- of accusative, since the external argument can be extracted in a meng- clause and an underlying direct object in a di clause. (53)
a.
yang mem-beli Siapa Comp Act-buy who 'Who bought the book?'
b.
Apa yang what Comp 'What did Ali buy?'
di-beli Pass-buy
buku-nya? book-Def Ali? A
However, case agreement cannot account for applicative constructions�� the applicative afftx is taken to be an instanciation of oblique case agreemen�\ll as it is in Tagalog, then it is a mystery as to why extraction (and agreement) i$�� possible only when the applicative cooccurs with the passive prefIx and not witli'i: the active prefIx. If
(54)
a.
b.
Ali *Siapa yang Comp A who 'Who did Ali buy a book?'
mem-beli-kan Act-buy-App
di-beli-kan yang Siapa Comp Pass-buy-App who 'Who was bought a book for by Ali?'
buku book
buku? book oleh by
Ali? A
The case agreement analysis therefore cannot be extended as is to Indonesian. Naturally, Rackowski (2002) makes no claim that it can. However, her proposal is made much less attractive by its lack of exportability, since the A' -extraction restriction is a pervasive phenomenon among Austronesian languages. Rather than stipulating mechanisms for individual languages, it is p�eferable to develop analyses with general applicability and broad typological implications. 8. Conclusion The primary purpose of this paper has been to propose an account of the absolutive restriction on A'-extraction in Tagalog. This goal was accomplished by positing an ergative analysis of Tagalog syntax in which absolutive case is checked by either v or T, depending on the transitivity of the clause. The extraction asymmetry is accounted for by allowing an EPP feature to appear on I Though Indonesian is now
an accusative language, I assume the extraction restriction to be a
remnant feature of earlier ergative syntax. Fo r reasons of space, I will not debate this issue here.
2S
Phase theory and absolutive extraction in Tagalog
v only when the verb is transitive, thereby ensuring that internal arguments can move to the vP phase edge only when they have absolutive status . Additional ly, I have shown in this paper how my account of the A' extraction restriction in Tagalog can be extended to Indonesian, which suggests that it may have general applicability in accounting for this common characteristic of Austronesian languages.
References Aldridge, Edith (2 0 04). Ergativity and Word Order in Austronesian Languages. Ph.D. dissertation, Cornell University. Baker, Mark C. (1 988). Incorporation: A Theory of Grammatical Function Changing, Chicago: University of Chicago Press. Bittner, Maria and Ken Hale ( 1 9 96). The structural detennination of case and agreement Linguistic Inquiry 27: 1 -68. Chomsky. Noam (2000). Minimalist inquiries. In R. Martin, D. Michaels, J. Uriagereka (eds), Step by Step: Essays in Minimalist Syntax in Honor of Howard Lasnik. Cambridge, MA: MIT Press. pp. 89-1 55 . Chomsky, Noam (200 1 a). Derivation b y phase. In M . Kenstowicz (ed). Ken Hale: A Life in Language. Cambridge, MA: MIT Press, pp. 1 -52. Chomsky, Noam (200 1b). Beyond explanatory adequacy. MIT Occasional Papers in Linguistics 20. Chung. Sandra (1998). The Design ofAgreement. Chicago: University of Chicago Press. Diesing, Molly (1 992). Indefinites, Cambridge, MA: MIT. Dixon, R.M.W. (1 994). Ergativity, Cambridge, UK: Cambridge University Press. Georgopoulos. Carol ( 1 99 1 ). Syntactic Variables, Dordrecht: Kluwer Academic Publishers. Keenan, Edward and Bernard Comrie (1 977). Noun phrase accessibility and universal grammar. Linguistic Inquiry 8(1): 63-99. Larsen, T. W. and W. M. Nonnan (1 979). Correlates of ergativity in Mayan grammar. In F. Plank (ed.), Ergativity: Towards a Theory of Grammatical Relations, Academic Press, pp. 347-370. Manning, Christopher ( 1 996). Erga/ivity: Argument Strocture and Grammatical Relations, Stanford: CSLI. Marantz. Alec ( 1984). On the Nature of Grammatical Relations. Cambridge. MA: MIT Press. Marantz, Alec ( 1 993). Implications of asymmetries in double object constructions. In S. Mchombo (ed.), Theoretical Aspects of Bantu Grammar. Stanford: CSLI Publications, pp. 1 1 3-1 50. Murasugi, Kumiko G. (1 992). Crossing and Nested Paths: NP Movement in Accusative and Ergative Languages. Ph.D. dissertation, MIT. Nunes, lairo (1 999). Linearizat ion of chains and phonetic realization of chains. In S. Epstein and N. Hornstein (eds.), Working Minimalism . Cambridge, Mass.! MIT Press, pp. 2 17-249. Payne, Thomas E. (1982). Role and reference related subject properties and ergativity in Yup'ip Eskimo and Tagalog. Studies in Language 6(1 ): 75-106. Pesetsky, David (2000). Phrasal Movement and Its Kin. Cambridge, MA: MIT Press. Pylkkanen, Liina (2002). Introducing Arguments. Ph.D. d issertation MIT. ,
26
Edith Aldridge
Rackowski, Andrea (2002). The Structure oJ Tagalog: Specificity. Voice. and the Distribution oJArguments. Ph.D. dissertation, MIT. Richards, Norvin (2000). Another look at Tagalog subj ects. In 1. Paul, V. Phillips, L. Travis (eds.), Formal Issues in Austronesian Linguistics. Dordrecht: Kluwer Academic Publishers, pp. 1 05-1 1 6. Richards, Norvin (20 0 1 ). Movement in Language: Interactions and Architectures. Oxford University Press. Schachter, Paul (1 976). The subject in Philippine languages: Topic, actor, actor-topic, or none of the above. In C. Li (ed), Subject and Topic, New York: Academic Press, pp. 49 1 -5 1 8. Ura, Hiroyuki (2000). Checking Theory and Grammatical Functions in ,Universal Grammar. New York: Oxford University Press.
27
1 1 1 1
1 1
1 1 1
1 1 1
1 1 1
1 1
1 1 1
1 1 1
1 1 1
1 1
1 1 1
1 1 1
1 1 1
1 1 1
1 1
1 1 1
1 1 1
1 1 1
1 1
1 1 1
1 1 1
1 1 1
1 1
1 1
1
Word-level phases: Evidence from Hebrew
Maya Arad Stanford University This paper points out cenain semantic and phonological differences between two groups of Hebrew verbs, those that are traditionally taken to be root-derived and those independently known to be noun-derived. These differences are derived if we assume the first category head (nominal or verbal) with which the root merges creates a point of spell out, or a phase. Beyond motivating low-level phases, the Hebrew data further support the Root Hypothesis. and the syntactic approach to word formation.
Phases play an important role in current generative theol)', yet there seems to be an unresolved ambiguity regarding their nature. They are simultaneously taken to constrain movement and to define a point of semantic and phonological spell out. Most of the work on phases has concentrated on the first property, and so far there have not been cases where the two properties actually stand in contradiction. In this paper I present a clear case for a head defining a semantic and phonological spell-out without (apparently) constraining movement. I suggest that if we take seriously the spel1-out property of phases, then we have to allow phases to be not only (relatively high) maximal projections (vP, CP), but also categol)' heads merging much lower in the tree, at the level of the first categol)' head merging with the root. The empirical evidence in favor of this view is drawn mainly from Hebrew. The discussion of word-level phases interacts, of COUlBe, with a theory of word fonnation. The empirical data bear not only on the word-level phase hypothesis, but also on the Root Hypothesis, which will be explained and illustrated in the following section. I. The Root Hypothesis
and the status orthe Hebrew consonantal root
This paper takes as its starting point the existence of two groups of Hebrew verbs. The first group is assumed, traditionally, to consist of verbs made from roots (atomic lexical elements). The second group is independently argued to be derived from nouns. The two groups differ both semantically and phonologically. Root-derived verbs are free to acquire multiple meanings and may exhibit phonological peculiarities. Noun-derived verbs are tied in their meaning to the noun they are derived from, and do not exhibit phonological idiosyncracies (their phonological fonn preserves the phonology of the noun). I argue that the differences between the two groups are immediately derived if we adopt the idea that spell-out (or a phase) may apply at the level of the first category head that merges with the root (nominal, adjectival or verbal). Thus, when the root merges with a nominal head, its semantic interpretation and phonological incarnation are fixed. Further derivation takes as its input not the MIT Working Papers in Linguistics 49. 29-47
Perspectives on Phases CO 2005 Maya Arad
Word-Level Phases
underspecified core of the root, but the semantic and phonological incarnation of the noun. While providing motivation for the word-level phase hypothesis, the Hebrew data also strongly support the Root Hypothesis, that is, the assumption that languages have underspecified lexical kernels that serve as building blocks for nouns, verbs and adjectives (see Halle and Marantz 1 993 and subsequent work, Marantz 1 997, 2000). While the idea that in all languages words are made from roots is relatively new, in Hebrew, and Semitic languages in general, words are traditionally taken to consist of tri-consonantal roots. It is precisely the existence of this tri-consonantal root that has been subject to debate recently (Bat EI 1 994, 2001 , Ussishkin 1 999). By pointing out the differences between root-derived and noun-derived verbs, this paper provides further support for the existence of the Hebrew consonantal root. Furthermore, it will be shown that the arguments against the Hebrew consonantal root, when examined carefully, actualJy support the existence of the root. This paper is organized as follows. In the remainder of this introductory section, both the traditional view of root-based Hebrew word formation and the recent challenges to this view are presented. Following this, 1 point out, in sections 2 and 3 , the semantic and phonological differences between root derived and noun-derived verbs. In section 4 I make the central claim, that the semantic and phonological differences are immediately captured under the hypothesis that phases may apply at the level of the word. Finally, in section 5, I return to the original argument against the consonantal root and show how the evidence accumulated so far strongly motivates the existence of the consonantal root. Semitic words are traditionally taken to consist of tri-consonantal roots. Roots, on their own, are unpronounceable (it is not possible to pronounce three consonants in isolation), nor are they assigned a fixed meaning. It is only when they are put into palternst arrays of vowels and consonant slots, that roots become actual Hebrew nouns and verbs. Consider the case of the root "xsb and the multiple patterns it is associated with:
(1)
Root
a. b. c.
d. e. f.
"Jdb ""xsb "xSb ""db
""xsb ""xsb
Qattern (nlv momheme} CaCaC CiCCeC hiCCiC CiCCon .maCCeC taCCiC
nlv
xasav xisev he:div xeibon 1 maxiev taxiiv
'to think' 'to calculate' 'to consider' 'arithmetic' 'computer ' 'calculus'
The root "db is a semantic and a phonological core. The semantic core is easily distinguished. All the words created from that root are related to some menta] activity. The verbs this root creates are verbs naming some mental activity (although this activity is underspecified and may be realized as thinking, calculating or considering). The nouns this root creates are the outcome of mental activity (calculus) or instruments used for that activity (calculator). Note that the phonological core of the root is also evident. All the words derived from I
; turns into e before gutturals. resulting in xeibon instead of ·xiSbon.
30
Maya Arad vxsb contain the consonants [x]t [5] and [b]. Furthennore, just as the precise interpretation of the root-as 'think', 'calculate' or 'considert-is given in the environment of the pattern, so is its phonological fonn. It is a well-known fact that h, k and p are spirantized in Hebrew in post-vocalic positions. The root consonant [b] is spirantized if the consonant position of the pattern follows a vowel (as in CaCaC, CiCCeC, hiCCiC, maCCeC and taCCiC), but remains [b] when the consonant slot foHows another consonant (as in CiCCon, xe§hon, in (I d». In the case of the root ;Inpl below, the core of the root is just as evident: (2)
a. b. c. d. e. f.
vnpl CaCaC hiCCiC maCCaC maCCoCet maCCaCa CCoCet
naJal hipi/ mapal mapo/et mapala nefolet
'to fall' 'to drop' 'waterfall 'avalanche' 'defeat' 'fallouCz •
All the words created from the root ;Inpl share a semantic core of falling, or directed motion. They also share the phonological core-all contain the root consonants [n], [P] and [1]. The pattern with which the root combines detennines not only the meaning assigned to the root in that particular environment, but also its phonological fOIm. The root consonant [n] assimilates before another consonant, thus yielding na/al (2a) and nefolet (2£), but hipil (2b) and mapal (2c). The root consonant [p] becomes spirantized in post vocalic positions- yielding na/al (2a) and nefolet (2f).3 The Hebrew data-though partial {for a fuller account of the root see Arad 2005}--motivate a view of the Hebrew root as an underspecified semantic and phonological core. This underspecified core is semantically and phonologically incarnated in the environment of the pattern. Both its semantic interpretation and its actual phonological form are given to the root relative to the environment of the pattern. This view of Hebrew word fOImation is not uncontroversial. An alternative view, originating with Bat El's (1 994) seminal paper (see also Bat EI 200 1 , Ussishkin 1 999, 200 1 ) denies the existence of the Hebrew consonantal root. According to this view, there is no decomposition of words into atomic elements. The Hebrew lexicon contains lexemes (which may be related to one another through correspondence rules). We shall examine the evidence in favor of this view in what foHows. For the time being, let us concentrate on the structure of the argument. The argument consists of two steps. Bat EI ( 1 994) examines a sub-group of Hebrew verbs, denominal and borrowed verbs. These verbs strongly tend to preserve the phonological fonn of their base (nominal prefixes, preservation of consonant clusters, choice of verbal pattern). Such preservation cannot be explained if we take these verbs to be root-derived. On the other hand, if we assume that they are derived from existing words, through 1 Note epenthetic schwa following the first root consonant, avoiding a cluster that
Hebrew.
is illicit in
] Note that after n assimilates, p is not spirantized, as the position is not post-vocalic anymore.
31
Word-Level Phases
an independently motivated process of Stem Modification (of the base word), followed by melodic overwriting, then the properties of these verbs are immediately and elegantly explained. The second step in Bat EI's argument is extending the Stem Modification analysis to all Hebrew verbs. This way, we get a unified theory of all Hebrew verbs, with less theoretical machinery (no need to postulate an abstract consonantal root). I accept Bat EI's argument regarding borrowed and denominal verbs. Indeed, the phonological evidence indicating their being word-derived in very convincing. In the following two sections, however, I show that there is a clear semantic and phonological distinction in Hebrew between word-derived and root-derived verbs. Therefore, one must not generalize from word-derived verbs to root-derived verbs. By showing the effect of roots in word formation, I provide an argument in favor of the existence of the root itself. Before examining the effect of roots in word fOImation, a word is called for regarding the nature of the argument made here. What is at stake is the status of the consonantal root itself. Therefore, we must not presuppose the existence of the root in making the argument Rather, I will show that there are two groups of Hebrew verbs, which differ semantically and phonologically. Interestingly, these groups coincide with the group of verbs traditionally taken to be root-derived, and the group that has been independently shown to he noun derived. The Root Hypothesis captures the difference between the two groups, providing support for the status of the root. For ease of presentation, however, I will refer to the group traditionally taken to be root-derived as to "root-derived verbs", keeping in mind that their root-derived status still requires proof. It is through comparison of these verbs with noun-derived verbs that their root derived status is fully exhibited. 2. Root-derived verbs differ from word-derived verbs semantically
While most Hebrew verbs are derived from roots, some are derived from nouns. When we compare root-derived verbs with noun-derived verbs, a striking contrast appears: while root-derived verbs may take up multiple interpretations, noun-derived verbs must be tied in their meaning to their base noun. To illustrate this, consider the root ..JxSb, mentioned above. The root creates several nouns, as well as three different verbs-xaSov 'think', xiSev 'calculate' and hex.siv 'consider'. Now, when the basis for the derivation is not the root "x�b, but one of the nouns derived from that root, max.sev ( 'computer', 3f), the result is the verb mixSev ('computerize', 3g); (3)
32
Root
a. b. c. d. e. f. g.
l2attem (nlv momheme) CaCaC -.Jdb CiCCeC -.Jxsb hiCCiC CiCCon ..Jdb teCCiC ..Jdb ..Jxsb maCCeC � maxsev 'computer'
\ldb
nlv
xasav xisev
hetiiv
xesbon laxi;v
maxiev
mixSev
'to think ' 'to calculate' 'to consider' 'arithmetic' 'calculus' 'computer' 'computerize'
Maya Arad Consider the relation between the noun maxsev and the verb derived from it, mixsev. The derivational relation between the two is clear. The verb contains not only the root consonants, [x], [8] and [b], but also the nominal prefix m-, which is carried into the verb. I assume that the derivation proceeded as in (4) below first, the root ""xsb combines with the nominal head (spelled out as maCCeC), creating the noun maxSev 'computer', as in (4a). This noun then combines with the verbal pattern CiCCeC, to create the verb mixSev 'computerize', as in (4b):4 (4)
a.
N
--------=-
b.
maxSev 'computer'
""db
Nmacccc
Vmix5cv
�
VCiCCeC
Nmax!cv
�
N
xSb
mixsev 'to computerize'
But the verb mixSev 'computerize' resembles its base noun, max.fev 'computer' not only morphologically, but also semantically. The noun denotes an instrument, or artifact-a computer. The verb derived from that noun means 'equip with a computer' (in the case of computerizing an office) or 'feed into the computer' (in the case of computerizing a database). It seems reasonable enough that a verb derived from the noun computer' should mean something to do with computers. But now compare the variety of words in (3a-f) with the noun maxiev (3f) and the verb derived from it, mixSev. While the root ""xsb may create words with varied meanings, the noun maxSev, computer, may only create a verb whose meaning is tied to the meaning of that noUD. Although the verbs contains the root consonants, it cannot have access to the root -.Jxsb or to any of the interpretations assigned to it in other environments. Before we account for these facts and see how they bear on the status of the root, it should be noted that the same phenomenon occurs in all other cases where a verb is derived from a noun. Consider the case of the root ""xzq, which specifies a core of holding, or possessing: •
changes in verb-derivation from nouns, see Bat EI (1 994). Melodic overwriting in which the verb appears. Stem modification may resyllabify the noun so that it fits the prosodic structure of the
4 For phonological
replaces the vowels of the noun with those vowels typical of the verbal pattern pattern-including erasure of suffixes.
33
Word-Level Phases
(5)
a. b. c. d. e. f.
g.
h.
..Jxzg CaCaC CiCCeC hiCCiC CiCCa CCaCa CoCeC taCCuCa CiCCeC
adjt 'strong' 'to strengthen' Vt 'to hold' nt '(in mathematics) power' nt 'a holdt custody' n, 'strength' n, 'maintenance·5 v, 'maintain' -note t- from D. pattern
xazaq xizeq
v,
hexziq xezqa xazaqa xozeq taxzuqa tixzeq
While the root -.Jxzq may acquire numerous interpretationst the verb tixzeqt derived from the noun taxzuqa 'maintenance' t which is itself derived from -.Jxzq, can only mean 'maintain'. As in the former case, the verb derived from the noun has no access to the root, only to the noun. Other similar cases are given in below:
(6)
a. b. c. d. e.
Root 'Ixzr -.Jlkd ...Jspr ...Jqsr -.Jr;sh
noun
maxzor 'cycle' malkodet 'trap' mispar number '
'
tiqsoret 'communication' lafas;ya 'industry'
(6)
denominal verb mazer 'recycle' milked 'trap' misper 'number' tiqser 'communicate' Iifes 'industrialize'
Each of these roots creates numerous nouns and verbs . Verbs derived from a noun made from that root have access only to the noun. There seems to be some barrier between the verbs derived from the noun and the root. This barrier, I argue, is the nominal projection itself. It interferes between the root and the verbs because of the following constraint:
(7)
LocaUty constraint on the interpretation of roots: roots are assigned an
interpretation in the environment of the first category-assigning head with which they are merged. Once this interpretation is assigned, it is carried along throughout the derivation.
The generalization above holds for all noun-derived verbs in Hebrew (for a more complete account, see Arad 2003). We shall return to its theoretical status in section 4, but note, now, that if this generalization is correct, then it should hold not only for noun-derived verbs, but for any element that is derived from a word, rather than a root. This is indeed the case in Hebrew. Compare the variety of interpretations assigned to the root ..Jyld in (8) with the words derived from the noun yeled 'child' in (9):
suffixes, unlike prefixes, are not earned into the verb and may be truncated (see Bat EI 1 994 for similar cases). In this case, the suffix -0 in laxzuqa 'maintenance ' is truncated when the noun creates a verb, in order to make the stem fit into the bisylJabic verbal pattern. S Nominal
34
Maya Arad
I
(8)
c.
d. e.
�'
t r
a. b.
";yld ";yld "yld "yld ";yld
CiCoC CCuCa CCeCa CaCiC CeCeC
yilod yeluda leda yalid yeled
'newborn' 'birthrate' ' childbirth . 6 'a native' 'child'
Any heads that further combine with the noun yeled 'child' (8e), made from this root, take as their input the semantics of that noun and not of the root: (9)
a. b. c. d. e. f.
g.
"yld + CeCeC yeled + ut yeled + ut+i yeled +ut +i+ut yeled + on yeled +a yeled +hitCaCCeC
yeled yaldut yalduti ya/dutiyut ya/don yalda hityaled
n. 'child, boy' n, 'childhood' adj, 'childish' n, 'childishness' n, 'little child' n, 'girl' v, 'act like a child, be childish'
It should be stressed that when we say that the meaning of the root is fixed in the environment of the nominal pattern, it does not mean that it cannot be altered in any way. Nominal and adjectival heads combining with the noun may affect it in a number of ways: create an abstract noun out of a concrete noun (9b) or out of an adjective (9d), create a property from the noun (as does the adjectival head in 9c), form a diminutive (ge), or change gender (9t). The meaning of the root is fixed because all these heads take as their input, and operate on the basis of the noun yeled 'child', and not on the basis of the root "yld. They therefore retain the meaning of 'chiJd'-which they may affect in the ways mentioned above but have no access to the root ";yld or to any of the inteIpretations of that root in other environments. It is impossible to add any of these suffixes to the noun yeled 'child' and get, for instance, yeluda 'birthrate' (8b). Finally, the locality constraint on root interpretation works trivially with borrowed nouns or verbs. Consider Hebrew verbs that are derived from foreign words:
(1 0)
� B;:: as�e:...:f4= orm �
eb ew .: ve=r.:c. b ---,H = ,,-=z;::. � til/en 'telephone' hiqliq 'click' xarap 'snooze' jikses 'fax'
____
a. b. c.
d.
leleJon 'telephone' qliq 'click' xrop 'snooze' Jales 'fax'
Borrowed verbs are always semantically related to the word they are derived from (with some possible narrowing or specialization in meaning). This is expected-what is borrowed are words, not roots. Let us recapitulate the argument. One group of Hebrew verbs, traditionally taken to be root-derived, contains a triconsonantaJ lexical and phonological coret and shares some aspects of its semantics (e.g. mental state) and phonology (e.g. containing [xl, [8] and [b]) with other words sharing that core. Another group of verbs, independently shown to be derived from nouns, 6
Initial root consonanty drops, resulting in ledo (*yleda).
35
Word-Level Phases
must be tied in its meaning to that noun, and may not have any access to the root from which the noun is derived. or to any of the words containing that root. The difference between these groups is easily explained if we assume that nouns are decomposable into roots, and that. further, a locality constraint on the interpretation of roots such as the one in (7) holds. Still, at this point, a proponent of the rootless approach to Hebrew word formation might argue that the data reviewed so far do not point to the existence of the root. Rather, they merely show that verbs derived from nouns, such as mixsev in (3) above, must be semantically tied to that noun. But this does not mean that all the other words in (3) are necessarily derived from the root "xsb. They might as well be basic lexemes, separately listed in the Hebrew lexicon, which are not further decomposable into roots. While this is logically possible, I believe this would be missing the point. For the point is precisely the fact that words containing a certain phonological core-the consonants [x] [8] and [b]-also share some semantic core, related to mental activity or reasoning. It is not a coincidence that all the words made from this root are assigned a meaning related to mental activity, and not to directed motion, manner of speaking or change of state. Leaving the words in (3a-f) as separate lexemes, we leave their phonological and semantic relatedness as a coincidence, just like that of mist and mast in English. But the existence of these three consonants is crucial. If you replace one of them with another consonant-you get a very different lexical core. Replace [x] with [q], and you get "qsb, which creates the verb hiqsiv 'listen'; replace [b] with [q], and you get "�q, which creates the verb xasaq 'desire'; replace [5] with [r], and you get "Vxrb, which creates the verb hexriv 'destroy' . None of these verbs is related to a mental activity. The fact is that, on the one hand. the words containing the consonants [x], [§] and (b] also contain some lexical core, but, on the other hand, their meanings are quite varied. This is easily explained if we assume that the three consonants are a manifestation of the underspecified root "db, and that furthermore, this core is underspecified, it may acquire multiple interpretations in the environments of different patters. The Root Hypothesis thus enables us to capture the fact that words created from the same root are semantically related without there being a derivational relation between them. Now, if the root is indeed both a semantic and a phonological core, then we expect to find not only semantic differences between root-derived and noun derived verbs, but also phonological differences. This is indeed the case, as pointed out in the following section. ,
3. Root-derived verbs differ from word-derived verbs phonologically
The group of verbs traditional1y taken to be root-derived exhibits a number of phonological peculiarities. These peculiarities depend on the identity of the root and the phonological fonn of the pattern. Crucially, verbs made from adjectives or nouns do not exhibit any of these peculiarities. Instead, they retain the form of the word from which they are derived.
36
Maya Arad
Consider, first, assimilation of n, which was noted in section I above. When n is part of the root, it assimilates before stops and coronal fricatives and affricates:
( 1 1)
i
t
"npI "ngt;
c.
vn�q
� �
�
natal (CaCaC, 'faU '), hipi/ (hiCCiC, 'drop' ; *hinpil) naga f ( CaCaC 'touch-past'), yigaf( ' touch-f future'; otyinga I), higi fa (hiCCiC. 'reach'; *hingifa) nasaq (CaCaC, 'touch on'), hiSiq (biCCiC, 'be tangential'; *hinsiq) ,
We thus get naiaq in ( l I e), when the root consonant precedes a vowel (CaCaC), but hisiq, rather than *hiniiq, when the root consonant precedes another consonant (hiCCiC). In a theory that denies the existence of the root, such behavior of verbs remains a mystery. On the other hand, if these words have an underlying root, and if phonological rules can affect that root, then n assimilation is straightforward.7 The contrast becomes much clearer when we compare these verbs to verbs made from nouns that happen to contain the consonant n in them. In such cases n does not assimilate. The verb retains the phonological fonn of the noun it is derived from (modulo the phonological changes required for making the base fit into the verbal pattern, such as stem modification and melodic overwriting): ( 1 2)
I
a. b.
a. b. c.
Base noun
neged 'opposite' necax 'eternity'
nevel ' sprout'
derived verb
hingid 'put in opposition'. hiCCiC, cf. *higid hinciax 'eternalize', hiCCiC, cf. "'hiciax hinbil 'sprout', hiCCiC, cf. *hihi/
From neged, 'opposite', in (12a), a verb, hingid, 'put in opposition ', is derived. When put into the verbal pattern, n directly precedes the stop g, but this n does not assimilate. The consonants of the base noun, neged, are unaffected. The difference between the two groups is even more manifest when we compare minimal pairs of (alJegedly) root-derived and noun-derived verbs:
( 1 3)
a. b. c.
Root-derived hidl 'save', Vncl hib;t 'look', ..Jnbt
Noun-derived (no assimilation hingid 'put in opposition'; neged 'opposition' hinciax 'eternalize; necax 'eternity' hinbit 'sprout'; nevet 'sprout'
The contrast is clear: when n is part of the abstract root, it assimilates in well defined phonological environments. When the noun that serves as the basis for the derivation happens to contain n, this consonant is not affected, regardless of the phonological environment in which it appears.
7 The process of n assimilation is not completely productive in Hebrew, and in some cases the root consonant is not assimilated, or there exist two variants, one with assimilation and one without (Iinso fa and liso fa, 'travel', from "nsfi). What is important is that n that makes part of a noun never assimiJates-see the following.
37
Word-Level Phases
Another phonological process in Hebrew that was briefly noted above is spirantization of b, k and p in post-vocalic positions, yielding v, x and f, respectively. This process takes place precisely in those verbs that are traditionally taken to be root-derived:8
( 14)
a. b.
c.
"sbr "ktb
savor (CaeaC, 'break'), nisbar (niCCaC, 'break-passive') katav (CaCaC, 'write-pasf), yixtov (yiCCoC, 'write future') Vpxd -7 paxad (CaCaC, 'fear'), hifxid (hiCCiC, 'frighten') -7
-7
In noun-derived verbs, on the other hand, no spirantization alternation occurs. The fonn of the consonant present in the noun is retained in the verb:
( 1 5) a. b. c. d.
Base noun xrop 'snooze' faila 'flop' telefon 'telephone' koxav 'star'
derived verb xarap 'to snooze ', CaCaC, cf. *xorof fisel 'to flop', CiCCeC, cf. *piSel tilfen 'to telephone' , Ci CCeC , c[ *tilpen kixev 'to star', CiCCeC, cf. *kikev
Although the phonology of Hebrew requires spirantization after jl vowel, the verb derived from xrop, 'snooze', in ( 1 5a) is xarap, not *xaraf The mirror image of this verb is in ( I Sb): although in initial position no spirantization occurs, the verb derived from/oslo is fiSeJ, not *pisel. Roots containing middle glides (w/y) give rise to contracted verbal forms, as illustrated in ( 1 6) (note that w and y alternate): ( 1 6)
a. b. c.
d.
e.
"qwm "lewn wc xwl Vbws
�
-7 -7 -7 4 -7
qom (CaCaC, 'rise', c[ *qawom) hexin (hiCCiC, 'prepare', cf. *hixwin) hefic (hiCCiC, 'spread'; cf. *hifwic) xoleI (CiCCeC, 'cause, make happen', cf. *xiweJ) boses (CiCCeC, 'mire in', cf. *biyes)
When a glide is present in a noun, it is preserved in the verbal fonn derived from that noun, and no contracted fonn occurs:
(17)
a. b. c.
d.
Base noun xayal 'soldier' tiq 'file' bayit ' house' suq 'market'
derived verb xiyel 'conscript., CiCCeC, cf. *xolel tiyek 'file', CiC CeC cf. *toqeq bryet 'domesticate', CiCCeC, cf. *botet siveq 'market', CiCCeC. cf. *soqeq ,
A Like assimilation of n, spirantization is not always productive, and many speakers spiranlize
p
in certain positions where. according " correct" grammar, they should not (e.g.
viqai,; '1
b, k or
asked'
biqalti, xisit; 'I covered' instead of kisiri). Yet this phenomenon is rather limited. occurring mainly is initial position of the third verbal pattern, CiCCeC. but never in CaCaC (no . $ faxad for paxad 'fear') or hiCCiC (no ·hisvir for hisbir 'explain '). Most importantly, the form of instead of
word-derived verbs never changes according to spirantization rules, as illustrated in what follows.
38
Maya Arad
A similar effect exists with bi-consonantal roots-those roots containing two final identical consonants, which McCarthy ( 1 979) analyses as hi-consonantal. Such roots give rise to special verbal fonns: ( 1 8)
a. b.
.Jsbb .Jsbb
--7
c.
�rr
--7
d. e.
��
--7
--7
--7
sovev (CiCCeCt 'tum around ', cf. *sibev) hesev (hiCCiC, 'tum', cf. *hisbiv) porer (CiCCeC, 'crumble', cf. *pirer) hefer (hiCCiC, 'violate', cf. *hifrir) domem (CiCCeC, 'silence', cf *dimem)
When a verb is formed from a noun that contains two identical consonants, or when reduplication of one consonant takes place in order to make a monosyllabic noun fit into a bisyllabic pattern, the verb does not give rise to such special fonns:
( 1 9) a. b. c.
Base noun / adjective dam 'blood' basis 'basis' xam 'hot'
derived verb dimem 'bleed ', CiCCeC, cf *domem bises 'base', CiCCeC, cf *boses ximem 'heat', CiCCeC, cf. *xomem
When we compare minimal pairs, made from roots or from nouns, the contrast is particularly evident:
(20)
Root-derived
a. b.
domem 'silence', �dmm boses 'mire in ', 'l/bss
Noun-derived dimem 'bleed'; dam, 'blood' bises 'base' ; basis, 'basis'
If the underlying root is biconsonantal (..Jbss), it gives rise to a verb of a special form (boses). Verbs fanned not from roots, but from nouns containing two identical consonants (basis) do not have special forms (bises). The phonological peculiarities noted above are exhibited only by those verbs alIegedly derived from roots. There are other effects, noted by Bat EI ( 1994), specific to noun-derived verbs. All of these effects are related to the preservation of the phonology of nominal fonn, making it clear that derivation from nouns, unlike derivation from roots, must not affect the phonology of the noun any more than absolutely necessary in order to make it fit into the verbal pattern. The f1l'8t phenomenon, which was noted above, is nominal prefIxes carried into denominal verbs:
(2 1 )
a. h. c.
misgeret 'frame't 'l/sgr xe.fbon 'account', 'l/xSb taxzuqa 'maintenance' , 'l/xzq
--7 --7 --7
misger 'to frame' hitxaSben 'settle account' tixzeq 'maintain'
In addition, Bat El noted a strong tendency in noun-derived verbs denominals to appear in the pattern that best preserves the phonological fonn of the base noun:
39
Word-Level Phases
(22)
a. b. c. d.
qliq 'click' spritz splash ' jliq 'slap' xrop 'snooze' •
� � � �
hiqliq 'to click' hispritz 'to splash' hifliq 'to slap' xarap 'to snooze'
Verbs made from noun containing i appear in pattern five, hiCCiC, where th e vowel is retained even after melodic overwriting takes place. The noun made from xrop appears in pattern one, CaCaC, where its future form preserves the form of the noun (yaxrop). Finally, Bat El notes that consonant clusters in the base noun are preserved in the denominal verb:
(23)
a. b. c.
sinkroni 'synchronic' striptiz 'striptease' transfer 'transfer'
� � �
sinkren 'to synchronize' striptez 'to strip trinsfer 'to transfer' I
Such preservation effects, recall, were the initial motivation against the root based view of Hebrew verbs. Indeed the preservation of consonant clusters as in is very difficult to explain if we assume that these verbs are derived from abstract roots. It is evident that the way in which the consonant clusters in nouns are transferred into the verbs indicates a derivational relation between them. There is very convincing phonological evidence that some Hebrew
(23)
verbs are made from nouns. I fully adopt this aspect of Bat EI's account. I disagree with the second step, extending the root-less analysis to all Hebrew verbs. Because she concentrates on a sub-group of Hebrew verbs, Bat EI may be saying less about the bigger picture, that is, the fonowing three facts: First, that the preservation effect is evident only in a subgroup of Hebrew v�rbs. Second,
that this subgroup consists precisely of those verbs that do not exhibit phonological peculiarities typical of other Hebrew verbs. Third, that these verbs exhibiting phonological peculiarities are precisely those verbs that are traditionally taken to be root-derived, and which are not derivationally related to any noun. Once we look at Hebrew verbs more carefully, it becomes evident that their phonological behavior is not uniform. Verbs that are known to be noun derived show the preservation effect and no phonological peculiarities. Verbs that are traditionally called root-derived show phonological peculiarities and aie not derivationally related to any noun. Add to this the fact that these groups are, also characterized by different semantic properties, as illustrated in section 2. Those verbs exhibiting phonological peculiarities are also those which share some semantic core with other words containing the same core, or hypothetical root (e.g. -JxSb, mental state). On the other hand, those verbs that show the phonological preservation effect are also semantically tied to their base noun. To maintain the root-less view of Hebrew word formation, one would have to concede that there is some coincidence in the Hebrew lexicon-some verbs happen to have semantic and phonological peculiarities while others do not. Given the systematic nature of these correlations, and the coincidence of semantic and phonological properties in each group, it would be particularly undesirable to leave this as a matter of arbitrary property of verbs. On the other hand, the root hypothesis immediately distinguishes between the two classes and
40
Maya Arad
captures their semantic and phonological properties. Verbs made from underspecified core may be incarnated in a number of different ways, acqu several related (though varied) meanings and exhibit phonological peculiarit depending on the nature of the abstract root. Verbs made from actual nouns ta as their input not the abstract root, but the actual semantics and phonology of t noun. They are therefore semantically tied to the meaning of that noun and ha no access to the root, and do not exhibit phonological peculiarities typical roots, but rather, preserve the form of the noun. In the following section we shall see why this should be the ca� Before that, let us briefly look at roots in other languages. Recall that while , are examining the status of the Hebrew consonantal root, this topic bears on tl wider question, that of the universal existence of roots (Marantz 1 997, 2000). universally, words are made of atomic elements, then the same distincti( between root-derived and noun-derived elements should be manifest not only Hebrew. Two examples, from Georgian and Russian, illustrate this. Consider t1 Georgian root "ex and the words derived from it (example provided by Lea Nash, p.c.): (24) a.
h.
c. d. e.
acxobs namcxvari sicxe cxeli acxelebs
'to bake', v 'cake'; literally, participle of 'bake' 'fever', n 'hal', adj. 'to heat', v, from cxeli 'hoC
The words in (24a-d) are derived from the root "ex. The verb acxelebs (24e), on the other hand, is derived not from the root, but from the adjective cxeli (note the adjectival suffix -el which is carried into the verb). While the root "ex may be assigned numerous interpretations, the adjective-derived verb must depend in its meaning on that adjective. Georgian, like Hebrew, distinguishes word formation from roots and word formation from nouns or adjectives, and exhibits the same locality constraint on the interpretation of roots. The same phenomenon exists in Russian. A large number of Russian roots may acquire radically different interpretations in the environment of different prefixes. Consider one typical example, the root "kaz: (25)
t
r 1:
a. b. c. d. e. f. g.
"kaz skazat ' raskazat' otkazat '$ja dokazat ' prikazat ' okazat'sja pokazat '
'say' 'tell' 'refuse' 'prove' 'order' 'turn out' 'show'
The root may acquire many interpretations in combinations with different prefixes, but once its meaning has been assigned, in combination with the prefix, it is retained throughout all further derivation:
41
Word-Level Phases
(26)
a. b. c. d. e.
v, 'to show, indicate' n, 'an indicator, index' n, 'a showing, demonstration ' n, 'a show' adj., 'significant, revealing'
po kazat , pokazatel pokaz pokazuxa pokazalelniy
Finally, evidence from English, regarding zero-related pairs, suggests that in this language, too, there is a distinction between root-derivation and word derivation. It has been observed by a number of linguists (Hayes 1 98 1 , Kiparsky 1 982, Myers 1984, Raffelsiefen 1 993), that in zero-related pairs, different stress for the noun and verb correlates with tenuous semantic connection between the two (27). On the other hand, when the noun and the verb share the same stress pattern, they also show a tight semantic relation (28):
(27)
(28)
subjectN / subjectv, objectN / objectv, pr6j eciN / projectv, e6nductN / conductv, abstractN / abstractv . c6mbin� / combinev, exploitN / expl6itv, pr6duceN / producey, defectN / defectv. rec6rdv / recordN• permitv / permitN, convertv /convertN, c6ntestN / contestv, convictN / convictv, contractN / contractv disciplineN.V, balanc�.v, commentN.V. heraldN•v, contactN,v, d6cumentN.v
The relation between the noun subject and the verb subject, the noun combine and the verb combine, or the noun exploit and the verb exploit, is quite free. By contrast, those verbs that share the nominal stress pattern, such as comment, herald or contact, are also tightly related to the meaning of the noun. This difference is explained if we assume that in (27) the verb and the noun in each pair are derived from a common root:
(29)
a.
V
�ect
b.
N
�
fect
In this case, each element deri ved from the root is free to be assigned its own stress and its own interpretation. On the other hand, in (28), the verb is derived from the noun. It thus inherits both the phonological pro�erties of that noun (they share the stress pattern) and its semantic interpretation:9
(30)
a.
N
�
tact
b.
v
-------
V
N
-------
N
con�tact
9 I concentrate here on those English pairs where both phonological and semantic evidence exists
regarding their status as noun-derived or root-derived. See Arad (2003) for a full account of English zera.related pairs. such as lape.jiie, hammer etc.
42
Maya Arad
The relation between root-derived and noun-derived verbs in English is particularly striking in those cases of doublets-two verbs, one of which derived from a root, the other-from a noun:
(3 1)
a. b.
c. d. e. f.
permitv affixv protestv digestv compoundv contnictv
permitN affiXN protestN digestN compoundN contractN
� � � �
� �
pennitv
affixv
protestv digestv compoundv c6ntractv
The relation between the verb permit and the noun permit is close, yet none contains the meaning of the other. On the other hand, the verb permit, derived from the noun permit, inherits both the stress pattern of that noun and its interpretation: it may only mean 'issue a pennit for someone'.
4. Word-level phases Let us recapitulate the argument so far. We noticed two groups of Hebrew verbs. One group exhibits phonological peculiarities (assimilation, spirantization. contracted forms) and its members are semantically related to verbs containing the same phonological core (e.g. xasav 'think', xiSev 'calculate'). The verbs of the second group are semantically and phonologically tied to the noun from which they are derived. I suggested that the differences between the two groups are easily explained if we make the following two assumptions. First, distinguishing roots (that is, atomic lexical kernels) from words (that is, complex elements composed of roots and features). Second, postulating a locality constraint on the interpretation of roots, which poses a constraint on word formation from non-roots. Drawing on the data accumulated so far, we can refonnulate the generalization in
(32 )
(7) above as the following:
Locality constraint on the interpretation of roots: roots are assigned both semantic interpretation and phonological incarnation in the environment of the first category-assigning head with which they are merged. Once this interpretation is assigned, it is carried along throughout the derivation.
The generalization above seems to hold not only in Hebrew, but also in Georgian, Russian and English. The question arises, whether this is an ad-hoc generalization, or whether it is a result of a more general grammatical principle. Locality plays such a substantial role in all aspects of grammar, that it is possible to make this generalization follow from a number of different principles. In a lexicalist theory, where word formation is performed by the lexicon, it could follow from a lexical rule governing the structure of complex words (note that for this to work we would still need to distinguish between roots and words that is, words are not the smallest lexical elements-an assumption contrary to lexicalism). Alternatively, this generalization could be stated as a syntactic locality constraint operating in the lexicon. The word fonnation component,
43
Word-Level Phases
separate from the computational system, is still subject to the same structural constraints. The third alternative, which 1 will advocate here, is postulating a locality constraint on word formation in the syntax. That is, I suggest adopting the view known as the single engine hypothesis, according to which the lexicon contains only atomic elements (roots and features), and all computational processes, whether of word, phrases or sentences, are performed by the syntax (see Marantz 1 997 for empirical evidence for that view). While I agree that the data could, in principle, be accounted for also under different theories, as sketched above, I believe that there are two main motivations for the third view. First, note that the domain for semantic and phonological special meaning is structural. It is exactly the first category head-nominal or adjectival-that merges with the root, which defines the domain for semantic and phonological peculiarities. Second, the domains for semantic special meaning and phonological peculiarities coincide: it is the same head that defines the border between semantic peCUliarities and semantic relatedness, and between phonological peculiarities and phonological preservation. This fact, I believe, is evidence for spel1-out, or a phase (Chomsky 1999 and subsequent work; see Marantz 2000, who defends this claim for English). A phase, recall, is a stage in the derivation where the properties of the element built so far are sent to the interface level, to be interpreted. Everything that merges above the phase level does not have any access to the elements contained in the phase. Phases are normally taken to be maximal projections vP, CP (and, perhaps, DP). I would like to explore here the possibility that a phase can be determined at a lower level-by the first category head merging with the root-n, v etc. This requires some modification of the term " phase"-l will discuss this question below. Let us now come back to the Hebrew example in (3) above and see how the word-level phase hypothesis makes the locality constraint on roots in (32) follow. Once the underspecified root -.Jdb merges with the first nominal or verbal head, a closed domain, or phase, is formed, and the noun or verb is sent off to be semantically and phonologically incarnated:
____LF: __ interpretation: 'computer'
N
(33)
------
Nmaccec
-.Jdb
____ PF: pronunciation: !maxsev!
Anything that merges with that noun no longer has any access to the semantics or phonology of the root-because the phase defines a closed, inaccessible domain. Rather, the elements combining with the noun must take as their input the semantics and phonology of the noun:
(34)
Vmixsev
______
V
NIl'UI�ev
�
�
NmacccC
-.Jdb
V merges with the noun ma:dev 'frame',
and has no access to ..JxSb N defines a phase. Interpretation is assigned to the noun at LF
-7
The word-level phase hypothesis explains why noun-derived verbs must share the interpretation of their base noun. Their input is not the lexical core of the
44
Maya Arad
root, but a particular interpretation assigned to the root in a specific environment; they have no access to other interpretations of the root. This way, the existence of two groups of Hebrew verbs noted above is anything but arbitrary: root-derived verbs, but not word-derived verbs. may have a special meaning. The phonological properties of denominals are also easily explained. Noun-derived verbs cany nominal prefixes, seek to preserve the phonological form of their base noun, and do not exhibit phonological peculiarities precisely because they have no access to the phonological core of the root, only to a particular phonological incarnation of a noun made from that root. Therefore, phonological rules applying to roots (assimih.tion, spirantization) will not affect them. Bat EJ and Ussishkin are right in their claims regarding borrowed and denominal verbs: these verbs are indeed created through modification of existing words, rather than from roots. This is exactly what we predict, given the phase hypothesis: the nominal projection creates a barrier for special meaning and special phonology, so that borrowed and denominal verbs cannot have access to the root. But, crucially, this analysis cannot be extended to all Hebrew verbs. Clearly, there are semantic and phonological differences between noun-derived and root-derived verbs that make a unified analysis unlikely. lo Before concluding, let us return to the question touched on above, regarding the word-level phase hypothesis. Phases seem to play a double role in current theory. On the one hand, they constrain movement, in a way somewhat reminiscent of barriers (Chomsky 1 986): only certain projections may be phases ( or barriers), and the phaseJbarrier may be bypassed under specific conditions. On the other hand, phases are defined as points of semantic and phonological spell-out. The case discussed here sharpens the contrast between the two supposed properties of phases; clearly, the first category head merging with a root does not have any effect on movement. On the other hand, there is plenty of evidence that this head serves as a point of semantic and phonological spell-out. At this point it is precisely such work on the nature of phases, incorporating data from a variety of languages and concentrating on both types of properties, that is needed in order to clarify their status. The point of this paper is to suggest that if we take seriously the idea that phases are points of spell out, then the first category head merging with the root clearly answers the definition of a phase. I I
10
Pesetsky (1 982) points out certain cases in Russian where the locality constraint above seems to be
violated. that is, a suffix seems to
be
able to "look back" into the opaque domain defined by the
combination of the root with another suffix. An example is sud' 'judge' -+ sud'ba 'fate' -+ sudebnyi 'related to court'. Note, however, that it is not necessary that sudebny; is derived from sud'bo, and
not directly from sud', with the
-eb being pleonastic.
More regular derivation in Russian, such as the
combined with its prefix, is subject to 2003): vkaz -+ pokazat ' 'show'. dokazaf '
secondary imperfective that is derived after the verb root has the locality constraint on the interpretation of roots (Arad
'prove' , zakazat' 'order', prilaat' 'command' -+ pokanvat ' zakazivo, ' ·order-impf.·, prikazival' ·command-impf.'.
'show.impf.'. dokazivot ' ·prove-impf. ·,
the status of the first head merging with the root. Is for the first merge, or is every application of merge cyclic. At least as far as
II A question that arises at this point regards
there any specific status
Hebrew is concerned, it seems that the first category head merging with the root has special properties, and that further heads merging with the noun or verb do not have an effect on
semantics or phonology of the derivation in
the same way (see Arad 2 005
for verb-derived verbs).
the
45
Word-Level Phases
5.
Summary
The Hebrew data presented here strongly support the Root Hypothesis and the word-level phase hypothesis. In a theory that denies the existence of the root, there is no way to explain the differences between the two groups of verbs those semantically and phonologically related to a noun and those semantically .and phonologically related to an abstract consonantal core. The different seman tic and phonological behavior of these two groups remains a coincidence. If we adopt the Root Hypothesis, the difference between the two groups becomes obvious: the first group consists of verbs made from roots, that is, underspecified semanti c and phonological cores. The second group consists of verbs made from nouns, that is, specific semantic and phonological incarnations of roots. Adding the word-level phase hypothesis to the Root Hypothesis gives us a principled reason why verbs made from nouns may not have any access to the root below the nominal projection, either semantically or phonologically. The semantic and phonological characteristics of each group are immediately derived. Both the Root Hypothesis and the word-level phase hypothesis are currently contested. It is through more research into the nature of semantic and p honological properties of roots and into the nature of spell-out, as well as through wider scope of empirical evidence, that their status can be corroborated or refuted. The goal of this paper is to provide such empirical support for both hypotheses.
References Arad, Maya (2003). Locality constraints on interpretation: The case of Hebrew denominal verbs. Natural Language and Linguistic Theory 2 1 : 737-778. Arad, Maya (2005). Roots andpatterns: Hebrew morpho-syntax. Dordrecht: Kluw er. Bat
EI,
Outi (1 994). Stem modification and cluster transfer in Modem Hebrew. Natural
Language and Linguistic Theory 12: 571-596. Bat EI, Outi (200 1 ). In search of the roots of the Root. Paper presented at the Root and Template Morphology Workshop, University of Southern California. Chomsky, Noam (1 986). Barriers. Cambridge, Mass.: MIT Press.
Chomsky, Noam ( 1 999). Derivation by Phase. MIl' Occasional Papers in Linguistics 1 7, MIT Working Papers in Linguistics.
Halle, Morris, and Alec Marantz ( 1 993). Distributed Morphology and the pieces of inflection. In Kenneth L. Hale and Samuel J. Keyser (cds.), The new from BU ilding 20. Cambridge, Mass.: MIT Press, pp. 1 1 1 - 1 76. Hayes, Bruce ( 1 98 1 ). A metrical theory of stress rules. Doctoral dissertation, MIT.
Distributed by Indiana University Linguistics Club. Kiparsky, Paul ( 1982). Word formation and the lexicon. In Fred Ingeman (ed.), Proceedings of the Mid-America Linguistics Conference. University of Kansas. Marantz, Alec ( 1 997). No Escape from Syntax: Don't Try Morphological Analysis in the Privacy o f Your Own Lexicon. In Alexis Dimitriadis et at. (eds.), Penn Working Papers in L inguistics 4.2: Proceedings of the 2P' Annual Penn Linguistics Colloquium. Department of Linguistics, University of Pennsylvania. Marantz, Alec (2000). Roots: The universality of root and pattern morphology. Paper presented at the Conference on Afro-Asiatic Languages, University of Paris VII.
46
Maya Arad
McCarthy. John ( 1 979). Fonnal problems in Semitic phonology and morphology. Doctoral dissertation, MIT. Distributed by MIT Working Papers in Linguistics. Mye rs, Scott ( 1 984). Zero Derivation and Inflection. In Margaret Speas and Richard Sproat (cds.), Papers from the January 1984 MIT Workshop in Morphology. MIT Working Papers in Linguistics 7. Pesetsky. David ( 1 982). Russian mOlphol ogy and lexical theory. Ms., MIT. Raffelsiefen, Renate ( 1 993). Relating Words: A Model of Base Recognition, Linguistic Analysis 23: 3-159.
Ussishkin. Adam ( 1 999). The inadequacy of the consonantal root: Modem Hebrew denominal verbs and Output·Output Correspondence Phonology 1 6.3 : 40J-
442.
.
Ussishkin, Adam (2001 ). Fixed prosody and the epiphenomenon of root·and.pattem morphology. Paper presented at the Workshop on Root and Template Morphology, University of Southern California.
Center for the Study of Language and Infonnation 220 Panama Street, Stanford University Stanford CA 94305 USA marad@Stanfordedu
47
j j j
j j j j
j j
j j j
j j j
j j j
j j
j j j
j j j
j j j
j j
j j j
j j j
j j j
j j j
j j
j j j
j j j
j j j
j
j
f."
\1.
�.
}
y
Split-DPs, generalized EPP and visibility*
Alastair Butler, University ofAmsterdam, Eric Mathieu, University College London We ask why the folJowing statements hold: (1) the EPP feature associated with C is universal, i .e. visibility is needed; (if) pied-piping is required when an intervener is present in a split-DP construction; (iii) pied-piping is still possible when no intervener is present Our main claim is that the visibility requirement in the C domain has a functional role: it ensures that an overt signal is made to the effect that the speaker's utterance should be interpRted as an information question. Second, the nominal in a split-DP c onstructi on must sometimes raise to the left field domain to meet interpretability requirements. Finally, syntax is oblivious to when such movement is needed, and so thc gtammar leaves this as an available option.
1.
Introduction
From a minimalist perspective, split-DP constructions like (1a) are interesting because, a priori, it is not clear why the nominal can also pied-pipe as in (1 b).
(1)
[DP Ci de fivres]]7 of books
a.
[cp CombieDi
h.
[cp [DP CombieD de livres]i as-tu
as-tu Iu how-many have-you read
Ius Ci] how-many of books have-you read-AGR 'How many books have you read?'
The reasoning is as follows: the EPP is a general uninterpretable feature requiring visibility to be erased (Chomsky 2000, 200 1 ). Visibility is (minimally) achieved in the split alternative (la) by movement of the bare operator. Being uninterpretable, the EPP feature is not accessible to the computation once eliminated. Economy should thus block the pied-piping of the nominal in ( l b), since it is not required for convergence. Let us call this the Pied-piping puzzle. We will relate the possibility of pied-piping in (1 b) to the fact that scopal elements like negation block the relation between the bare operator and the stranded nominal in the split alternative, as the contrast of (2) illustrates (cf. Obenauer 1 976, 1 983. 1994; Rizzi 1990).
·We wish to thank Paul Dekker, Klcanthes Grobmann, Hans Van de Kool. the anonymous reviewer, and the participants of the workshop for their questions and comments.
MIT Working Papers in Linguistics 49. 49-67 Perspectives on Phases © 2005 Alastair Butler and Eric Mathieu
Split-DPs, generalized EPP and visibility (2)
a. ?*[cp Combien; n'as-tu pas how-many NE-have-you not b.
Iu [tIP e; de livres]]? of books read
Combien de livres]; n'as-tu how-many of books NE-have-you 'How many books have you not read?'
[cp [op
pas not
Ius e;p' read-AGR
We will argue that the full DP raises to avoid a situation where the Logical FOnD representation cannot be evaluated. In other words, pied-piping in ( l b) and (2b) is semantic Last Resort to avoid crashing at LF. Such reasoning seems weU motivated for (2b): pied-piping is the only option. However, this does not tell us why pied-piping should also be okay in (lb), where: (i) pied piping is not the only option, and (ii) pied-piping is arguably the more costly option. To motivate the pied-piping in (lb), we will argue that syntax is blind to the semantic constraints responsible for the contrast of (2). Thus the grammar allows vacuous movement 'just in case'. In sum, syntax is flexible enough to cater for the needs of semantics. These ideas are implemented within the framework developed in Butler and Mathieu (2004). This ties a Predicate Logic based semantics to a Minimalist syntax. The paper is organized as follows. Previous solutions to the Pied piping puzzl e are introduced and rejected in section 2. Then we tum to address the paper's three central questions: in section 3, we ask why (i) the EPP feature associated with C is universal (or in earlier versions of Minimalism, why the C related D feature is universally strong, Watanabe 1992, Chomsky 1 995); in section 4, we ask why (il) pied-piping is required when an intervener is present; and in section 5, we ask why (iii) pied-piping is still possible when no intervener is present. With section 6 we conclude the paper. 2. Possible Solutions to the Pied-piping Problem 2.1 Distributed deletion
Let us first consider a solution to the Pied-piping puzzle in tenns of partial deletion as advocated by Fanselow and Cavar (2002) (henceforth F&C). Assuming Chomsky's (1 995) approach to movement, where raising of elements in the phrase-marker leaves copies behind which are subsequently deleted, F &C argue that the deletion operation may affect both copies (hence the term 'distributed deletion'). F&C claim that this is the best way to account for split XPs. Under a strict deletion theory of movement, either the upper or the lower copy deletes. F&C argue that a strict theory cannot be correct, as there are apparent constructions where both copies are pronounced. For example, there are so-called 'Copy-constructions,' where both WH-phrases are spelled out (cf. Hiemstra 1 986):
(3)
50
[cP 1
Wer denlest du denn [en wer du bist]]? who think you that who you are 'Who do you think you are?' (Fanselow and Cavar 2002:83)
I
Alastair Butler and Eric Mathieu
Moreover, certain resumptive pronouns can be viewed as reflecting the failure c copies of movement to delete completely (cf. Pesetsky 1 998. Bianchi 1 995 Boeckx 2003, among others). For example, Egyptian Arabic (4) illustrates eithe left-dislocation or relativisation depending on whether or not thl complementizer illi 'that' is present (cf. Demirdache 1 997). 1 (4)
iI-walad (illi) Laylaa �aafit-ub imbarih def-boy (that) Laila see -him yesterday 'The boy (that) Laila saw him yesterday.' (Demirdache 1 997: 1 99)
Accepting such evidence, F&C argue that it is entirely feasible for some, but not all, of the upper copy to delete while similarly some. but not all, of the 10wer copy deletes. Such an idea seems attractive: it offers a unified account of both PP-extraction as in (Sa) (Serbo-Croatian) and noun extraction as in (6a) (Gennan). (5)
(6)
a.
[cp Na jaki Marek dach on what-kind Marek roof
b.
[a Na jaki d ach Marek kociL]? on what-kind roof Marek jumped 'On what kind of roof did Marek jump?'
a.
[cp BOcber habe ich keine gelesen]. books have I none read 'As for books, I haven't read any.
kociL]? jumped
•
b.
[cp Keine Biicherj babe ich Ci gelesen). none books have I read 'No books, I have read. '
(7) and (8) give the derivations for (Sa) and (6a) respectively:
(7)
(8)
a.
Marek na jald dach kociL?
b.
Na jald dach Marek D a jaki dach kociL?
c.
Na jaki (dach) Marek (na jaki) dach kociL? (distributed deletion)
a.
Ich babe keine Bucher gelesen.
b.
Keine BOcher babe ich keine BUcher gelesen.
c.
(Keine) Bucher babe ich keine (BUcher) gelesen.
(movement)
(movement) (distr. deletion)
I Note that Demirdachc's analysis of resumptive pronouns is rather different from Pesetsky's and
Boecla's. On her view a resumptive pronoun is the in-situ counterpart of a null-operator.
51
Split-DPs, generalized EPP and visibility F&C further suggest that in split-DPs. a OP or a PP is split up in cases where its phonetic material is nonnally linked to at least two different pragmatic features. On the assumption that focus is checked in a specific position, they argue that the two spread elements split up because the feature that they each bear cannot be checked in the same position. One bears a +WH or a +TOP feature and is checked in a higher interrogative position, while the other carries a Focus fearore that is checked in a lower focus position. (9) and ( 1 0) illustrate the two options.
(9)
[F1"WH Na jaki [+WH] Marek [F+Foc what-kind on Marek 'On what kind of roof did Marek jump?'
( 1 0)
[F+TOP Bucher [+TOP] babe ich [F+Foc books have I As for books, I haven't read any.'
dach [+Foc] kociL]]? roof jumped
keine [+Foc] gelesen]]. none read
•
In cases where full movement is instantiated, the idea behind F&C's proposal is that there is only one feature involved, namely the +WH feature for the case of (9) and the +TOP feature for the case of( 1 0). 2.2 Remnant movement
Another possible account for the existence of split-OPs is a remnant movement account (see Starke 2001 and Kayne 2002. for a remnant movement approach to split com bien constructions). ( I I) is an abstract representation of what remnant movement may consist of. First, an element X moves out of a constiroent. Second, the whole constituent, which now contains an empty element (i.e. the trace of X), raises to its designated position.
Along these lines, Androutsopoulou (1 997) argues that. in split adjectival constructions in Modem Greek the nominal raises to a topic position (higher than IP) and then the whole complex containing the trace of the nominal raises to its dedicated position. A variant of the remnant movement approach (3 la Poletto and Pollock 2004) would postulate movement to Spec-CP of combien de livres just like the case of the non-split structure, and then subsequent raising of the whole IP. This is illustrated in ( 1 2). (12)
Combien [IP as-tu lu ei] de livresi ej ej ?
There are good reasons to believe that the remnant approach to split-DPs is not on the right track. In the following section, we give arguments against such an approach. The problems identified will also hold for the distributed deletion analysis.
52
Alastair Butler and Eric
Mathieu
2.3. Problems with these approaches
There are a number of crucial differences between the split variant and the full movement alternative. We identify six main differences that we dub problems: (i) the scope problem; (ii) the agreement problem; (iii) the adjunct problem; (Iv) the thematic ranking problem; (v) the reconstruction problem; (vi) the stress problem. Each of these is a problem, since it is wtclear how approaches based on either distributed deletion or remnant movement can account for them. First, Van Geenhoven ( 1 998) shows that the scope of the split nominal in a split-topic construction is fixed. For example, (1 3a) shows that the indefinite cannot achieve wide scope over negation. Rather than being unambiguous, (13a) is ungrammatical: einige 'some', being a positive polarity item, cannot take scope inside negation. On the other hand, (1 3b) is ambiguous: Schwarze Spinnen 'black spiders' can scope either wtder or above negation. ( 1 3)
a.
?* [0' Schwane Spinnenj hat Lisa im Keller einigee; black spiders has Lisa in-the cellar some nicht gesehen]. not seen As for black spiders, there are some that Lisa has not seen in the cellar: •
b.
Lisa hat im Keller einige schwarze Spinnen nicht Lisa has in-the cellar some black spiders not gesehen. seen
'Lisa has not seen some black spiders in the cellar. ' 'There are some black spiders that Lisa has not seen in the cellar.' In addition, the splitipartiaJ movement construction in (14a) receives only a pair-list interpretation, while the full movement construction in (14b) receives both a pair-list and an individual reading. (14)
a.
hat jedes Kind ffinf e; geseben]. cats has every child five seen 'As for cats, every child has seen five such animals. ' # 'As for cats, there are five such cats such that every child has seen them. '
b.
[cp FOnf Katzenj jedes Kind ej hat gesehen]. five cats every child has seen 'As for cats, every child has seen five such animals. ' 'As for cats, there are five such cats such that every child has seen tbem. ' (Van Geenhoven 1 998: 1 25)
[cp
Katzenj
53
Spi i t D Ps genera li zed EPP and visibility -
.
With split combien con struction s, scope interaction with universal qu anti fi ers is also l imited. De Swart (1992) notes that the no n-split ( 1 5) has two distinct readings in which (i) the u niversal scopes over the WH-phrase, and (ii) the WH phrase scopes over the universal. Reading (i) asks of each persor. how many books they read. This is the pair-list reading: 'John read 3; Mary, S ; Peter. 7.' Re ading (ii) asks for a single number, i.e. 'how many books are such that everyone read them?' This is the individual reading.
(15)
[ep [IP
Com bien de fivres]i ont-ils how-many of books have-they 'How many books have they all read?'
tous
all
Ius
read-AGR
Ci]?
(de Swart ] 992: 403)
In contrast, the split construction equivalent ( 1 6) has only reading (i). De Swart calls this a scope island: the WH-phrase is incapable of taking wide scope.
(16)
[ep
Combieoj ont-ils tous lu how-many have-they all read 'How many books have they all readT
[Of> ej
de livres]]? of books (de Swart 1992:403)
Second, there is a difference in past participle agreement between the split variant and the full movement version (on French past participle agreement, see Obenauer 1 983, 1994; Kayne 1 989; Deprez 1 998). This is illustrated in (17). It must be noted that there is dialect variation with regard to past participle agreement i n French: not every French speaker accepts agreement. However, what is clear is that no agreement is possible when the WH constituent is split.
(17)
a.
b.
[ep [Ill'
Combien de boites]i how-many of cans
as-tu ouvertes ei]? have-you opened-AGR
Combieni as-tu ouvert*es how-many have-you opened-* AGR 'How many cans have you opened?
[ep
[[I' ej
de boites]]? of cans
'
In a split construction, if the nominal were to raise to the specifier of a topic position (as suggested by Androutsopoulou 1997), then the nominal would have to go through the Spec of AgroP (or the outer specifier of vP in more recent versions of Minimalism), movement being cyclic. But we have evidence against the nominal moving through Spec-AgroP: no agreement shows up, implying that the nominal does not raise out of the DP. Aware of this problem, Starke (2001) claims that combien is inherently Case-marked in split combien constructions. Generally, inherent Case does not trigger agreement. But this claim is strongly stipulative: no independent evidence is given for the idea that combien is inherently Case-marked. Third, on the distributed deletion account, it remains a mystery why some PPs can split, but not others. In (l 8b) extraction of the indirect object a combien 'to how many' is possible. but in (19b) extraction of en combien 'in how many' is not (we owe this example to Sophie Heyd).
54
Alastair But1er and Eric
( I 8)
(19)
a.
[cp [PI'
b.
[cp
A combien de personnes]i to how-many of people
as-tu ecrit ei ]]? have-you written
A combieni as-tu ecrit [op � to how-many have-you written 'To how many people have you written?'
de personnes]]'
of persons
[cp [pp En combien d'annees]i as-tu fini 18 these e; in how-many of-years have-you finishedyour thesis
a.
b.
Mathieu
...
[cp En combieni as-tu fini ta these [IP e; d'annees]]? in how-many have-you finished your thesis of-years 'How many years have you taken to finish your thesis?'
If the split alternative is simply the equivalent of full movement, full movemer being masked by independent processes made available in the grammar, theu these contrasts are not expected The difference between (I 8b) and (19b) has to do with differences in thematic relations. What raises in (19b) is an adjunct, whereas it is an argument that raises in (1 8b) (for other contrasts of this sort in split combien constructions and an account of these facts, see Mathieu 2004). Fourth, whereas direct objects can be split freely. as shown in (20b), indirect objects can be split only if no lower ranked noun (e.g. a theme) is present in the sentence. Compare (2 I b) and (22b) (see Mathieu 2004 for details). (20)
(21)
(22)
a.
donne a [cp [IP Combien de livres]i as-tu how-many of books have-you given to
b.
donne [op e; de livres [ep Combieni as-tu of books how-manyhave-you given 'How many books have you given to Jean? '
a.
[cP [oP De coin bien de livres]i of how-many of books
b.
besoin [pp e; de Bvres]]? [cp De combieni as-tu of how-many have-you need of books 'How many books do you need?'
a.
[ep [I'P A combien de personnesli as-tu donne to how-many of persons have-you given
Jean Jean.
e;]?
a Jean]]? to Jean
as-tu besoin Cj]? have-you need
un livre e;p' a book
55
Split-DPs, generalized EPP and visibility
b.
* [cp A combieni as-tu donne un livre to how-many have-you given a book
[pp � de personnes]]? of persons
'To how many people have you given a book?' Fifth, reconstruction phenomena provide us with direct evidence that the nominal in a split construction is not in the Comp area. In (23a) coreference between Jean and the pronoun if is possible (the R-expression is in an adjunct), whereas in (23b) it is not. (23b) is a Principle C violation. (23)
a.
[ep [DP Combien how many -
avait had
b.
* [cp
IIi
he
acbetes1
baught-AGR regrette
regrets
[DP combien
how-many
des
des
of-the
tableaux que paintings that
regrette-t-ili regrets-he
Jeaoi
Jean
avair have-INF
revendus Cj]?
resald-AGR
avair
revendu have-INF resold tableaux que
of-the paintings that
Jeanl avait achetes]?
Je�n had
bought-AGR
'How many of the p aintings that Jeanj had bought did hei regret having sold again?' As shown by (24), (23b) is ungrammatical because of the coreference indicated by the coindexation of Jean and the pronoun ii, not because the question is unavailable independently: (24)
[ep
II regrette avoir revendu he regrets have-INF resold
que
Marie avait
achetes]?
that
Marie
bought-AGR
had
tableaux combien des how-many of-the paintings
'How many of the paintings that Marie had bought does he regret having sold again? '
As shown by (25), in the split version, coreference between Jean and it is impossible. We take this fact to indicate that the nominal complex is not in the Camp area. Rather, it is in-situ below the VP area (see Cecchetto 1999 for arguments of this sort in defense of a non-remnant movement approach to Clitic Right Dislocation).
56
Alastair Butler and Eric Mathieu
(25) *[cp Combien; regrette-t-il; how-many regrets-he
revendu avoir have-INF resold
des tableaux que Jean, avait achetes? of-the paintings that Jean had bought
'How many of the paintings that Jean; had bought did hCi regret having sold again?' Finally, under a remnant movement approach, it is not clear what drives movement of the nominal. It is unlikely that it raises to a topic position (as argued for Greek by Androutsopoulou 1 997) because the nominal is clearly not interpreted as 'presupposed', 'given' or 'old'. Instead, it introduces a novel variable. A question such as (26) can be answered in the negative. (26)
A.
Com bieni as-tu lu Ci de livres? how-many have-you read of books 'How many books have you read?'
B.
Aucun. none 'None.'
To swnmarize. neither the distributed deletion nor the remnant movement analysis can account for why there are important differences between a full movement and a split variant. One natmal way to account for the difference between a split and a non-split construction is in terms of movement of a bare operator with stranding of the restrictive material versus pied-piping. Before we tackle the question of why pied-piping of the nominal should ever be possible on minimalist grounds, we tum to the first main issue of the paper: why the EPP feature associated with C is universal (or in earlier tenns, why the C-related D feature is universally strong). 3. Visibility
While methods for checking the EPP feature associated with QIWH vary cross linguistically, it appears to be the case that such a feature must somehow be checked in all interrogatives in all languages (cf. Watanabe 1 992; Chomsky 1 995). Recently, this condition has been encoded in the grammar as follows: the EPP is a general uninterpretable feature requiring visibility to be erased (Chomsky 2000, 200 1 , see also Plat7ack 1 998 who explores the consequences of a visibility condition on the C domain in Germanic languages and Italian). Obvious examples of such checking/visibility satisfaction are by Merge: in-situ languages like Chinese have Q particles that can be base generated in the C domain; and Move: in a language like English one question word raises to Spec-CPo Since French does not have question particles with constituent questions� it is usual for material to move and thereby check the EPP feature in C. That is� by default a WH element should raise. The case of split combien 57
Split-DPs, generalized EPP and visibility
constructions is a good illustration of this. However, it is also possible for no morphological element to be present in Spec-CP, as (27) illustrates. (27)
Elle a vu qui? she has seen who 'Who has she seen?'
In a case like (27), a special intonation is required instead. We will suppose that this special intonation satisfies the visibility requirement. More specifically, we can suppose that the special intonation pattern of (27) signals the presence of a null operator in Spec-CP that checks the EPP feature aosociated with QIWH. In the type of French described in this paper, the special intonation pattern is associated with deaccenting, not with heavy stress on the WH word as in Cheng and Rooryck (2000). 2 The symbol L% in (28) stands for a low boundary tone; it illustrates the fact that a French WH phase in-situ has a falling pitch movement in the contour associated with the utterance (for similar phonological effects, see Oiry 2004). Deaccenting is also argued to be a WH-scope marking strategy in Japanese (Ishihara 2002). (28)
[Spec-CP Op Ell'a vu qui?] L%
This predicts a split configuration in French that is much like the split configuration we find with split combien constructions. Evidence to support this view comes from the fact that such in-situ examples share the syntactic properties of split combien constructions: they give rise to intervention effects (see (37) and (40) below, as well as Mathieu 1 999, 2004; Butler and Mathieu 2004 for details). It fonows that, in French, visibility is not only achieved morphologically but is also achieved phonologically. Of course the fact that visibility is always achieved does not tell us why there should be a visibility requirement associated with C. We will suppose that this visibility requirement has a functional role: it ensures that an overt signal is made to the effect that the speaker's utterance should be interpreted as an information question, as opposed to a reprise question. A reprise question comes in two bnu:tds: (i) an 'echo' question as in (29b); and (ii) a 'reference' question as in (30b). Bolinger ( 1 987) introduces the tenn 'reprise' for both uses and this is also the term used in Ginzburg and Sag (2001). This term captures the fact that in order to determine the meaning of such a question, one must in some sense have access to the preceding utterance. Engdahl (200 1) uses the term 'reprise' as the general term for context dependent
Z Cheng and Rootyck's account
is based on the idea that the intonation in (27) is comparable to
that found in a yes-no question like C'est un livre? 'Is it a book?'. Both types of questions are
claimed to have a rising contour. Their idea is that WH phrases in situ are licensed by the yes-no question operator; this applies to both French and Chinese. The difference between French and
Chinese is that in French WH feature movement is necessary to set the value of the Q morpheme, which is otherwise underspecified, to QIWH.
58
Alastair B utler and Eric Mathieu questions. [n (29b), B asks for clarification of what A said. In (30b). B asks fOI further infonnation concerning the intended referent of them. (29)
(30)
Mary is
going to visit [inaudible]
a.
A:
b.
B: She is going to visit WHO?
a.
A: Mary is going
b.
B: She is going to visit WHO?
to
visit them.
What is crucial is that interrogative reprises must contain an accented WH phrase. This differs from ordinary in-situ information questions in French, where an accent on the WH phrase in-situ is impossible. Instead, as already noted, a falling pitch movement in the contour is obligatory. Finally, we note that in the case of split combien constructions, combien checks or saturates the relevant feature to satisfy the visibility requirement.
4. Last Resort and Semantics
We now tum to the second question with which this paper began: Why is pied piping required when an intervener is present? We take as our starting point the dynamic semantics stance on inteIpretation offered by, among others, van Rooy ( 1997), Stalnaker ( 1 998), Zimmermann (l 999). Kamp ( 1 990), and Dekker (2002). This starts with the observation that, for a discourse to be interpretable, information pertaining to the discourse itself is needed. We will caU this usage information. An archetypal example is the contrast between (3 1 ) and (32) (due to Partee). (3 1 )
I dropped ten marbles and found all o f them except for one. I t i s probably under the sofa.
(32)
] dropped ten marbles and found only nine of them. nIt is probably
,
under the sofa.
The first sentences i n (3 1 ) and (32) are truth-conditionally equivalent: they provide the same infonnation about the world. But unlike (3 1 ), the first sentence of (32) does not give sufficient information to aJlow for the resolution of the anapboric link. To capture the distinction between (3 1 ) and (32). we can suppose that the occurcren e of one in the first sentence of (3 1) comes with an 3 usage instruction. This makes available an 'intentionally present' individual (i.e. the lost marble) that the pronoun in the second sentence can take as its referent. In contrast, (32) is bad because no such intentionally present individual is made available. Note that the referential intentions associated with the use of indefinites like one in (3 1) are to be attributed to the speaker, who is supposed to be able to support what she says. The hearer has no such requirements, and so
59
Split-DPs, generalized EPP and visibility
can generally take an 3 occurrence to introduce a new subject. Now consider the speaker' s support for (33).
(3 3)
There i sn ' t a tiger in the cage.
This can be characterized as a ban on an update of her state with (34). (34)
A tiger is in the cage.
Consequently, the requirement of referential intentions is replaced by the requirement to have evidence that she, as a hearer, bans any update with (34), no matter who, with whatever intentions, would try to attempt to bring about such an update . This situation arises because the 3 usage instruction a tiger makes available is assumed to only be visible from withi n the scope of the negation. From outside negation's scope, 3 is opaque, accounting for the absence of referential intentions, whi ch in tum accounts for the absence of any potential anaphoric pick up. We wish to claim that there is a related visibility requirement on the C domain for interrogative sentences to meet. In this respect, the approach we advocate is in line with the existential disclosure approach of Honcoop (1 998). We propose that this visibility requirement arises because cross-linguistically interrogatives come with a bare interrogative operator Q base generated as the highest element. The reader is referred to Butler and Mathieu (2004) for the teclmical details. Thus all interrogatives give rise to split configurations. To have values under question to impart to variables, Q is taken to rely on the presence of 'wh' usage instructions. Whenever WH-phrases cany wh , they must be in a visible relation with the C domain, in the sense of not falling underneath a scopal operator, for Q to function as a binder of WH variables. This explains why (35) and (36) are okay. (35)
(36)
Op Qui,' est-ce que tu Q wh is-this that you Q x yOll_did_not_see(x)
n'as
NE-have
vu pas scopal-op seen
O p Tu as vu qui? Q you have seen wh Q x you_saw(x)
It also explains why (37) is bad, with Q left to ask a question without the support required to bind any variables. (37) * Op Tu n'as pas vu qui? Q you NE-have scopal-op seen wh Q you_did_not_see(x) That is, (37) i s incoherent. On the one hand, because of the intervening scopal operator, it fails to show any wh usage information, and thus no values under question are introduced (hence the plain Q in the interpretable representation), and on the other hand, it really does have a WH-phrase primed to receive values
60
Alastair Butler and Eric Mathieu
under question (the free x). The approach generalizes as in (38) and can be see: as a rationalization of Pesetsky's (2000) Intervention Effect Condit ion and othe related proposals. (38)
Op [Motrix
•••
( * scopal_op)[ ... [R=sIriclion
•••
usage-instruction . . . ] . . . ] . . . ]
As a further illustration of the configuration in (38), we will g' through the details of the split combien construction in (39), and contrast thi with the all-fronted option of (40).
(39)
(40)
Combienj as-tu lu Cj de livres? how-many have-you read of books 'How many books have you read?' Combien
de fivresi
as-tu
Ius
Cj?
how-many of books have-you read-AGR 'How many books have you read?' In addition to their overt structures, the two options differ in terms of the locality effects to which they are susceptible. Notably the split option displays intervention effects, as (41 ) shows, while the all fronted option escapes such effects, as (42) illustrates. pas Iu (41 ) *Combienl n'as-tu how-many NE-have-you not read 'How many books have you not read?'
(42 )
�
pas de livresi n'as-tu how-many of books NE-have-you not 'How many books have you not read?' Combien
de livres? of book s
Ius Cj? read-AGR
To account for this difference we will analyze combien de DPs as foll ows :
(43)
[IP combien de livresJ := [3y[ wh x = IYI] books(y)]
That is , combien de livres is taken to introduce a set of books y, the number of which x is brought into question by the presence ofcombien. By itself, combien is taken to contribute:
(44)
combien ;= [wh x = tvl]
Thus, combien is not a question operator. Rather it contributes a usage instruction wh (providing the information that the sentence contains at least one WH-pbrase), and restrictive material x = tvl (the condition that brings into question the number of books).
61
Split-DPs, generalized EPP and visibility
4.1 The split option Combining our analysis of combien de DPs in (43) with our assumption that all interrogatives come with a null Q operator gives (39) the following LF:
(45)
[cp Op Combien i as-tu lu [IJI' � de livres]]? Q [wh x= IYI) have-you read [3y books(y)] Q x [wh x = IY/](you_read(y) [3y books (y»))
Here, Q has visible information pennitting it to bind a variable. Nevertheless, the LF is bad: combien's movement leaves an occurrence of y outside its binder's scope. The situation is salvageable if the restrictive content of combien reconstructs. We indicate this by splitting combien into 'eembieft' (the residue of movement) and 'combien' (all the restriction material), as in (46). Notably, in (46), all y occurrences are in the scope of their binder.
(46)
[cp Op ealBhien as-tu lu [a' com bien de Iivres]]? have-you read [3y[wh x = lYll books(y)] Q a x (you_read(y) [3y[wh x = IYIJ books(y»))
This analysis tells us why intervention effects occur. (41) is ruled out because reconstruction of combien's restrictive content is forced for the same reasons that it was forced in (46). As (47) shows, this has the undesirable consequence of placing at LF the usage infonnation that comes with combien Wlder the scope of the intervener pas 'not.' Thus, (47) is out for the same reason as (37): there are no visible occurrences of wh from a, and so a is not permitted to bind any variables, with the result that x fails to denote. pas lu (47) "'Ccp Op eamhien n'as-tu [DI' combien de livres]]? NE-have-you Neg read [3y[wh x = IYI]boohy] Q a Neg (you_read(y) [3y[wh x IYI]books(y)]) =
4.2
The all-fronted option
We continue with the same assumptions for the all-fronted option. The difference from the split option is that this time the whole combien de DP moves to the clause initial position. As a consequence, in (48), which is the LF for (40), y 's binder moves with combien. This removes the need for reconstruction, with all variables being bound (Q is able to bind x thanks to the visible wh).
(48)
[0' Op [a> Combien de livres]i as-tu Ius Cjp. Q [3y[wh x = Iv\] books(y)] have-you read-AGR a x[3y[wh x = IYI1 books(y)] youJead(y)
It also gives (42) the LF (49). which is likewise interpretable, with the essential usage infonnation wh occurring in the matrix Spec-CP position and hence outside the scope of the potential intervener pas 'not·.
62
.
I1"
Alastair Butler and Eric Mathieu
(49)
[0> Op [IP Combien de livres]i n'as-tu Q [3y[wh x tvl] books(y)] NE-have-you a x[3y[wh x tvl] books(y)] Neg you_read(y) =
pas
Neg
Ius
�]?
read-AGR
=
4.3 Scope Effects Split combien constructions provide a particularly c1ear illustration of the need to account for intervention effects in terms of scope. Consider (50). (50)
tous lu read all
a.
Combieni ont-ils how-many have-they
b.
tous Com bien de livresi ont-its how-many of books have-they all 'How many books have they all read?t
e;
de livres?
of books
e;? Ius read-AOR
As noted already, de Swart ( 1 992) observes that (SOb) is ambiguous. Either the universal takes wide scope over the WH-phrase (we ask all persons how many books they read), or the universal takes narrow scope (we ask for a single number: the number of books everyone read). In contrast, (50a) only has the reading where the universal takes wide scope. Attempts to derive the reading where the universal takes narrow scope can only result in LFs like {5 1}. This gives an intervention effect: there are no visible occurcren es of wh from a because of the intervening universal, and so a is not permitted to bind any variables, with the result that z is left without a denotation. (51 )
a [Vx person(x)] (read(x y) [3y[wh z = [vll books(y)]) ,
Note that (SOb) encounters no such problem because it can yield the LF in (52) where the usage information wh is visible from a, that is, outside the scope of the intervener tous 'all. ' (52)
Q z [3y [wh z = lvi] books(y)] [Vx person(x)] read(x, y)
The question that now arises is why the wide scope reading for the universal is nevertheless available when other interveners such as negation lead to complete ungrammaticality, cf. (4 1). To explain this. we can follow Krifka (2001 ). who argues that a universal can scope out of a question act (and quite generally all speech acts) because such a scoping has the effect of giving rise to a conjunction of question acts (more generally speech acts). Thus we can view a pair-list question like (53a) with its universal quantifier as abbreviating a conjoined question, e.g. (53b).
63
Split-DPs. generalized EPP and visibility
(53)
Which book did every linguist read? a.
[[01' every linguist]i [Q-ACT which book did ej read]]
b.
[Q.ACTI which book did John read] and [Q.ACT2 which book did Mary read] and .. .
Thus, in the LFs of readings where a universal takes wide scope there is no universal present between Q and WH to act as an intervener. Negation, on the other hand, cannot scope out of question acts (for example, pair-list readings are never available in questions such as What did no one see?, see also Chierchia 1 992), which means that it is always present between Q and WH and consequently negation acts like an intervener.
5. Vacuous Movement We now turn to the third question posed at the outset, i.e., why raising of the nominal in French remains possible when bare operator movement can satisfy the visibility requirement. In the previous section we established the idea that intervention effects arise with interpretation crashes. It is reasonable to suppose that syntax has no access to infonnation from the interpretation procedure. It follows that syntax itself cannot be sensitive to when intervention effects will occur. The optionality syntax displays can therefore be viewed from the perspective of syntax blindly permitting potentially sub-economical movements to meet coverage demands. For example, full movement in (40) is not required, as the speaker can ask the same question with a split-DP. namely (39). In contrast, full DP movement is necessary in (42); else the speaker would lack the ability to ask the question, there being no interpretable split alternative «4 1 ) is ruled out since it brings about an intervention effect). With this reasoning we can suppose that syntax allows a potentially vacuous movement (as in (40» 'just in case' the worst happens and an intervener is present in the structure. 6. Conclusion
Our account has left us with two visibility requirements in the C domain. The the clause, that is, signaling that an infonnation question interpretation is required. We implemented this idea as the introduction of Q into the Logical Fonn. The second visibility requirement corresponds to the need for usage information to support the interpretation, much like we saw with Partee' s marbles example of (3 1)-(32). We implemented this idea as the introduction of wh into the Logical Form.
first involves typing
References
Androutsopoulou. A nton ia (1 997). SpIit-DPs, focus, and scrambling in Modem Greek. West Coast Conference on Formal Linguistics 1 6: 1-16. Bianchi, Valentina ( 1 999). Consequences of antisymmetry: Headed relative clauses. Berlin: Mouton de Gruyter.
64
Alastair Butler and Eric Mathieu
Boeckx, Cedric (2003). Islands and Chains: Resumption as Stranding Amsterdam: John Benjamins. Bolinger, Dwight L. ( 1 987). Echoes reechoed. American Speech 3: 26 1-279. Butler, Alastair and Eric Mathieu (2004). The Syntax and Semantics of Sp/i Constructions: A Comparative Study. BasingstokelNewYork: Palgrave, MacMillan. Cecchetto, Carlo ( 1 999). A comparative analysis of left and right dislocation ir Romance. Studio Linguistica 53 : 43-67. Cheng, Lisa and Johann Rooryck (2000). Licensing WH-in-situ. Syntax 3 1-1 9 . Chierchia, Gennaro (199 1 ). Functional WH and weak crossover. In Dawn Bate! (ed.), Proceedings of WCCFL 10, Stanford. pp. 75-90. Chomsky, Noam ( 1 995). The Minimalist Program. Cambridge, Mass.: MIl Press. Chomsky, Noam (2000). Minimalist inquiries: The framework. In Rogel Martin, David Michaels, and Juan Uriagerelca (eds.), Step by step; Essays on minimalist syntax in honor of Howard Lasnik, Cambridge, Mass.: The MIT Press, pp. 89- 1 55 . Chomsky, Noam (20ot). Derivation by phase. In Michael Kenstowicz (00.), Ken Hale. A Life in language, Cambridge, Mass.: MIT Press, pp. 1-52.
Dekker, Paul (2002). Meaning and use of indefinite expressions. Journal of Logic, Language and Information 1 1 : 14 1-1 94. Demirdache, Hamida (1 997). Dislocation, resumption and weakest crossover. In Elena Anagnostopoulou, Henk Van Riemsdijk and Frans Zwarts (eds.), Materials on Left Dislocation, Amsterdam: John Benjamins. pp. 1 93-23 1 .
Deprez, Viviane ( 1 998) . Semantic effects of agreement: The case of French participle agreement. Probus 1 0: 1-67. Engdahl, Elisabet (200 1 ). The role of syntactic features in the analysis of dialogue. In Christian Rohrer, Antje RoBdeutscher and Hans Kamp (eds.), Linguistic Form and its Computation, Stanford: CSLI, pp. 1 1 1-142. Ginzburg, Jonathan and Ivan Sag (2001 ). English Interrogative Constructions. Stanford: CSLI. Fanselow, Gisbert and Damir Cavar (2002). Distributed deletion. In Artemis Alexiadou (ed.), Theoretical Approaches to Universals, Amsterdam: John Benjamins, pp. 65-107. Hiemstra, Inge (1 986). Some aspects of WH questions in Frisian. North Western European Language Evolution 8: 97-1 1 0. Honcoop, Martin ( 1 998). Dynamic Excursions on Weak Islands. Leiden UniversitylHIL. Ishihara, Shinichiro (2002). "Invisible but audible Wh-Scope Marking: Wh constructions and deaccenting in Japanese. In Line Mikkelsen and Christopher Potts (eds.), WCCFL 21 Proceedings, Somerville, Mass., Cascadilla Press, pp. 1 80-1 93. Kamp, Hans ( 1990). Prolegomena to a structural theory of belief and other attitudes. In C. Anthony Anderson and Joseph Owens (eds.),
Propositional A ttitudes:
The Role
of Content in Logic
,
Language and
Mind, Stanford: CSLI, pp. 27-90. Kayne, Richard ( 1 989). Facets of Romance past participle agreement. In Paula Beninca (ed.), Dialect Variation and the Theory of Grammar, Dordrecht: Foris, pp. 85-103.
65
SpJit-DPs, generalized EPP and visibility
Kayne, Richard (2002). On some preposItionS that look like DP internal: English 'of and French 'de'. Ms., NYU. Krifka, Manfred (200 1). Quantifying into question acts Natural Language Semantics 9: }-40. Kuno, Susumu ( 1 972). Functional sentence perspective: A case study from Japanese and English. Linguistic Inquiry 3: 269-320. Mathieu, Eric ( 1 999). French WH in situ and the intervention effect. ueL Working Papers in Linguistics 1 1 : 44 1-472. Mathieu, Eric (2004). The mapping of form and interpretation: the case of optional WH-movement in French. Lingua 1 1 4: 1 090-1 1 32. Obenauer, Hans Georg (1 976). Etudes de la Syntaxe Interrogative du Franfais. TUb ingen: Niemeyer. Obenauer, Hans-Georg ( 1 983). Une quantification non-canonique: La quantification a distance. Langue Franfaise 58: 6�88. Obenauer, Hans Georg ( 1 994). Aspects de la Syntaxe A-Barre. Un ive rsity of Paris VII: These de doctorat d'Etat Oiry, Magda (2004). Les questions exceptionnelles en franyais ou
. Ms. University of Nantes. Ouhalla, Jamal ( 1990). Focusing in Berber and Circassian and the V2 phenomenon. Talk presented at University College London. Pesetsky, David (1998). Some optimal ity principles of sentence pronunciation. In Pilar Barbosa, Danny Fox, Paul Hagstrom, Martha McGinnis, and David Pesetsky (eds ), Is the Best Good Enough?, Cambridge, Mass.: MIT Press, pp. 337-383. Pesetsky, David (2000). Phrasal Movement and Its Kin. Cambridge, Mass.: MIT Press. Platzack, Christer ( 1 998). A vis ibility condition for the C domain . Working Papers in Scandinavian Syntax 6 1 : 53-99. Poletto, Cecilia and Jean-Yves Pollock (2004). On the left periphery of some Romance WH-questions. To appear in Luigi Rizzi. The Structure of IP and CPo The Cartography of Syntactic Structures. Volume 2. Oxford: Oxford University Press. Rizzi, Luigi ( 1990). Relativized Minimality. C ambridge, Mass.: MIT Press. - Stalnaker, Robert (1998). On the representation of context. Journal of Logic. Language and Information 7: 3-19. Starke, Michal (200 1 ). Move disolves into Merge: A Theory of Locality. Doctoral dissertation, University of Geneva. de Swart, Henriette ( 1 992). Intervention effects, monotonicity and scope. Proceedings of SALT 2: 387-406. Van Geenhoven, Veerle ( 1 998). Semantic Incorporation and Indefinite Descriptions: Semantic and Syntactic Aspects 0/ Noun Incorporation in West Greenlandic. Stanford: CSLI. Van Rooy. Robert ( 1 997). Attitudes and Changing Contexts . Doctoral disse rtation IMS, Stuttgart. Watanabe, Akira (1 992). Subjacency and structure movement of WH-in situ Journal 0/ East Asian Linguistics 1 ; 255-29 1 . Zimmennann, Ede ( 1999). Remarks on the epistemic role of discourse referents. In Lawrence Moss, Jonathan Ginzburg, and Maarten de Rijke (eds.), Logic. Language and Computation II. Dordrecht: Kluwer Academic Publishers, pp. 5 1 3-540. .
-
-
.
.
-
66
.
I
Alastair B utler and Eric Mathieu
Alastair Butler ILLClDepartment of Philosophy University of Amsterdam Nieuwe Doelenstraat 15 1 0 1 2 CP Amsterdam The Netherlands
A [email protected] http://staff.science. uva.nl/-ahutler/ Eric Mathieu Department of Phonetics and Linguistics University College London Gower Street London WCI E 6BT United Kingdom
eric@ling. ucl.ac.uk htlp:llperso. wanadoofrleric.mathieul
67
! Tbe pbase structure of tense·
lonny Butler Universitiit Stuttgart
I argue that phases should
I
be defined as domains
of quantificational closure.
propose that a pbase consists of a core predicative category (V. T), topped
by I-syntactic 'little' heads
(v, I) that introduce situation variables; this whole CP level which closes off the variables intro
suuCttlre then topped off by a
duced by the little heads. This derives a theoJ)' of temporal construal akin 10 that of Stowell
(1 996). The system is
extended to cover aspecrual notions like
perfect/progressive. where evidence from their interaction with
modality sug
gests that perfect and progressive aspect should be considered to head their own phases.
1. Intro This paper argues for a view of phase structure (Chomsky 1 999; etc.) where phases are defined in tenns of quantification: each phase corresponds to one domain of quantifica tional closure, such domains being provided by a series of quantification-encoding CP levels interspersed through the clause (cf. Starke 1993.2001 ; Hallman 2000; Brody & Szabolcsi 2003; Butler 2003a). Specifically. the purpose of these CPs is to close off vari ables introduced by I-syntactic 'little' heads above core predicative categories V, T and N-little v, t. and n. For v and I, these are situation (s) variables; for n, individual (x) variables.2 This is shown to derive a system of tense construal along the lines of that pro posed by Stowell ( 1996). wherein tense is a temporal ordering predicate, serving to order the situations denoted by the closed off s variables of v and I (corresponding to. re spectively. the Event Time E and Speech Time S of Reichenbach 1947). The system is extended to cover (outer/viewpoint) aspectual notions (perfect/progressive) also, leading to the proposal that there are in fact more phases than the vP and CP phases generally assumed. The structure of the paper is this: §2 goes through a particular view of I-syntax within vP (Hale & Keyser 1993; etc.), and justifies the topping off of vP with a CP layer. §3 extends this view of I-syntax to TP. attributing parallel structure to the verbal and temporal layers of the clause. §4 then goes through the system of tense construal in more detail. §5 extends it to aspectual construal, providing evidence that structure equivalent to that proposed for the verbal and temporal layers of the clause also obtains -Thanks arc due to David Adger. George
Tsoulas. and Tim Stowell for discussion; additional thanks to Tun
for giving the title his bleSSing. Also to the audience at the workshop on which this volume is based, and an earlier audience in York. The final draft of this paper was completed some time ago: for an updated and (1 like to think) better treatment of some aspects. see my doclonll dissertation (Butler 2004b).
2) assume for DPs thai D and C arc essentially the same categoJ)' (cr. Cardinaletti & & Torrego 200) . though in fact i don't consider the putative DP phase in detail here.
MIT Working Papers in linguistics 49, 69-85 Perspectives on Phases © 2005 lonny Butler
StiUXe
1999; Pesersky
The phase structure of tense
in the aspectual layers, performing the same quanlificational function. §6 introduces an analy sis of modality that in conjunction with the aspectual story p rovides further evidence for the aspectual phases proposed. §7 is a conclusion.
2.
The Structure of the vP Phase
2.1. Situation structure and argument structure
I assume a structure for the verbal layer of the clause where it is decomposed into a V level and a v level. V, I take to be a lexical Root (Pesetsky 1995; etc.), which gives us the big 'meaning' semantics: basically, it is like an encyclopaedic entty that tells us what property is being predicated. So a Root like EAT tells us that we're talking about eating; a Root like MANOUEVRE tells us that we re talking about manouevring. But it doesn't tell us much else: in particular, it doesn't tell us anything about situational or argument semantics. What does tell us this is the vP layer (Harley 1995; Marantz 200 1; Pylkkanen '
2002).
v, I use as a shorthand for a more complex structure of 'little' heads, which syntactically encode situation (=event) sttUcture. Each of these heads encodes one sub· situation. which when composed describe one macro-- situation This macro-situation is fonnalized as a situational variable s corresponding to something like a Davidsonian event argument (Davidson 1 967). When this variable is given a reference (details of which below), it corresponds to something like the Reichenbachian Event Time E. Arguments are merged as the specifiers of these little heads. The interpretation an argument receives will depend on what little head it is specifier of thus deriv ing theta· roles: so if we have say, a CAUSE head, the argument in its specifier will be interpreted as a causer, or agent. Argument structure is thus a reflex of situation structure. Different versions of l·syntax assume different kinds/combinations of little �ads. For my purposes here it doesn't really matter what the internal structure of this layer is; all we need to know is that the complex v level provides us with a (macro )situational variable s. I assume standardly that Root V moves up to the highest little v head in the structure. .
,
,
2.2. Closure There are various stories around in the literature relating to what happens to the situation variable in vP next They largely claim that it is subject to existential quantification ; what performs this quantification is where the variation comes in. Probably the most common proposal is that it is subject to the EXISTENTIAL CLOSURE operation of Heim (1982); Diesing (1992). This is an operation which takes place at the mapping from the syn· tax to the semantics; its function is to insert an existential quantifier into the semantic representation, which closes off any stray variables not subject to binding by other quan· tificational elements. For Heim, this operation takes place at an abstract 'text' level-that is, over whole (possibly multi-sentential) utterances. Diesing influentially proposes that it rather takes place inside the clausal structure. with the quantification being over VP. In an l·syntax story, this reinterprets straightforwardly as quantification over vP. Several analyses propose a simil ar kind of 'default' existential quantification over vp, but rather than Diesing's abstract closure mechanism, they posit that the quan.
70
Jonny Butler
I 1
tifter is in fact represented structurally in the syntax. Of panicular relevance here ar:
(1991) and Stowell ( 1 996). Stowell ( 1 99 1), following Kratzer ( 1 995), proposes that something like Davidson's event argument is syntactically real, and
the analyses in Stowell
is introduced as the outermost argument of the verb in vP. Specifically, he takes the verb
to assign an EVENT theta-role, whose position is filled by an ex.istential QP over events.
Closure is. then. in a sense part and parcel of the event position itself: if a verb has an event argument3• then it has existential quantification over that argument by default. and this quantification is syntactically represented.
Stowell (1 996) gives an alternative analysis for such quantification. Here, h;.. assumes that all verl:/$, stage-level or not, select
i
I
I
an
event argument (though 'event' her:
means something more general like 'situation' , or 'eventuality ' . I use 'situation ' here) Stowell seems more agnostic about whether the argument is real or abstract in this paper: his trees suggest he takes it to be real, but to introduce an indefinite argume�t, and thu;' a variable (as Heim 1 982), rather than a closed off QP as in Stowell (1991). He posit:· that this variable is given reference by a category he labels Z(EIT)P, which sits on top of
vP, and is basically a kind of existential detenniner over situations: 'ZP bears a structural
NP' (Stowell 1996; 280). Again, then, something like existential closure is represented overtly in the syntax.
relation to [v]P that is analogous to the relation that OP bears to
though this time it is a result of some higher operator, rather than pan of the denotation · of the situation argument itself. In Stowell's story, the higher operator is of category Z, a selective binder of
situation variables, serving specifically to give the situation argument a reference. In fact, it serves to give it the reference of a time, and this time is the internal argument
of the temporal predicate
T.
The proposal I make here is that this is unnecessary: we
don't really need to give the situation argument the reference of a time, since we
can
get this just by virtue of the fact that it is an argument of T. If the operator-variable construction only denotes a particular situation, and the function of
T
is to order this
situation temporally relative to some other situation, then this in itself will have the effect of making the situation (seem to) correspond to a time. I suggest that rather. the operato is simply an ordinary existential, with no particular temporal relevance. The immediate bonus of this is that we can take the operator not to be a selective binder of situation variables, but rather an unselective binder of variables generally (cf. Sportiche
2002). This allows it to correspond exactly to Diesing's vP closure operator
closing off both the situation variable
s,
and any
x
variables introduced by existential
indefinite DP arguments.
Given the general usefulness of this operator position, 1 will adopt it here. though since I generalize its purpose away from situation/temporal related to more general clo sure, I won't adopt Stowell's ZP label. Instead. I will take this position to be a clause
internal CP position. This label doesn't really matter-it could equally be labelled OP
or more non-commitally XP-but for the idea that there is a CP-like position here, see among others Starke
(1993,200 1 ); Belletti (2001); Jayaseelan (2001); Butler (2oo3a).
I take this quantificational4 closure to define the EDGE of the vP phase, since any
loose variables etc. should be dealt with by it. Anything that needs to escape closure will have to move out of the scope of this existential operator, to [Spec, CP]: this suggests an 3Stowell. following Krcltzer, assumes in this paper that only stage-level predicates assign an EVENT tbera
role. 4Quantificational ralber than existential because I assume olber quantificationa) operators are sited here also·
sec §6.
71
The phase structure of tense
analysis where successive cyclic movement is movement to escape closure. The general structure of the vP phase. then, is as shown in ( I ) (i rre levant details omitted).
(1)
CP
�
3i
vP
-------
subject
Vi
� SITUATIONi
VP
6.
3. The Structure of the T Phase A fi rst point to avoid confusion: given that I
now take both the vP and the CP phases
to be CPs. I will refer to them by the label of their Root-the V phase and the T phase. respectively.
I take the structure in (1) to generalize to the T phase, and in fact to phases in general: a phase consists of a predicative Root H. topped by one or more I-syntactic little heads h encoding situation structure. topped by a CP which closes off variables introduced in the h layer. The structure of the T phase will thus be as in (2). (2)
CP
�
3i
tP
� t
SITUATIONi
TP
6.
T. like V, is a Root category that says what the predicate is about. In this case, it predicates a temporal relationship [±PAST] between two situations. which I will Re ichenbachesquely refer to as the event situation E and speech situation S. More detail follows in §4. t is like v: a little head that introduces a simation variable. There is of course a lot of syntactic evidence for some projection in this position. commonly labelled Agr(S) (Pollock 1989; etc.; etc.). This area of the clause is argued to be further decomposed by Cardinaletti (1 997.2000) ; Cardinaletti & Roberts ( 199 1); Sponiche ( 1 996); Cardinaletti
& Starke (1999); Manzini & Savoia (2002); which perhaps points to a finer grained structure for t analogous to the fine grained structure that v is standing in for here. C heads the standard CPo Its sali ent purpose here is again existential quantifi cation over the situation variable introduced by t. giving it reference. and giving us the speech situation S.
72
Jonny Butler
4. The Phase Structure of Tense In short, the above discussion reduces to this: vP is immediately dOminated by :.. which contains some operator element taking in the situation variable associated with and giving it referential status; this operator-variable pair gives us something like th� situation time (i.e. Reichenbach's Event time E). We assume the same thing happer.�· with external CP and the situation variable associated with t. giving us something like the Speech time S. The function of T is to specify some temporal ordering relationship between these two objects. We end up, then, with a structure like that given in (3). __ .
(3)
CP
�
tP
�
SITUATIONj
TP
�
T
CP
�
31
vP
�
SITUATIONi
VP
6.
As far as the interpretation of (3) is concerned, we can, allowing for certain differences in structure, follow the ideas set out in Stowell (1996). Stowell decribes tenses as 'dyadic predicates of temporal ordering ... they take two time denoting phrases as their arguments, and it is in these categories that the referential properties associated with tenses reside' (Stowell 1996: 279). In Stowell's system, these phrases are ZPs; here, they are the operator-variable constructions associated with CPs and llv. The tense predicate 'temporally locate[s] the denotation of E in relation to the denotation of its external argument .. PAST means "after" . . and PRESENT means "simultaneous with" or "overlaps'" (pp.280-1). This clearly carries over easily to the structure given here. To exemplify, (4a) would have the structure (4b). This is interpreted along the lines of 'there exists a Speech situation S, and there exists an Event situation E, E a situation of raconteurs and roustabouts saying "Buddy, come on in"; S is located temporally after E' . This is basically the interpretation we want. As the bones of an approach to tense, then. this is promising. But it has various specific details that need to be dealt with. First, it involves only two times, as opposed to (neo)-Reichenbachian approaches which often make use of three (Speech time S, Ref erence time R, and Event time E). The reason a third time gets invoked is to deal with 'complex tenses' such as the English perfect. which can't be dealt with straightforwardly by appealing to just two time points. Therefore it needs to be shown that this system is theoretically capable of dealing with the kind of empirical data that led to the Reichen bachian three-time system. .
I
.
•••
73
U III
Jj
I<-. o
� E
�
�
�
.s:: Q. u
t=
(4) CP
a. The raconteurs and roustabouts said 'Buddy. come on in'
h.
the rs & rs
t'
______
tP
------
3J
TP
------
I
SITUATI O N j
CP
------T[+PASTI
vP
------3i
v'
-----tsubject
�
�VP SITUATION i
said buddy come on in
r:t
lonny
1\
I I
! I I �
Butler
s. Aspect 5.1. Complex tenses Reichenbach's ( 1 947) original proposal to treat tenses as relating three times rather than just two was motivated by the fact that certain constructions can '[ be dealt with using just two. In particular, the use of three times allows perfective constructions to be represented. (5-7) illustrate the English past, present, and future perfect respectively, and it is clear that the differences in interpretation among these three sentences can't be captured with reference to just two times.
(5)
Susan had gone to sleep
(6)
Susan has gone to sleep
(7)
Susan will have gone to sleep
In the past perfect (5), E is taken to be ordered before R which is before S: that �i is, the Event situation is being evaluated as past from a time which is itself in the past,:� with regard to the Speech situation. In the present perfect (6), S and R are taken to bell1n simultaneous, with E preceding them. Note that this isn't the same as simple past, wheren E precedes just S: there is a distinct difference between (6) and (8), and it is captured well \\{ ' by the use of three time points.
�
(8)
Susan went to sleep
In the future perfect (7), S and E are taken to be unordered with respect to one another. with R following both (cf. Comrie 1985; Hornstein 1 990). There is no way we can capture this with only two times: we need three. In a system where T relates only two time points. this problem obviously has to . be dealt with along different lines. Here, I follow Vikner ( 1 985); Zagona ( 1 990); Stow ell (1 996); Demirdache & Uribe-Etxebania (2001) in taking the obvious way out and treating perfect aspect as essentially equivalent to a [+PAST] tense predicate. This means perfect tenses generally express a relation between two separate tenses: have will order E relative to its own situation argument, which will basically be R; the denotation of this argument will be fixed as in the usual case of a subordinate clause in Stowell's system, by
being controlled by the closest c·commanding situation argument. The implicit situation argument of (the head hosting) have. Stowell claims, is a result situation, so that ·the higher tense [TJ provides its usual temporal argument structure, locating the time of the result state (the time of "having") in relation to ... the Speech Time ... The past partiCiple provides an additional past tense. which orders the thematic verb's [Situation] Time in relation to that of the result state.' (Stowell ) 996: 285) What we have, then, is a means of getting a Reichenbachean 3-time system when we need to. and not when we don't, which is a good result.
5.2. Aspectual phases I am taking the perfect/progressive aspect to be fundamentally equivalent to tense, in that they introduce a new situation into the representation, and place this situation temporally relative to the situation denoted by their complement. Perfect aspect correlates with past tense, ordering the situation it introduces as after the situation denoted by its comple ment; progressive aspect correlates with present tense. placing the situation it introduces
75
The phase structure
of tense
temporally within the situation denoted by its complement. In general this makes perfect sense and is more than reasonable; see for example Stowell ( 1996) (and the discussion thereof above); Demirdache & Uribe-Etxebarria (2001); Butler (2003b) for extensive justification of what the system can get us. For this story to run in the present framework, though, it is necessary also to postulate the same general structure for perfect/progressive elements as for the V and T phases: namely, a core predicative category (PERF/PROG), topped by I-syntactic situation stucture (perj/prog), which is in tum sunnounted by a quantificational CP layer-that is. 'a sentence like (9a) must have a structure like (9b) (words omitted).
(9)
a. The piano has been drinking b.
CP
�
C
IP
�
TP
�
T
CP
�
C
perfP
�
perf
PERFP
�
PERF
CP
�
C
progP
� prog
PROGP
�
PROG
CP
�
C
vP
A
v
VP
6.
Two questions arise here: 1 . What empirical (syntactic or semantic) evidence do we have for these 'extra' CP levels? 2. Do we take these CPs to be phases or not? §5.2. 1 will deal with question I, showing that ne gatio n, argued in Starke (2001); Butler (2003a). to be connected with CP levels. provides syntactic evidence for the existence of these CPs. Further arguments based on interpretive facts to do with
76
Jonny Butler
modal/aspectual interaction will be prese nted in §6. after a theory of modal interpretatio .. has been briefty introduced, which will show that in addi ti on to negation. qu antification must also be active in the aspectual CPs . As to w hether these CPs should be considered phases or nOl, this depends on how e xactly we end up defining phases; discussion of this will be reserved till §7. For now, I will simply assume that the aspectual CPs are phases on a par with the T and V phases. 5.2. 1 . Negation
In Starke (2001); Butler (2oo3a), it is argued that CP l ayers host a negative operator sited by Butler for concreteness in Ri zzi ' s ( 1 997) Foc position. Thus, the standard Negp is replaced by the internal FocP; the external FocP provides a wide·scope position for negation. Given that there are two additional CPs represented in (9b), we make a straight forward prediction that negation should be able to appear in (the Foc proj ection 00 thest CPs also. This prediction is borne out, as shown in ( 1 0).5
(10)
[cp hp the piano might [cp (nol) [PERFP have rcp (not) [PROGP been [ep ( notl [vp drinking]])]J])] ':{�f: ,
\
;
II:!·'
j' !
I
There are various points raised here: one is that we don't see negation overtly' (' ' in the hig hest CP; this seems to be a fact of English though-negation does appear here in many languages (Italian, for example), and it i s certainly an available scope position for negation in Engl ish. Another is that we don ' t tend to see all these negation s surfacin� simultaneously, though there is nothing in the theory to force this ; really it seems most likely to be a processing problem, since with appropriate context and stress patterns w� can make the example reasonably fel icito us. A third is that the two highest positions for negation, i.e. those either side of T, seem to give sentential negation whereas the lower positions seem to give constituent negation. In fact this falls out of the story in Butler (2oo3a), where subject movement pl ayed a crucial role . Canonically, I take the subject to sit in [Spec, tPl, having moved up from its base position in [Spec, vP]. In order to interpret the subj ect in this position, though. we have to create a A·abstract immediately below it ( see Heim & Kratzer 1998 for w hy, and a way of doing this; see Butler 2004a for a variation on Heim & Kratzer's story whereby the A·abstract is the semantic reftex of the EPP feature on t). The A-abstract so created constitutes the main sentential predicate. Once the topmost CP is sorted, we then have a real proposition. If negation appears in the topmost CPo we of comse have propositional, i.e. se ntential . negation. If it appears i n the next CP down, it will basically negate the predicate; this negative predicate will then apply to the subject, and this will also give you sentential negation; see Hom (1 989) for details, especially chapters 2 and 7 . If, however, the negation is below this, it will be too low down inside the predicate to give you either propositional or predicate negation in the relevant sense , so you get only the constituent negation reading in these c ase s. This, then, seems to be good syntactic evidence that we have at least a Foe projection on lOp of the proposed PERF and PROG phases. 5 Might is added in (10) as a neutJal probe for the syntactic position of T.
77
The phase structure of tense
6. Modals In Butler (2oo3a) I argued that modality is connected to the CP levels proposed here. There, I took modals to be quantifiers over possible worlds (Kratzer 1 977. 1 98 1 , 1 991). operating over some abstract world variables, whose provenance I left vague. Here. I will assume that in fact modals quantify not over world variables. but over situation vari ables; specifically, the ones introduced in the T/PERF/PROGN phases.6 This derives a situation semantics reinterpretation of Kratzer's world semantics story (cf. Ponner 1 992). The means of quantification, I take to be quantificational operators sited in the CP level of phases. extending the story for existential closure developed above to include other quantifiers too, most saliently a universal.
6.1. Propositions I follow Kratzer in taking necessity modals like must to quantify universally. possibility modals like mighllcan to quantify existentially. I also follow her in taking medals to be propositional operators. I depart from Kratzer, though, in what I mean by 'proposition' . One definition o f phases familiar from Chomsky (200 1 ) i s that they are proposi tiona elements. CP, Chomsky describes as propositional basically in the usual semantic sense (evaluable for truth, etc.). vP, he describes as propositional in the sense that it 'has full argument structure' (p. 22). Now. there is an argument to be had here as to whether these two sense of 'propositiO.lal' are really similar enough to be brought in as an argu ment for the unified nature of phases. but certainly the 'full argument structure' definition is a legitimate one in terms of logic; and in fact. we can boil both senses down to just this one. The V phase has full argument sttucture in that it has all its theta roles filled (or however we want to formalize this notion); the T phase also has full argument structure, in that it is a dyadic predicate needing two situational arguments for saturation, which are provided by the V phase and by t's (closed off) situation variable (dino for the PERF and PROG phases). It just so happens that this phase really is propositional in the usual. truth-evaluable sense, as well, because it predicates a temporal relation between two sit uations. This is exactly the case we need for a set of truth conditions. This being so. the real propositional parallel is ,the one referring to full argument structure. If we take it that this definition of propositional is the one that matters to propo sitional operators like modals, we predict that modals will be able to operate not just over the whole sentential proposition. but over any expression which is propositional in the relevant sense. Here, this means over any phase. This gives us a prediction with regard to scope: in a simple clause. with just the T and V phases (i.e. no aspect). we have two positions in which modals can scope: one at the top of the clause, in the T phase's CP, and one in the middle. in the V phase's CPo These two positions should be easily distinguishable by their interactions with other scope bear ing elements in the sentence; most relevantly. with non-presuppositional weakly quanti fied subjects (in the sense of Milsark 1 974).
6.2. Subjects It is conunonly accepted. following Diesing ( 1992). that there arc two basic scope posi tions for weakly quantified subjects. which can be distinguished by their different read6Modals must also relativise these variables to possibility, so we are talking about possible siruations rather man just situations; we don't need to go into me technicalities of this here.
78
Jonny Butler
)
t
i
I'r f.
i
t
ings. Diesing takes non-presuppositional (existential) weakly quantified elements to i · ·
troduce variables into the representation. which need to be quantified over by some higher
element--either oven elements like temporal adverbs (always. never. sometimes, etc.), c : by abstract closure operators like her existential closure operator discussed in §2.2 abov&;: here reanalysed as a syntactically represented existential operator in the edge of the V
phase. She takes presuppositional (quantificational) weakly quantified subjects not to b�
variables but to have their own quantificational force. and thus not to be associated wim existential closure.
The two scope positions associated with these elements are [Spec,
vP] for th�
existential cases, where the subject is existentially dosed, and canonical [Spec. IP1 for th� presuppositional cases. where it is outside the scope of closure. Given these assumption�
we predict that:
l;
I·w �:
.. . .•. .
1 . Modals in the edge of the T phase will scope over the subject whether it is inter preted as existential or presuppositional, since they are higher than both subjec. scope positions.
2. Modals in the edge of the V phase will scope above existential subjects, since they are higher than [Spec.
.'.·.'
l
f
lower than [Spec. tP] .
( 1 1)
a
:ril
Some philosophers go to those seminars Presuppositional reading: 'some philosophers (specifically, Quine. Camap, and Socrates) go to those seminars' . Existential reading 'there exist philosophers who go to those seminars, though
;
vP}. but below presuppositional subjects, since they ar��f'
I don't know who ' .
b . Some philosophers must g o t o those seminars (epistemic necessity) PresuppositionaYquantificational reading: 'it must be the case that (at least) some philosophers go to those seminars (because Quine, Carnap, and Socrates all told me they were intending to)' . Scope: modal > subject
Non presuppositionallexistential reading:
'it must be the case that there
are philosophers who go to those seminars (because they're about possible worlds)'. Scope: modal > subject
c. Some philosophers might go to those seminars (epistemic possibility) PresuppositionaVquantificationaJ reading:
'it might be the case that (at
least) some philosophers go to those seminars (because Quine, Camap. and Socrates all told me they were intending to) ' . Scope: modal > subject
Non presuppositionaVexistential reading:
unable to distinguish relative
SCOpe.7
7There are going to be gaps in the parodigm here, since we assume possibility modals instantiate existential
quantification: we can't therefore make any concrete predictions with regard to no�presuppositionallexistential readings of weak subjects and possibility modals since we can'l distinguish the relative scope of two existential quantifiers (May 1985).
79
The phase structure of tense
d. Some philosophers must go to those seminars (root necessity) Presuppositional/quantificational reading: 'Quine, Camap, and Socrates are required to go to those seminars' .
Scope: subject > modal
Non presuppositionallexistential reading: it is required that some philoso phers go to those seminars. as a condition on our being given money to run them' . •
Scope: modal
>
subject
e. Some philosophers can go to those seminars (root possibility) Presuppositional/quantificational reading: 'some philosophers-specifically Quine, Camap, and Socrates-are able/allowed to go to those seminars' . Scope: subject > modal Non presuppositionaJ/existential reading: scope. 7
unable to distinguish relative
The data in ( 1 1 ) confirm that the two positions posited are indeed available for modals; moreover, they show that the two scope positions for modals correspond to the two major readings for modals: those at the edge of the T phase are interpreted as epis temic, those at the edge of the V phase are interpreted as root. This is an interesting result; it correlates with the claim often found in the literature that epistemic and root in terpretations for modals differ semantically along the lines that 'epistemic modals modify a sentence [= proposition] and deal with the truth value of that sentence; root modals re late ... [a subject] to an activity and deal with permission, obligation and ability' (Cook 1978: 6). This is basically what we would expect epistemic and root modals to do if they correspond to the two scope positions I claim here, since the higher, epistemic modals will invariably scope over a tensed (Le. truth evaluable) proposition, including subject, whereas the lower. root modals will commonly intervene between a subject in [Spec, tP] and its vP predicate (the exception being where the subject is interpreted existentially, as above). This analysis also derives the old intuition from Ross ( 1 969) that epistemics are lexically one-place (intransitive) predicates. somewhat like raising verbs, while roots are two-place (transitive) predicates, somewhat like control verbs. For more extensive discussion of this story, see Butler (2003a); here. it is suf ficient to note that epistemic readings for modals correspond to modals scoping above tense and the subject. while root readings correspond to modals scoping below tense and the subject.
6.3. Modal/aspectuaJ interaction What does this have to do with aspect? If modals are quantificational elements in the edge of a phase operating over sit uational variables in the domain of that phase, and if perfect/progressive aspect introduce situation variables and phases, then we make another prediction about where modals can scope: not just in the edge of the T and V phases. but also in the edge of the PERF and PROG phases. We aren' t going to find evidence like that seen above relating to subject interpre tation, since subjects don't get interpreted in the aspectual phases. The clearest evidence. then. would come from the interaction of modality with aspect itself; and in particular. with perfective aspect. 80
Jonny Butler We take PERF to be a temporal predicate l i ke [+PAST] T. ordering two situatior:- temporally relative to one another-specifically, ordering the result situation associated with perf and its own CP as after the situation denoted by its complement (which will b� either a V phase or a PROG phase, depending) as in ( 1 2). ( 12)
CP
-------
�
perf
SITUATION j
PERFP
�
PERF[+PASTj
CP
�
3i
vP
�VP
SITUATION1
6
There are two CP levels in (1 2) that could act as the scope positions for modals the CP of the PERF phase and the CP of the V phase. There are therefore two times at which the modal evaluation could be understood to take place: either in the past, if the modal is associated with the V phase, or in the 'now' of the PERF phase. We can think of this informa1ly in terms of scope--the modals should be able to scope either over or under the perfect. depending which CP they are in. Condoravdi (2001) and Stowell (2004) show that this is so. Condoravdi shows that for cenain modal utterances containing perfect have. like ( 1 3), the modal receives different readings depending on whether it scopes above or below the perfect: if it scopes above it receives an epistemic reading (14), if it scopes below it receives what she calls a 'metaphysical' reading (I S), which doesn't relate to the epistemic state of the speaker (or whoever) but simply refers to 'how the world may tum out, or might have turned out. to be' (Condoravdi 200 1 : 3). This is essentially the reading labelled by Palmer (1990) 'dynamic' modality: dynamic possibility he describes as expressing 'neutral possibility, simply to indicate that an event is possible'. (p. 83). Butler (2003a) and Palmer both treat dynamic interpreted modals as a subcase of roots, a generalized version of what are often described as 'ability' modals. ( 13)
He might have won the game
( 1 4)
He might have (already) won the game (#but he didn't)
(15)
At that point he might still have won the game but he didn't in the end. (Condoravdi 2001 : 4; her (6b}-(7b»
As it stands, Condoravdi 's analysis doesn't actually say much about the structure in (9b) or ( 12). since all we see are an epistemic and a root reading. which were taken above to be associated with the T and V phases, respectively. However, Stowell (2004) takes Condoravdi's basic insight and shows that root interpreted modals can scope over PERF also ( 1 6- 1 7). in which case the modal must be associated with the PERF phase's 81
The phase
structure
of tense
CPo We have already seen that root modals can scope under
PERF
in
(1 5); ( l 8)
gives ·
another example. In this case, the modal must be associated with the V phase's CPo This
is exactly the kind of contrast we are looking for.
( 1 6)
( 1 7)
You should have bought that book when you had the chance Max ought to have kept his mouth shut at the meeting (Stowell forthcoming; his (22a.b»
( 1 8)
To be eligible for this loan,
you
must never have been turned down for credit
before
As Stowell notes, it is 'more plausible to suppose that in
[( l�17)]
the relevant
deontic obligation held at the past times in question, rather than obtaining at the utterance time (obligating the subject at the utterance time to have arranged things in the past in a particular way)' (p. 23). In ( 1 8), on the other hand. it is clear that the requirement for the subject never to have been refused credit does hold at the utterance time. Given that the modals in all of ( 1 �1 8) are clearly root interpreted. we may conclude that modals
are able to appear not just in the CPs of the
V
and T phases. as claimed above, but also
in the CP of the PERF phase.8 This then is more good evidence that this CP exists, and moreover that is is fundamentally equivalent in its function to the CPs of the T and
phases: that is, that these CPs are just as much phases on the current analysis as the and
V phases.
It is harder to use the same kind of reasoning to show that the
V T
CP of the PROG
phase exists, since the temporal relation5hip at stake (present rather than past) doesn't allow such an obvious distinction to be made; but it seems entirely reasonable to infer that if the PERF CP exists. then so too does the PROG CPo
7. 0utro
I have argued here that phases should be defined in terms of quantificational closure: the
T and V
phases (standard CP and vP phases) both involve quantificational (usually
existential) closure over situation variables; the putative N phase (standard
DP),
though
I haven't discussed this here, involves quantificational closure of individual variables (or perhaps sometimes situation variables, say with gerunds). More specifically, I have argued for the general structure to represent this shown in ( 1 9).
CP
(19)
�
3i
hP
h�
S ITUATIO N i
HP
6.
81t may possible to fonnaJize on this basis the distinction between what are known in the literature as 'ought to be' and 'ought to do' root modals, along the lines of: 'ought to be' modals are those where the modal scopes over aspect, but under the subject (thus having both roOI and epistemic propenies)� 'ought to do' modals are those where the modal scopes under aspect.
82
Jonny Butler
A Root head H is at the core of each phase; this is basically a property denoting category, telling us what property is going to be predicated. The predication itself is introduced by I-syntactic little heads h above H. Each of these introduces a sub-situation, which then compose to give a macro-situation variable s, this being something like a generalized Davidsonian event argument. The I-syntactic h layer basically spells out event structure syntactically, and any arguments of the predicate so derived are introduced as specifiers of the little heads. This is topped off by a CP layer, which is quantificational in nature. serving to close off variables introduced below. for temporal (§4), aspectual (§5), and modal (§6) interpretation. This CP layer corresponds to the EDGE of the phase, everything below is its DOMAIN. This story ends up deriving us more phases than the classic two (three if we count DP): we are led to propose that (outer/viewpoint) aspect (perfect/progressive) should be analysed along the same lines. with a Root PERF/PROG each introducing a temporal predicate and situation argument thereof, exactly parallel to T. Is this to be seen as a problem? No: the usual tests for phasehood. inasmuch as they work and are coher ent (see Matushansky 2003 for a critique), generally relate to matters of argumenn:E placement/interpretation-reconstruction. QR, expletives, etc (cr. Legate 2003). all these things really boil down to, though, is quantification: quantificational effects justJif happen to be more visible where you have a visible DP argument. Where you don't have �i!! ' such an argument (as far as interpretation is concemed)-e.g. in the aspectual phasesyou don't see the same effects DP-wise, but you do see the same effects more generally: i.e. there must be quantification going on in order for the temporallaspectuallmodal inter· pretation to come out right (§6).
Whar:�1
References Belletti, Adriana (200)). &InvelSion' as focalization. In Aafke Hulk & Jean-Yves Pollock {eds.} Subject Inversion in Romance and the Theory oJ Universal Grammar. Oxford: Oxford Univer sity Press. pp. 60-90 Brody. Michael & Anna Szabolcsi (2003). Overt scope in Hungarian. Symax 6: 1 9-5 ) Butler, Jonny (2003a). A minimalist treatment of modality. lingua 1 1 3: 967-996 Butler. Jonny (2003b). Sequence of tense phenomena and non-eventivity. Ms .• University of York Butler, Jonny (2004a). On having arguments and agreeing: Semantic EPP. York Papers in linguistics (series 2) 1 : 1-27 Butler. Jonny (2004b). Phase Structure, Phrase Structure, and Quantification. Doctornl disserta tion. University of York Cardinaletti , Anna ( 1 997). SUbjects and clause structure. In Liliane Haegeman (ed.) The New Com parative Syntax, London: Longman, pp. 33-63 Cardinaletti, Anna (2000). Towards a cartography of subject positions. Ms., SSLMIT University of Bologna - University of Venice Cardinaletti, Anna & Ian Roberts ( 1 99 1 ). Clause structure and X-second. Ms., University of Venice - University of Geneva Cardinaletti, Anna & Michal Starke (1 999) . The typology of structural deficiency: A case study of the three classes of pronouns. In Henk van Riemsdijk (ed.) Clirics in rhe Languages oj Europe: Empirical approaches 10 language typology, Berlin: de Gruyter. pp. 145-233 Chomsky, Noam (1 999). Derivation by Phase. MrrOPL 1 8 Chomsky. Noam (2001). Beyond explanatory adequacy. Ms., MIT Comrie. Bernard ( 1 985). Tense. Cambridge: Cambridge University Press Condoravdi. Cleo (200 1 ). Temporal interpretation of modals: Medals for the present and for the past. In David Beaver, Stefan Kaufmann, Brady Clark, & Luis Casillas (eds.) Stanford Papers on Semanrics. Stanford: CSLI Publications, pp. 1-30
83
The phase SlrUcture of tense
Cook. Walter A. ( 1 978). Semantic structure of english modals. TESOL Qua rterly 1 2: 5- 1 7 Davidson. Donald ( 1 967). The logical form of action sentences. I n N . Rescher (ed.) The Logic oj Decision and Action. Pittsburgh: University of Pittsburgh Press Demirdache, Hamida & Myriam Uribe-Etxebarria (200 I). The primitives of temporal relations. In Roger Mnnin. David Michaels. & Juan Uriagereka (eds.) Step by Step: Essays on Minimalist Syntax in Honor ofHoward Lasnik, Cambridge. Mass.: MIT Press. pp. 1 57- 1 86 Diesing. Molly ( l 992). lntiejinites. Cambridge. Mass.: MIT Press Hale. Ken & Samuel 1. Keyser ( 1 993). On argument structure and the lexical expression of syntactic relations. In Ken Hale & Samuel J. Keyser (eds.) The Viewfrom Building 20: essays in linguistic.� in honor of Sylvain Bromberger. Cambridge. Mass.: MIT Press. pp. 53- 1 09 Hallman. Peter (2000). The Structure of Predicates: interarrions ofDerivation, Case and Quantifi cation. Doctoral dissenation. UCLA Harley. Heidi (I 995}. Subjects, events, and licensing. Doctoral dissertltion. MIT Heim, I rene ( 1 982). Tile Semantics ofDefinite and Indefinite Noun Phrases. Doctoral dissenation. University of Massachusetts at Amherst Heim, Irene & Angelika Kratzer (1 998) . Semantics in Generative Grammar. Oxford: Blackwell Hom. Laurence ( 1 989). A Natural History ofNegation. Chicago: University of Chicago Press Hornstein. Norben ( 1990). As Tune Goes By: Tense and Universal Grammar. Cambridge. Mass.: MIT Press Jayaseelan. K.A. (2001 ). IP-internal Topic and Focus Phrases. Studia Linguistica 5 5 : 39-75 Kratzer. Angelika ( 1 977). What 'must' and 'can' must and can mean. Linguistics & Philosophy I : 337-355 Kratzer. Angelika (1 98 1 ). The notional category of modality. In H.J. Eikmeyer & H. Rieser (eds.) Words, Worlds, and Contexts: New Approaches in Word Semantics, Berlin: de Gruyter. pp. 38-74 Kratzer, AngeJika ( 1 99 1 ). Modality. In A. Von Stechow & D. Wunderlich (eds.) Semantics: an Inremational Handbook ofComemporary Research. Berlin: de Gruyter. pp. 639-650 Kratzer. Angelika ( 1 995). Stage and individual level predicates. In Greg N. Carlson & Francois J. Pelletier (eds.) The Generic Book. Chicago, III.: University of Chicago Press. pp. 125-175 Legate. Iulie Anne (2003). Some interface p rop erties of the phase. linguistic Inquiry 34: 506-5 16 Manzini. M. Rita & Leonard M. Savoia (2002). Parameters of subject inftection in Italian dialects. In Peter Svenonius (ed.) SUbjects, Expletives, and the EPP. Oxford: Oxford University Press. pp. 1 57-199 Marantz. Alec (200 1). Words. Ms., MIT Matushansky, Ora (2003). G oing through a phase. Paper presented at the lAP Workshop on phases and the EPP, MIT, January 2003 May, Roben ( 1 985). Logical Form in Natural Language. Cambridge. Mass.: MIT Press Milsark. Gary ( 1 974). Existenrial Sentences in English. Doctoral dissertation. MIT Palmer. frank ( 1 990). Modality and the English Modals. 2nd edition. London: Longman Pesetsky, David ( 1 995). Zero Syntax: Experiencers and Cascades. Cambridge, Mass.: MIT Press Pesetsky. David & Esther Torrego (200 1 ). T to C movement: causes and consequences. In Michael Kenstowicz (ed.) Ken Hale: a life in language. Cambridge, Mass.: MIT Press. pp. 355-426 Pollock, Jean-Yves ( 1989). Verb movement, Universal Grammar and the structure of IP. Linguistic Inquiry 20: 365-424 Portner, Paul ( 1992). Situation theory and the semantics of propOsitional expressions. Doctoral dissenation. University of Massachusetts at Amherst Pylklciinen. Liina (2002). Introducing Arguments. Doctoral dissertation, MIT Reichenbach. Hans ( 1 947). Elements ofSymbolic Logic. London: Collier Macmillan Rizzi. Luigi ( 1 997). The fine structure of the left periphery. In Liliane Haegeman (cd.) Elements of Grammar: Handbook in Generative Syntax, Dordrecht: Kluwer. pp. 281-337 Ross. JOM ( 1 969). Auxiliaries as main verbs. In W. Todd (ed.) Studies in Philosophical Linguistics, series one. Evanston. Ill.: Great Expectations Press. pp. 77-102 Sponiche. Dominique ( 1 996). Clitic constructions. In Johan Rooryck & Laurie Zaring (eds.) Phrase Structure and the Lexicon. Dordrecht: Kluwer. pp. 2 1 3-276 Sportiche, Dominique (2002). Movement types and triggers. paper presented at the TILT workshop. GLOW 2002, Utrecht Slarke. Michal ( 1 993). En deuxieme posi tion en Europe Centrale. memoire, University of Geneva Starke, Michal (2001 ). Move Dissolves into Merge: a theory of Locality. Doctoral dissertation.
84
Jonny Butler University of Geneva Stowell. Tim ( 1 99 1 ). Alignment of arguments in Adj ective Phrases. In Susan Rothstein (ed.) Per sperrives on Phrase Structure: Heads anti Licensing. Syntax and Semanlics 24: 105- 1 35 . San Diego: Academic Press Stowell. Tim ( 1 996). The phrase structure of tense. In Johan Rooryck & Laurie Zaring (eds.) Phrase Structure and the Lexicon. Oordrecht: Kluwer, pp. 277-29 1 Stowell. Tim (2004). Tense and modals. In Jacqueline Guemn & Jacqueline Lecarme (eds.) The Syntax ojTime. Cambridge. Mass.: MIT Press. pp. 621-637 Vikner, Stcn ( 1 985). Reichenbach revisited: One, two. or three temporal relations? Acta Linguislica Hafniensia 1 9: 8)-98 Zagona. Karen ( 1 990). Times as temporal argument structure. Ms . • University of Washington. Seat tle
Universitat Stuttgart Institut fUr Linguistik: Anglistik KeplerstraBe 17/4b (KII) D-70174 Stuttgart Germany
[email protected]
85
A phase-geometric approach to multiple marking systems
·
Andrew Carnie University ofArizona
Multiple marking systems refer to case, agreement. and word order patternings that make usc of more than two positions. agreement patterns or case markings, such as the phenomena of split ergativity and differentia) object marking. It is proposed here that such systems arc sensitive to a particular version of phase theory. In particular it is argued that each phase consists of a single argument, the predicate that introduces it and a temporal operator. Since each phase is independently sent to the LF and PF interfaces, it is claimed each phase contains a restrictor and a nuclear scope for the argument in question. It is mapping to these phase internal scopa) positions that gives rise to the appearance of argument (accessibility) hierarchies that seem to govern multiple marking systems.
1. Introduction Chomsky (2000) I defines a "strong phase" as either a vP or CP; the two functional projections he identifies as "complete propositions". While this characterization has had some important empirical and theoretical results (see for example the other papers in this volume), there is at least one conceptual problem and a family of empirical problems with this claim. I mention the conceptual problem here, and then wiJ) not talk about it again. The vP and CP do not, in fact, both represent " complete propositions" in any uniform sense of the words "complete" or "propositions". They are quite different animals. The vP represents a verb and its arguments. By contrast, the CP represents a speaker's intentions, beliefs and attitudes towards the predication, along with a temporal operator that locates the utterance relative to the speech time. While these a11 make important contributions to the truth conditions of the sentence, the two kinds of phase represent very different kinds of "complete propositions." It isn't at al1 clear (to me at least) what they have in common. Turning now to the family of empirical issues. Consider the predictions made when we assume both Chomsky's (2000) definition of phase, along with a Diesing (1 992)-style tree-mapping analysis of specificity effects (defined where the VP is the domain of existential closure.) In such a system, specific and definite NPs must have raised out of the VP, so that they the map to the �, Andy Barss, Tom Bever. Phil Cash Cash. Heidi Harley, Eloise • J would like to thank Judith Aiss Jelinek, Simin Karimi, Terry Langendoen. Martha McGinnis, Norvin Richards and the audience at the MIT workshop on Phases and the EPP for their input on this paper. As usual. any errors in this paper are the fault of my cat, Pangur. I
See also Chomsky (2001a. b).
MIT Working Papers in Linguistics 49. 87-102
Perspectives on Phases
© 2005 Andrew Carnie
A phase-geometric approach to multiple marking systems
restrictor. In phase theory, this means that only the vP phases have a domain of existential closure. There can only be one domain of existential closure per clause, simply because the vP phase is sent to LF before the CP phase comes into play.
Barss and Carnie (2003) address this problem with respect to English existential constructions and adverbial p lacement. They assume a typical view of Phase theory, where there is no covert movement; relationships previously taken to be covert movement are checked through the Probe-Goal relation (via AGREE). This means that the surface position of a phrase also marks its position with respect to scopal relations.2 Consider the sentence in ( I ):
1)
The woman always drove her car with her gloves on.
Taking temporal adverbs to mark the left edge of the vP (Emonds 1 976), we have a definite NPs surfacing in the nuclear scope. Such a structure can't be rescued without covert movement. A similar problem is seen in the example in ( 1 )3:
2)
There was some guy kissing the linguist (when I walked into the room).
This sentence is acceptable with a non-specific reading of some guy, taken to be the canonical reading of a there existential sentence. In this sentence we have a clear marker of the vP edge: some guy. We can thus also conclude that vPs mark the top end of the nuclear scope of the clause. What is surprising here is the acceptability of the sentence given the presence of the definite the linguist, also presumably within the nuclear scope. Under standard assumptions about how the mapping principle works (see Diesing 1 992, and much subsequent work), specific indefinites and definites such as the linguist should raise out of the vP in order to escape the nuclear scope. In versions of the minimalist program prior to Phase Theory, this was accomplished by covert raising of the DP. In this version of Phase theory, however, this option is not available. Word order considerations alone demonstrate that if we assume that surface order fixes scope and there are
no covert operations, then it is impossible for the linguist to have moved outside the nuclear scope, yet have some guy remain inside it (3) (an arch here indicates the nuclear scope. The first phase is the vP and all it dominates.)
! Wh-in situ, Quantifier scope ambiguities and other phenom ena commonly analyzed as covert
movements might seem at first to be prima-facie counterexamples to this claim. However, it isn't at
all clear that covert movements arc the appropriate analyses of such phenomena in any case. For
example. Cooper-stack analyses of quantifier scope am biguiti es seems to have a better account of
the full range of the phenomena - see Sag, Wasow and Bender (2003) for discussion.
3 Sec Barss and Carnie for extensive arguments that these constructions do not involve a reduced
relative clause rather than an associate/tail existential construction as is crucial to the story here.
88
Andrew Carnie
TP
3)
,...,.� ... T'
There
T
�
When the LF for the lowest phase is created, there is no way for the linguist to move to create an appropriate variable. That is, on a theory with no covert movement, the Mapping principle is seemingly violated, since the definite OP the linguist has to be interpreted in situ, and it is within the nuclear scope. :,' Fol1owing in the spirit of Percus ( 1 993), and building upon Barss Carnie (2003), and Jelinek and Carnie (2003), I propose that the definition phase is relativized to each argument as set out in (4). Each phase consists of an �ir argument, the predicative element that introduces the argument (V or vP) and a '[f functional category that represents a temporal operator that locates the predicate ' in time or space (Asp, T, etc.)
amtil o�m
1:
1f
(4)
Phases consist of: a. a predicative element (v or V) b. a single argument c. a temporal operator that locates the predicate and argument in time and space (Asp or T)
This gives rise to phases such as those seen in (5)4 :
(5)
Theme Phase Goal Phase Agent Phase
a. b. c.
[.�spP [Asp' Asp [vp t h e m e V]]J [EftdP [End' En d [v go a l [v' v . J]]]5 [TP h T [vp agent [v v . ]]]
..
..
The clausal architecture created by such phases is essentially that of Travis' ( I 991) inner aspect approach:
4 I am assuming here. of course. a Hale and Keyser (1 992. 1 993, 2002) analysis of theta roles. where
theta role is detennined by syntactie position.
J also leave aside the question of how CPs fit into this
system. Speculatively, the is a fundamental difference between CP and phases and the phases in (5).
such that CP phases define A-bar relations. whereas those in (5) represent A relations. I'm not
prepared at this time to show how phase edging would work under such a conception. 50 I leave it for future work. S The functional category End is meant to represent a temporal operator that detennines
grammatically defined aktionsanen. Sec Travis (2000). Borer ( 1 994) and van Hout (2000) for discussion of the motivation for such a functional category. The ordering oftbis phase relative to the Theme phase
is Wlimportant to the story presented here.
89
A phase-geometric approac h to multiple marking systems
(6) TP
/". /". T
AGENT PHASE
vP
/". agent /". v
EndP
GOAL PHASE
/". En
('
vp
/". goal � v
spP
THEME PHASE
/". /".
Asp
VP
/".
theme
Vi
Each of these phases has its own domain of existential closure (the vP or VP) and it's own nuclear scope. This means that the interpretation of each NP wilJ be logically independent of its clausemates. Using relativized phases, I am going to argue for an account of multiple (or split) marking systems, where the split is triggered by some kind of specified or markedness hierarchy effect. The most familiar kind of split system are split case systems; although I use the term "marking" systems to extend the empirical domain to agreement and word order marking systems. This includes both differential object marking (DOM) (Aissen 1 999, 2003) and differential subject marking (DSM). In the next section, I tum to a brief description of such systems.
2. The Larger Goal: Differential Marking Systems. There are at least two distinct notions of 'hierarchy' present in grammatical theory today. One is the hierarchical constituent structures created by multiple applications of the Merge operation. The other conception, which is more prevalent in the typological literature, but sometimes bleeds into generative work as well, concerns relational or markedness hierarchies. The so-called theta hierarchy (see, for example, Grimshaw 1 990) is perhaps the most salient of these markedness/relational hierarchies. Scholars working outside the realm of mainstream Indo-European languages also have to draw upon other less well known hierarchies. For example, argument ordering in Navajo apparently governed by an animacy hierarchy, such that animate arguments precede inanimate ones (independent of their thematic or grammatical relation); the interpretation of the nominals is determined by active/inverse marking on the
90
Andrew Carnie 6 verb (Willie 199 1 ). Similarly, the vast literature on split ergativity has shown the importance of person and number hierarchies with respect to ergative/absolutive split case-marking systems. Typically this second kind of hierarchy is presented as a ranked list. Some example of such rankings are given in (7): (7)
Animate > inanimate Definite > indefinite specific > indefinite non-specific Istl2nd person > 3rd person Topical > focal Agent > Experiencer > Goal > Theme > InstrumentlBenefactive Subject > Object > Indirect object
a. b. c. d. e. f.
There are two important things to note about these kinds of hierarchies. First, they represent tendencies rather than hard and fast rules. It is possible to ,� override a markedness hierarchy with, for example, a particular morphological marking or operation (e.g., special voice marking, clefts, scrambling etc.),;l¥ Second, as pointed out to me by Marantz (p.c.), it isn't at all clear what thell ontological status of such hierarchies is for an app�oach such as generative.jIi grammar. They aren't rules; they aren't constraints. Nevertheless they have ' widespread exponence in the syntax of the world's languages. In this paper (and other work), I aim to reduce all independent semantic/relational/accessibility hierarchies to the constituent hierarchy, whereby appearing to the left on the relational hierarchy corresponds to positioning high in the constituent tree. This approach to hierarchies is not particularly new. Hale and Keyser ( 1 992, 1 993. 2002) claim that thematic roles can be directly be determined by the argument's initial position in the constituent tree. The "thematic hierarchy" is an artifact of the fact that, for example. agents are generated higher in the tree than themes. Jelinek (1 993) presents a similar account of split ergativity in Lummi, exploiting Diesing's mapping principle. 1st and 2nd person subject pronouns are inherently specific, so must appear outside the nuclear scope, which results in a
jjl !
f r
I
6 See for example, Abraham ( 1 996),
Bittner and Hale ( 1996a,b), Bobaljik ( 1 993), Bok·Bennema
(199 1 ), Campana (1 992), DeLancey (198 1 ), Dixon (1 972, 1 979, 1 994), DuBois (1 987), Hoopa- and Thompson ( 1 980),
Isaak
(2000),
Jelinek ( 1 993), Johns
( 1 992), Levin (1 993),
Levin
and
Massam
(1 995), Murasugi ( 1 992). Silverstien ( 1 976), Ura (200 1 ), and Woolford ( 1 997) to name just a
sampling .
7 Aissen ( 1 999) and (2003) attempts to recast these hierarchies as universal constraint rankings iII
OptimaJity Theory. There are,
however, a number of significant problems with her account First, it
relies significantly on morphological markedness constraints, where unexpected mappings between
relational hierarchies must be realized with overt morphology. This is contraindicated by the fact
that such mismatches can be realized with non·morphologieal means, such as word order. Second, . some of her crucial arguments are based on a faulty empirical characterization of Dyirbal (she
assens that Dyirbal Nom/Ace patterns are unmarked for casc-this is false , class markers in n;'irbal
arc overtly marked for case). Finally, her constraint rankings are "derived" via hannonic alignment
from the relational hierarchies themselves. This requires that the
I. !
i
,
re lational hierarchies be some kind
of primitive whose ontological status is still in doubt. Similar proble�s can be found
with
the
systems described by Isaak (2000).
91
A phase-geometric
approach to mUltiple marking systems
nominative argument marking. 3rd person subject pronouns in Lummi are, by stipulation, non-specific and thus stay inside the VP. where they get an inherent ergative case. Diesing and Jelinek ( 1 995) extend this analysis to phenomena in other languages that appear to be sensitive to relational hierarchies. Under this kind of approach, we can derive certain kinds of split ergative system as schematized in 1 0. This approach follows a Murasugi ( 1 992) nested paths approach to ergativity. Under this view, presuppositionaJ, topical, definite, animate subjects -the relevant criteria depending upon the language-raise out of VP to create a variable within the nuclear scope, giving rise to a nominative accusative pattern as in (8a). (8)
a.
Nominative/Accusative pattern: "highly presuppositional subject " TP -----NOM T'
v' domain of 3 ' -----v' ACC v----VP ---
-------- V
b.
Ergative/Absolutive Pattern: "Asserted subject " TP -----ABS T'
domain of 3
v
�
-----t
V
By contrast asserted (non-specific, indefinite, inanimate) subjects are licensed within the VP, as seen in (8b). The agent receives a lexical ergative case and thus remains within the nuclear scope. This ergative case presumably blocks accusative checking. As a matter of last resort the object raises to the specifier of s TP where it gets nomlabs case . K The tree in (8b) seems to have a obvious violation of Shortest Move or the Minimal Link
Condition . The simplest solution to this problem is to say that inherently case-marked
92
Andrew Carnie
While this sort approach has a number of appealing qualities. it also na: some significant defects. First, in the system in (8), it appears as if the eventual. LF is driving the syntax. Second this system doesn't allow for cases witn identical interpretations of multiple arguments; for example, cases where tn.� subject and object are both interpreted presuppositionally or are both interpreted as novel. This can be seen by looking at the trees in (8), if the both the object and the subject are asserted, then they should both remain inside the VP. In such a situation, either the formal Case requirements of the object are not met resulting in a syntactic crash or the argument doesn 't appear in the correct position for interpretation, resulting in semantic anomaly. Similar effects would be found when both the subject and the object are topical. Such interpretations do occur, as seen in Nez Perce data in (9): (9)
a.
b.
Haama +0 man+0
hi+'wi+ye
wewiikiye
elk 3+shoot+asp "A man shot an elk" wewUkiye+ne pe+'wi+ye Haama+nm Man+NM 3/3+shoot+asp elk+NE "The man shot the elk" (modified from Woolford 1 997)
Third, while it captures "simple" cases of ergative splits, where nominativ; always co-occurs .with accusative and ergative always co-occurs with absolutiv� it fails to account for more complicated 3 and 4 way systems, where for exampJ;.. ergative and accusative co-occur, as is found in languages such as Dyirbal (Dixon 1 972), Finally, although it does so in a disguised form, it maintains tn� idea that there are two separate hierarchies in the system: a semantic one (tn� LF) and a syntactic one; the two hierarchies are related by a mapping relation. in an ideal situation these two kinds of hierarchies should be homomorphic. All but the first problem disappear in a system with relativized phasing. Problems (2) and (3) vanish because the interpretations of nominals are largely independent of one another, as each has its own phase. Problem (4) disappears in any phasin�. system where there is no covert movement, as . surface position will determin: scope, The first problem (the fact that interpretation drives syntax), does not necessarily require relativized phasing, but is easily overcome using another aspect of Chomsky ( 1 99S)-style minimalism: little v categories. Provided that we have an explicit enough set of lexical entries for functional categories and predicate categories, these lexical entries not only select for arguments but also detennine the functional structure that licenses the arguments, thus determininj! surface position and relative scope.
arguments-with perhaps Icelandic Quirky case marked arguments being the exception-arc not candidates for movement,
thus don't create intervention effects.
93
A phase.geometric approach to multiple marking systems
3.
Multiple Marking Systems
The relativized phase approach, when it is coupled with an explicit theory of formal features and lexical entries for functional categories, provides an explanation for the existence and range of variability in multiple argument marking systems.
Ergative Split in Dyirbal
3.2
Consider first the case of split case marking in Dyirbal. I will be discussing here a "simplified" form of Dyirbal, in that I'm only trying to account for the range
of phenomena already accounted for by other approaches. I'm not trying to account for all the case marking patterns that exist in the language. (See Dixon ( 1 972) for discussion ofthe fuller range of data). 2nd Dyirbal case marking is split along the lines of person . 1 $I and person subjects exhibit a nominative accusative pattern, 3M person subjects take an ergative/absolutive pattern. I ' m following Jelinek's analysis of person splits 9 here where 1 st/2nd subjects are more "presuppositionaI " than 3 M person subjects, which are typically asserted.
( 10)
a.
b.
c.
d.
banaga-nYu. retumed-NONFUT "You (pI) returned." nyura-0 Nana-na bu} a-no see-NONFUT I PL-ACC 2PL-NOM "You saw us." Numa-0] banaga-nYu. [bayi cl ass l .ABS father-ABS retumed-NONFUT "Father returned." [bayi Numa-0] [baNgun yabu-Ngu] class l .ABS father-ABS class2.ERG mother-ERG " Mother saw father." nyura-0
2PL-NOM
bu}a-n sawNONFUT
One means of encoding this type of split is through the use of different little v categories for introducing 1 /2 vs. 3rd subjects. This is a reasonable approach to take, since we will need to lexically specify that T subjects take a lexical ergative case in any event. It thus seems reasonable to assume tI�at these different functional categories also have different subcategorizational properties. The v that introduces 1 12 subjects subcategorizes for an accusative case assigning AspP, whereas the v that introduces :r subjects (as well as the ones that introduce passives and intransitives) selects for a defective AspP that lacks an accusative case feature. This is encoded in the lexical entries in ( 1 1 ) .
9
The nature of this "presuppositionality" and "asscrtedncss"
is of course a crucial open question. I
leave it as an open question requiring a precise formal characterization once the phenomena have been more thoroughly described and better understood.
94
Andrew Carnie
(1 1 )
a.
V3
<.J.. Asp'>
b.
Vll2
d.
Asp·
I
<.lLl. Aspa>
[+erg]
c.
Vinlrlln
< 1 /2/3, Asp�
, [+ACC]
< [+V] >, There are some important subtleties to these lexical entries. First consider tr. major similarities and differences between AspB and Aspd. The first thing to note is that, despite the fact that these are Aspect nodes, I have underspecified tn,.. semantic contribution of the head. Presumably each kind of Asp could itself vary in terms of the actual aspectual semantics it contributes. I'm concentratin! 10. a here only on the syntactic contributions of the Asp head Asp is the form the aspect marker that would be found i n nominative/accusative languages witli all transitive verbs. The transitivity here is formalized i n assuming that VP,; obligatorily branching. With respect to this functional projection, the signifi c8i.l difference between a Nominative/Accusative language like English and a split. one like Dyirbal, is that selection for this category is limited to the v categorY. marked for 112 person in a Dyirbal-like split ergative system, and but is found with all transitive vs in nom/acc languages. Next consider Aspd. This is the functional category that would nonnahy be found with intransitive or passive little v in nominative/accusative language� This aspectual head differs from Aspa in two crucial regards: First, it lacks tn,.:. ability to assign accusative case. Second, instead of specifying a VP as ItS complement, it selects for the more general [+V] category instead. [+V] includes both VPs and V heads with no complements (which are presumably just bar= verbs in the bare phrase structure system.) This allows it appear both in simp!� intransitive constructions and in passives in nominative accusative languages. What is unique to the split system of Dyirbal (and languages like it) is that this form is also used when V3 is present and the subject gets a lexical ergative case. Next, let us see how these lexical entries, combined with the relativized phases I proposed above explain the interpretive distinctions found in Dyirbal capturing JeJinek's insights. First, consider the nominative/accusative pattern that arises with 1 /2 person subjects. For reasons of space, I have combined the two phases into a single tree in (l 2). The phase boundary is marked with a
ai, !i
10 One might expect a multiplication in the number of aspecrual functional categories. in such a way that either significant semantic or significant syntactic generalizations might be missed, (i.e. if one is forced to posit separate perfective Aspd and Asp', and imperfect ASpd and Asp' and progressive Aspd and Asrf functional categories, then one seems to be missing the generalization that there are two kinds of syntactic BSpect (those that assign accusative case, and tbose that do noi), and three (or more) kinds of semantic distinctions. This kind of problem disappears if one assumes that the lexicon is defined in terms default inheritance hierarchy of the kind found in HPSG. Such an assumption is not at all incompatible with a minimalist Phase Theory analysis of the licensing, movement and interpretation or even a OM lexicon.
95
A p hase-geometric
approach to multiple marking systems
strai �ht line. The nuclear scope of the agent phasc is indicated with a curved 1 line . ( 1 2) Agent phase
Aspap Theme phase :::------Ace Aspa,
�
------
pa [AC C]
VP _______ theme V
In th is structure, the V ll2 is used. This v does not assign a lexi cal case, so the agent must move to the spec ifier of TP for case licensing, this means that, on the surface, it appears in the restrictor portion of the phase. VI12 also selects for
Aspa p. This aspect node assigns checks an accusative case feature in its specifier. Consider now the tree in ( 1 3), which represents an ergativelabsolutive structure. V3 inherentl y marks its agent with an ergative case, so the remains
inside the vP (and thus is interpreted non-presuppositionally). Note that the pragmatics are not detennining the movement here (unlike, for exampl e , Jelinek 1 993); instea d the case licensing detennines surface position, which in turn -detennines how the argummt is m�ped to infonnation structure after LF. Like Vinran. and Vpassive, v) selects for Asp P. This aspect node has no accusative case feature to check, so a s a matter of last resort. the NP must move to �et licensed in the next phase,
r esulting in Nominative (Absolutive) case checking. 2
II I have not indicated here the scopal properties of the embedded clause. How differing interpretations of objects work in Dyirbal requires additional data and study. 12 The tree in ( 1 3) and others in this paper ignore the Phase Impenetrability Condition. There is a paradox raised by the interaction of formal liccnsing fC(juirements with semantic requirements on the phase's LF. In split ergative systems, semantic considerations indicate that asserted OPs remain VP internal. but formal requirements (the PIC) force them to move out of nuclear scopc to the edge so that they can case check with TP in the higher phase. This either suggests that the whole enterprise I have sketched here is misguided, or we have to rethink the PIC. One possible compromise is to is assume that we don't actually merge all of a lexical items features on its first instance. Instead we merge only formal features. This is consistent with many recent proposals recasting Move in tenns of Merge (see for example Lasnik 1999). The trace, which never has phonological content, isn't linearized during the phase's PF. This means that when the next phase accesses the structure, it can target any null argument without adjusting the PF content of that lower phase. As such, edges become irrelevant. One way to encode the requirement
96
Andrew Carn ie
( 1 3)
AgentphasE Nom
3.3
Nez Perce 4-way case s ystem
In this section, we look at a different kind of situation. One where the behavior of each phase interacts with those above it. This data comes from the complicated case marking patterns found in the Sahaptian language Nez Perce, spoken in Idaho. This section is a partly updated version of the analysis 1 3 given in Cash Cash and Carnie (2003). Consider the data in ( 1 4).
( 1 4)
a.
b.
c. d.
Haama +0 hi+'wi+ye 3+shoot+asp man+0 "A man shot an elk" Haama+nm pe+'wi+ye Man+NM 3/3+shoot+asp
wewUkiye
t?J/t?J pattern
elk wewUkiye+ne
ERGINEpattern
elk+NE
"The man shot the elk" * NM/0 (*The man shot an elk) * 0INE (*A man shot the elk)
that traces not be phonologically realized would be a condition on the operation SPELLOUT, such that it would only merge phonological features into the structure when all the uninterprerabJe features have been checked (i.e., phonological
features could only be merged at the top of a
movement chain), otherwise the node is left null. Of course, this kind of proposal creates many marc questions than it answers. For example, work will need to be done to recapture the economy
character of the PIC-perhaps in terms of scope in chains. Similarly, I will eventually need an
account of those
situations where the phonological material is inserted into a lower member of the
chain (assuming such cases as wh in situ or resumptive pronouns do in fact involve a movement chain). Nevertheless, I hope that this kind of proposal provides a ftuitful alternative view about how a phasaJ theory of syntax might work. 13 For alternative analyses of Nez Perce, see Rude (1982, 1985, 1 986a.b. 1988) and Woolford
( 1 997).
97
A phase-geome tric
approach to multiple marking systems
There are a coup le of imp ortant things to note about these sentences. First, there is an interpretive difference between ( 1 4a and b), which in turn corresponds to a difference in case marking Note also that the case marking of the two nominals is linked. Mixed patterns are not anowed ( 1 4c and d). Next note that that the two patterns have di fferent agreement marki ng Fina lly note that is the fact that unlike many other languages, the so-ca lled erga tiv e case here is liked to specific/definite subjects and objects, not asserted ones. This suggests perhaps we are looking at a very different phenomenon from the kind of split ergativi ty found in languages like Dyirbal or the Polynesian languages. For this reason, I propose quite a different analysis of Nez Perce case marking. In particu lar I cla im that the traditional labels for these case markings are quite misleading, the l4 case marked -nm isn't really an ergative, but a nominative , and the case marked with a 0 is really an ergative! A s imi l ar switch in names is appropriate for the two obj ect cases of Nez Perce. Ne is traditionally called the "objective", and the «0 is called acc usati ve (for example, by Rude 1 95 and Woolford 1 997). In the system I propose here -ne is actually an accusative case, and the 0-marked obj ec t isn't case marked at all (in a sense to be defined below). A chart summarizing these names is given in ( 1 5). .
.
,
"
,
"
,
-
"
"
( 1 5)
markil!g 0 -nm 0 -ne
relation
traditional name
subiect
Nominative
Name eiven here Ergative
subject object object
E rgati ve
Nominative
Accusative
not case marked
Obiective
Accusative
While this has the potential to confuse, and smacks slightly of l inguistic revisionism, I believe it more accurately describes the case markings. Where confusion might arise, I give the Nez Perce morphology as a guide. The % pattern ( 1 4a) is, in many ways, similar to Nez Perce intransitives ( 1 6), in that for example it takes an intransitive agreement morphology ( 1 7), an d the subjects of intransitives o�l igatori ly take -0 case. ( 1 6)
ipi +0 he+0 '
hi +kll +ye
3 +go+asp
" (Data from Rude 1 982)
"He went"
(1 7)
Intransitive and 0/0 NMINE
3 hi
1 ,2 0 1 ,211,2 0
15
113 'e
2/3 'aw
3/3 pee
3/1,2 hi
14 One possible objection to this is that -nm also marks possessives, a typical property of ergative case markers. I leave this objection aside here. IS
The slash notation here means Subject/Object, so 2/3 means 2nd person subject,
etc.
98
3rd person object
Andrew Carnie l�
Cash Cash and Carnie claim that the object in ( 1 4a) is an NP rather than a Dp . and in the spirit of Massam (200 1 ) is thus caseless (it is "pseudo-incorporated"). Thus giving an intransitive pattern. These NPs have no case checki n � requirements, thus remain V P internal, and are subsequently interpreted a s novel information, from the perspective of the rest of the clause such VPs ar� essentially intransitive. To see how this works, let us again set up lexical entries for the vario!:- flavors of little v found in the language. Notice that the variation is alon: transitivity lines. (I 7)
a.
b.
Starting with the % pattern, we can see how the reJativized phasin: approach, combined with these lexical entries derives the patterns in ( 1 4). Th� agent marked wilh lexical --0 ergative case remains in situ in the nuclear scope.� since it doem' t have to raise for case licensing. The theme is a pseudc-;' ,� incorporated NP, and thus also remains within its own VP. Both arguments alj; interpreted as novel asserted information.
'�\j:
( 1 8)
ft}/f!J Pattern
Agentphase
Contrast this to the NMINE pattern. In these cases, transitive v is used. This v does not assign a lexical case. So the agent argument must raise to the specifier of TP for case checking. This results in it (obligatorily) being interpreted presuppositionally. This v selects for Aspap. The theme, being a DP (requiring) case moves to the specifier of ASpR for case checking, and thus also moves out of its own nuclear scope.
16
This is a bit of an oversimplification of Cash Cash and Carnic's accoWlt and of the data. Indeed.
certain kinds of constructions that may well have determiners, such as possessive-possessed case marking, nevertheless they are int cipreted by native constructions are found with the % speakers as being "of no importance" or " Iess relevant" and thus Icss likcly to be presupposed. For a
fuller discussion of this. see Cash Cash and Carnie.
99
A phase-geometric approach to multiple marking systems
( 1 9)
Agentphase
�� Aspllp
Asp"
As .
DP
Theme phase
:::--------
A C c-NE
[ACe]
VP
:.---__ -__ v
This kind of case marking system is a direct consequence of a relativized phasing combined with the language particular featural composition of functional categories. One important thing to note about the tree in ( 1 9) is that although the -ne marked argument is dominated by the vP defining the agent's nuclear scope, due to relativized phasing (and only relativized phasing - normal Chomsky-style phasing does not have this effect) the object can be construed as part of the restrictor of the clause, and thus construed presuppositionally. This is because the LF of the theme phase is constructed entirely independently of the LF of the agent phase. As discussed above in section 1 , this kind of interpretive pattern is extremely difficult to construct if each argument does not have it's own interpretive domain - an advantage presented by the cyclically constructed LFs of Phase Theory. 4. Conclusions I have argued that Phases relativized to argument structure allow an account of the interaction and non-interaction of the interpretation of nominals. This predicts the existence of syntactic marking systems, which indicate the position of nominaIs relative to their interpretation in their own phase. This in tum allows us to take a step towards deriving semantic hierarchy effects the phrase structure hierarchy. References
Abraham, W. ( 1 996). The aspect-case typology correlation: Perfectivity triggering split ergativity. Folia Linguistica 30: 5-34. Aissen, J. (1 999). Markedness and subject choice in optimality theory. Natural Language and Linguistic Theory 1 7: 673-71 1 . Aissen, J . (2003). Differential Object marking: Iconicity vs. Economy. Natural Language and Linguistic Theory 2 1 : 435-483.
1 00
Andrew Carnie
Barss, A. and A. Carnie (2003). Phases and Nominal Interpretation. Ms. Univel A rizona. Binner, M. and K. L. Hale. ( 1 996a). 'The Structural Determination of Ca Agreement'. Linguistic Inquiry 27:1-68. Bittner, M. and K. L. Hale ( l 996b). Ergativity: Towards a Theory of a H ete rog Class. Linguistic Inquiry 27: 53 1-604. Bobaljik, J. (1 993). On Ergativity and Ergative Unergat iv es. MIT Working Papers I Linguistics ] 9:45-88. Bok-Bennema, R. ( 1 99 1 ). Case and Agreement in Inuit. Dordrecht: Foris. Borer, H. 1 994. The p rojection of arguments. University of Massachusetts Occasional Papers 1 7, 1 9-47. Campana, M. (I 992). A Movement Theory of Ergativity. Ph.D. dissertation, McGill University.
on Interpretation . M� University of Arizona. Chomsky, N. (2000). Minimalist Inqu iri es: The framework. In R. Manin, D. Micbaei�_ and J. Uriagereka (eds.). Step by Step: Essays on minimalist syntax in honor of Howard Lasni/c. cambridge, Mass.: MIT press, pp. 89-1 56. Chomsky, N. (1 995). The Minimalist Program For Linguistic Theory. Cambridge M asS,.f Cash Cash, P and A. Carnie (2003). Nez Perce Case: a Note
;:�'
MIT Press.
;r
Chomsky, N. (2001a). Derivation by Phase. in Kenstowicz (ed.). Ken Hale: A Life ;#
Language. Cambridge Mass.: MIT Press Chomsky, N. (200 1b). Beyond Explanatory Adequacy. ms. MIT. DeLancey, S. ( 1 98 1 ). An interpretation of split ergativity and related patterns. Languag. 57:626-657. Diesi ng, M . and E. Jelinek. (1 99S). Distnbuting Arguments. Natural Language Semantics 3: 1 23-1 67 . Di esing, M. ( 1992). Indefinites. Cambridge Mass.: MIT Press. Dixon, R. M. W. (1 972). The Dyirbal Language of North Queensland. Cambridge. Cambridge University Press. Dixon, R. M. w. ( 1 979). Ergativity . Language 55, 59-1 38. Dixon, R. M. W. (1 994). Ergativity. Cambridge: Cambridge University Press. Du Bois, J. W. ( 1 987). The discourse basis ofergativity. Language 63: 805-855. Grimshaw. J. (1 990) Argument Structure. Cambridge Mass.: MIT Press.
Hale, K. L and S. J. Keyser. ( 1 992). Lexical Categories and the projection of argument structure. In L A. Lakarra and J. Ortize de urbina (eds.), Syntactic theory and Basque syntax, 147-1 73 (Supplements of the Anuario del Seminario de Filologia Vasea 28) Hale, K. L and S. J. Keyser. (1 993). On a rgument structure and the lexical expression of syntactic relations. In Kenneth L Hale and Samuel Jay Keyser (cds.) The View from Building 20: Essays in honor of Sylvain Bromberger, Cambridge Mass.: MIT Press. pp. 53-1 09. Hale, K. L. and S. J. Keyser. (2002). Prolegomenon to a Theory 0/ Argument Structure. Cambridge M ass . : MIT Press. Hopper, P. J. and S.A. Thompson. ( 1 980). Transitivity in grammar and discourse. Language 56: 25 1 -299.
I
f
I
Isaak, A. (2000). Split Case Marking and Prominence Relations. Ph.D. Dissertation. - University of Massachusetts, Amherst. Jelinek, E. and A. C arn i e. (2003). Argument Hierarchies and the Mapping Principle. In Andrew Carnie, Heidi Harley and MaryAnn Willie (cds.). Formal Approaches to Function in Grammar. Amsterdam: John Benjamins. pp 265-296. Jelinek, E. ( J 993). Ergative • splits , and argument type. In MIT Working Papers in
Linguistics 1 8: 1 5-42. Johns, A. (1 992). Deriving Ergativity. Linguistic Inquiry 23:57-87
101
A phase-geometric approach to multiple marking systems
Lasnik, H. ( 1 999). Minimalist Analysis. Oxford: Blackwell. Levin. B. ( 1 983). On the Nature of Ergativity. Ph.D. dissertation, MIT. L evi n, 1. and D. Massam. ( 1 995). Surface Ergativity: CasefTheta Relations Re-exam ined . In S. Berman et al. (eds.) Proceedings o.fthe North East Linguistic Society I S: 286-30 1 . Massam. D . (200) . Pseudo Noun Incorporation in Niuean. Natural Language and Linguistic Theory 1 9: 1 53- 1 97. Murasugi, K. ( 1 992). Crossing and nested paths: NP Movement in Ergative and Accusative languages. Ph. D. dissertation, MIT. Percus, O. ( 1 993). Mapping Indefinites. Generals Paper, MIT. Rude. N. ( 1 982). Promotion and the topicality of Nez Perce Objects. In M . Macaulay et al. (eds.), Proceedings o/the Eighth Annual Meeting o/the Berkeley Linguistics Society. pp. 463-483. Rude, N. ( 1 985). Studies in Nez Perce Grammar and Discourse. Ph.D. Dissertation. University of Oregon. Rude, N. ( 1 986a). Topicality, Transitivity and the Direct Object in Nez Pe rce. International Journal ofAmerican Linguistics 52: 1 24- 1 53. Rude, N. ( 1 986b). Discourse-Pragmatic Context for genitive Promotion in Nez Perce. Studies in Language 1 0: 1 09-136. Rude, N. ( 1 988). Ergative. passive and Antipassive in Nez Perce. In M. Shibatani (ed.), Passive and Voice. John Benjamins Publishing Company, Amsterdam. pp. 547-560 Sag, I., T. Wasow and E. Bender (2003). Syntax: A Fonnal lntroduction. Stanford: CSLI. Silverstein, M. ( 1 976). Hierarchy of Features and Ergativity. In R. M . W. Dixon (ed.) Grammatical Categories in A ustralian Languages; Atlantic H ighl ands: Humanities, pp. 1 1 2- 1 7 1 . Reprinted in Pieter Muysken and Henk van Riemsdijk (eds.) Features and Projections. Dordrecht: Foris. pp. 1 63-232. Travis, L. ( 1 99 1 ). Inner Aspect and the Structure of VP. Paper presented at NELS 22 Travis, L. (2000) Event Structure in Syntax. In C. Tenny and J . Pustejovsky (eds.) Events as Grammatical Objects. Stanford: CSLI p ublications, pp 1 45- 1 85. Ura, H . (200 1 ). Case. In M. Baltin and C. Collins (eds.) The Handbook o/Contemporary Syntactic Theory. Ox ford: Blackwell, pp. 334-373. van Hout, A. (2000). Event Semantics in the Lexicon-Syntax Interface: Verb Frame Alternations in Dutch and their Acquisition. In C. Tenny and J. Pustejovsky (cds.) Events as Grammatical Objects. Stanford: CSLl publications. pp 239-282. Willie, M. ( 1 99 1 ). Navajo Pronouns and Obviation. Ph.D. Dissertation, Un ivers ity of Arizona. Woolford, E. ( 1 997). Four-Way Case Systems: Ergative, Nominative, Objec tive and Accusative. Natural Language and Linguistic Theory 1 5 : 1 8 ) -227.
Department of Linguistics Douglass 200E University of Arizona
Tucson AZ 85 72 1
USA
[email protected]
1 02
QR in the Theory of Phases
*
Carlo Cecchetto Universila degli Studi di Milano-Bicocca
In this paper I explain why QR tends to be more local than other types of A-bar movement (i.e. in typica l cases, QR cannot take place out of a
finite clause). 1 assume (and offers evidence for) the Phase Impenetrability Condition and the condition that requires that each step of
(possibly successive cyclic) QR be m oti vated. After showing why QR is local in typical cases, I focus on new evidence, involving a counterpart of ACD in Italian, which indicates that QR does take place long distance. as other types of A-bar movement do. whenever each step is independently motivated. The conclusion is that the locality conditions of QR are not construction specific.
I. Introduction
In this paper, I am concerned with Q(uantifier) R(aising) and its role in a system like the theory of phases (cf. Chomsky 200 1 ) in which the access to the interface(s) is cyclical. I will start from a well known puzzle, namely the observation that, despite the fact that wh movement and QR share some important properties, QR has stricter locality conditions (i.e. in typical cases it does not take place out a fmite clause). I will offer an explanation for why QR is so constrained and more generally for its locality condition. The version of the theory of phases that I will assume is one in which only the access to the phonological component is cyclical. The access to the semantic component, on the other hand, will be taken to be a one step operation that takes place when the derivation is over. In addition to accounting for the locality condition of QR in more common cases, my proposal makes some fine grained predictions that I will systematically test. In particular, I will contrast the length of QR, as measured by using A(ntecedent) C(ontained) D(eletion) as a diagnostic, and the length of QR, as measured by using the availability of inverse scope as a diagnostic. It turns out that QR appears to be longer if the diagnostics based on ACD is adopted. So, surprisingly enough, the length of QR seems to depend on the tools that are used to measure it. I will explain this puzzling situation and I will also discuss an interesting interaction between ACD and inverse scope: ACD makes *
Preliminary versions of this paper have been presented at infonnal seminars at MIT, the University
of Siena, the University of Tlibingen, UCLA, as well as at WCCFL XXII (UCSD), at the XXVIII
Incontro di Grammatica Generativa. (University of Urbino) and at the lAP workshop on EPP and Phases
(MIT). J am grateful to the audience of these talks for their useful comment. Special thanks to
Ivano Caponigro, Gennaro Chierchia. Jon Nissenbaum and Orin Percus for very helpful observations
and criticism. This paper appears as a contribution in the Proceedings ofWCCFL XXII.
MIT Working Papers in Linguistics 49. 103-114 Perspectives o n Phases © 2005 Carlo CecchetlO
QR
in the Theory of Phases
possible the inverse scope reading in some contexts that nonnally do not allow it. 2. Tbe loc ality conditions of QR QR is the name for a phonologically null operation that is responsible for the inverse scope reading in sentences like ( 1 ) . The inverse scope reading is the one which obtains when the QNP every plane gets scope over the QNP a technician, that is, when the superficial scope configuration is inverted. A technician inspected every plane
(1)
A popular thesis about QR is that it is a transfonnation that takes place in the covert component of the syntactic derivation. Since Chomsky ( 1 98 1 ), a basic tenet of the generative framework is that there are no construction specific rules. Adopting this point of view, QR cannot exist as a distinct rule. It must obey the same restrictions that other instances of movement do and its idiosyncratic properties must be explained by the environment in which QR takes place, by its trigger or by some other external factor. In fact, it has been often claimed that QR and wh movement share some important properties and should be analyzed as two instances of the same type of operation. However, there is an outstanding objection against this claim: wh movement and QR do not seem to have the same locality conditions. For example, the fonner can occur across a rmite clause (cf. 2), while the latter cannot (at least in the typical case illustrated by 3, but see below for an important qualification). What did a technician say that I inspected? A technician said that I inspected every plane (out in the V!9 reading)
(2) (3)
If construction-specific rules do not exist and ultimately QR and wh movement are both instances of the same operation Move, why is QR more local than wh
movement? In this paper, I will address this question. 3.
QR in the tbeory of phases
Chomsky
(200 1) assumes that the phonological and the semantic component are
accessed cyclically, in a way which is determined by (4).
(4)
Phase Impenetrability Condition (PIC) The complement of a strong phase a is not accessible to operations at the level of the next highest strong phase �, but only the head and the edge of a are.
I cannot go into the motivation supporting the theory of phases and just focus on the features of the system that will be put in use in my -account of QR. First, let me give an extensional list of strong phases (see Chomsky (200 1 ) and Legate (2003) for motivation and discussion). The only categories that are strong phases
1 04
Carlo Cecchetto
are CPs and transltlve vPs (note that vPs lacking an external role, I un accusative and passive vPs, are not). The edge of a phase referred to in (4) position in the periphery of the phase which is either a specifier or an adjoi: position. The complement of a phase is the complement (in the sense of pht structure theory) of the head of the phase. For example, the complement of phase CP is IP and the complement of the phase vP is VP. PIC states that complement of a strong phase a (say the VP) stops being accessible to synta, operations when the derivation has reached the next highest strong phase (3 ( the CP), because by that time the complement of a has already evacuated syntactic derivation. To illustrate the mechanics of the system, consider a cas. wh movement of the direct object.
(5)
[cp Who did [IP you INFL [vp twho" tyou
V
[vp see twho' ]]]]
PIC states that the complement of the strong phase vP (that is, the VP) is accessible to operations at the level of the nex t highest strong phase. This me that, under PIC, VP is not accessible at the CP level (IP is not a phase, so strong phase which is next highest with respect to vP is CP). If the direct obj ect remains in situ by the time in which the phase vP is completed, it cannot be attracted by the interrogative COMPo The only legitimate derivation, illustrated
in (5), is one in which the wh phrase moves to the edge of the vP phase and is therefore visible from the COMP position, in accordance with PIC. If who did not move to the edge of the vP, it would have evacuated the syntactic derivation by the time the derivation reaches the CP level. PIC forces a very strong form of cyclicity. For example, the wh phrase in (6), in order to move from its base position to its final landing site, must pass through the edge of the embedded vP, the edge of the embedded CP and the edge of the matrix vP (for simplicity, in 6 I have indicated only the traces of the wh phrase).
(6)
[ep Who did INFL [vp t" v
To sum up,
[IP you INFL [vp t"" v [vp [vp saw l' ] ]]]]]]]
think
[cp
till that
[IP John
in order to remain visible in later stages of the derivation, a category
in the complement of a strong phase must move to the edge of the strong phase. Let us now consider covert movement more closely in this system. In the traditional (inverted) Y-Model, covert movement can be easily defmed as the type of movement that occurs after Spell-Out. However. if the access to the phonological component is cyclic, there is no single moment of the derivation in which Spell-Out takes place and i t is not totally clear how the difference between overt and covert movement can be captured. A lthough there are various ways to go, in this paper I will assume Nissenbaum's (200 1 ) version of the theory of phases, in which, unlike in Chomsky's original version, only the phonological component is accessed cyclically. In particular, I will stick to PIC in the version assumed above with the understanding that the complement of a
a is shipped to the phonological component when the next highest strong phase (3 is reached. However, I take the access to the semantic component to be a one step operation that takes place at the very end of the derivation. In this modified version of the phase system, QR is movement of a category that has already been spelled out. Like overt movement, QR obeys PIC (this
strong phase
105
QR in the Theory of Phases
assumption is motivated by the usual considerations: if every occurrence of movement is just one instance of the same operation Move, QR has to obey the same restrictions that overt movement does). Summarizing, the access to the semantic component is not done cyclically. Covert movement is movement of spelled out categories and obeys PIC. I 4. The length of QR measured by the inverse scope diagnostic There are two main diagnostics to measure the length of QR. The first one is the availability of inverse scope. In this paragraph I focus on this. Sentences like (3) above suggest that QR does not take place out of a finite clause. Let us now observe the extraction pattern of QR out of non finite clauses. As discussed in the literature, infmitival complements come in two varieties. Some of them allow long QR of the material inside the infinitival clause. This happens in (7): (7)
Someone wants to dance with every woman (the V;9 reading is OK)
Other infinitival complements are less permissive: (8)
Someone persuaded John to attend every class (out in the V;9 reading)
In order to account for this pattern, Hornstein ( 1 995) proposes that long QR is possible only in restructuring contexts. In (7), but not in (8), restructuring would occur. While this analysis is appealing, it suffers from an obvious drawback. English does not have an independent reliable diagnostic for restructuring. So, if restructuring contexts are defined as contexts that allow long distance inverse scope, the discussion is circular. Italian is a better choice to test Hornstein's proposal because it does have an independent diagnostic for restructuring, namely clitic climbing (cf. Rizzi 1 982). In fact, whenever an Italian construction allows clitic climbing, it also allows a QNP in the infinitival complement to take scope over the matrix subject. One indisputable example of restructuring verb is volere (tlto want"): (9)
Almeno uno studente vorrebbe licenziare ogni professore At least one student would like to fIre every professore (the V;9 reading is OK)
Object control structures never allow restructuring. In object control structures, the long distance inverse scope reading is never possible:
1 An alternative approach to covert movement is the so-called phonological theory of covert movement (cf. Brody 1 995. Fox and Nissenbaum 1 999, and references cited therein). Although, in this paper
I cannot show this. I believe that my analysis
of the locality condition of QR could be
refonnulated in the phonological theory of covert movement. as long as the latter is coupled with
PIC.
1 06
Carlo Cecchetto
( 1 0)
Almeno un poliziotto gli ha ordinato di sorvegliare ogni uscita ordered to control every exit
At least a policeman to him (the '1/:1 reading is out}
These examples are fully representative. Although I cannot report the entire pattern, the evidence based on inverse scope indicates that long QR of the material inside an infmitival clause is possible only in restructuring contexts ill Italian. 5.
The length of QR measured by the ACD diagnostic
Let us now srudy the locality conditions of QR by using the availability of the matrix reading in ACD structures as a diagnostic. The matrix reading is the one that requires that the elided verb phrase be the matrix one. For example, a sentence like ( 1 1 ) is ambiguous.
(11)
John wants to visit every city Bill does
( 1 1) can either mean "John wants to visit every city Bill visits" (this is the embedded reading) or "John wants to visit every city Bill wants to visit" (this is
the matrix reading). The terminology reflects the fact that, in order for the matrix reading to obtain, every city Bill does must QR to a position higher than the matrix VP to fix the infmite regress problem that ACD triggers. On the other hand, for the embedded reading to obtain, QR to the embedded VP suffices. Hornstein ( 1 995) argues that the matrix reading is possible only in restructuring contexts. However, Kennedy (1 997) challenges this by pointing out examples like ( 1 2) in which the matrix reading is possible although expect, arguably, is not a restructuring verb:
( 1 2)
If you expect to read the same articles Albert does �Epeet te reae, then you should get started now
Furthennore, Wilder ( 1 997) discusses examples seems to require QR even out of a fmite clause: ( 1 3)
in which the matrix reading
John said that you were on every committee that Bill did (say that you were on)
Let us look at ACD in Italian, in which restructuring can be more reliably detected. In the literature, there is no systematic discussion of ACD in Italian and for a good reason. VP (or vP) ellipsis is virtually absent in this language, so the configuration that triggers ACD in English cannot be exactly reproduced. However, a very similar configuration can be created by using ex amples that, instead of containing a gap for the elided VP, contain the pro-predicative c1itic
10, which is a pro-form that stands for the missing VP. The role of the clitie 10 in a simple sentence is illustrated in ( 1 4) . In ( 1 4) the inflected fonn offore (lito do") plays the role of the auxiliary do in English. Instead of a null category for the VP, 10 surfaces. 1 07
QR in the Theory of Phases
( 1 4)
Maria ha rimproverato Gianni e anche Piero 10 ha fatto Maria has reproached Gianni and also Piero it has done 'Maria reproached Gianni and Piero did, too' 10 [vp reproach Gianni] =
Now consider (1 5). The clitic 10 is again a pro-fonn for the missing VP. If ogni ragazzo che mi aveva chieslO di[arlo moves out of this VP (for example adjoins to IP, as indicated in ( 1 5'» , at LF the VP [inlerrogato t) replaces the clitic and the intended interpretation is derived. On the other hand, if ogn i ragazzo che m i a veva chieSIO di [arlo did not move, the usual infinite regress problem would arise. So, ( 1 5) seems to be a genuine counterpart of ACD, even in the absence of VP ellipsis in Italian2: ( 1 5) ( 1 5')
Ho [vp interrogato ogni ragazzo che mi aveva chiesto di farlo ] 1 have examined every boy that had asked me to do it [IP [ NP ogni ragazzo che mi aveva chiesto di farlo] [IP pro ho interrogato t]]] 10 [vp examine t]
[vp
=
Let us check the distribution of the matrix reading in Italian in non restructuring contexts. One example is ( 1 6). ( 1 6), in which I have altered the sequence of tenses in a way that makes the embedded reading inconsistent, is grammatical with the matrix reading. ( 1 6)
Ammettero di aver licenziato (senza giusta causa) ogni lavoratore che mi costringera a farlo I will admit to have fired (with no good reason) every employee who will force me to do that
2 The category whose interpretation potentially triggers the infinite regress problem is a gap in English but is a pronoun in Italian. Since a pronoun can easily refer to a non linguistic antecedent, I present a piece of evidence that shows that 10 in sentences like (1 S) actuaUy refers to the VP out of which QR must apply. Hornstein (1 995) has observed that the elided VP can be contained in the subject of a ECM structure but not in the subject of a finite clause. (i) John believes every suspect Bill
does to be guilty
(ii) •John believes that every suspect Bm docs is guilty
If the Italian cases we are considering are the genuine counterpart of ACD cases, we expect a similar contraSt to arise and this is what happens. (iii) is acceptable but (iv) is not (in the intended interpretation). (iii) Ho osservato anentamente ogni ragazzo che mi avcva chjesto di farlo danzare i l tango I have observed
carefully every boy who had asked mc to do .it dancing the tango
10 :::= [vp carefu lly observe t dancing the tango]
(iv) Ho osservato anentamente che ogni ragazzo che mi avcva chicsto di farlo danzava il tango I have
observed carefully that every boy who had asked me to do it was dancing the tango
*10 :::= [vp carefully observe that t was dancing the tango]
Whatever explanation is chosen for this panern, the parallelism betwcen Italian and Eng l ish suggests in both languages.
that same underlying configuration holds
1 08
Carlo Ce cch etto 10 = [vp adm it to *10 = [vp fired t]
have fired t]
So� sentences like ( 1 6) prove that the matrix reading is available in principle with non restructuring verbs. Not surprisingly, it is possible to build similar examples that require QR out of a fmite clause:
( 1 7)
(Davanti al giudice) Ammettero che volevo licenziare (senza giusta causa) quei lavoratori che mi costringeranno a farlo (In court) I will admit that I wanted to fire (with no good reason) those employees who will force me to do that 10 [vp admit that I wanted to fire t] =
Summarizing, the distribution of the matrix reading in I talian indicates that QR can take place out of restructuring contexts, out of non restructuring contexts and out of finite clauses as well.
6. Explaining the locality conditions of QR
6.1 Where we are
Let us summarize where we are. If we measure QR by using the availability of the inverse scope reading, QR seems not to be able to apply out of either a fmite clause or a non restructuring context. If we measure QR by using the availability of the matrix reading in ACD configuration, on the contrary, it is possible to show that QR applies out of finite clauses and non restructuring contexts. We are facing a near contradiction here. The length of QR seems to depend on the diagnostics used to measure it. I will now explain this puzzling pattern. . 6.2 Background assumptions
Remember that I have been assuming that QR is constrained by PIC and takes place only when the entire syntactic derivation is over, as discussed in paragraph 3 . In particular, QR applies to spelled out categories. Before presenting my analysis I have to introduce two other assumptions. The fIrst one, which plays a crucial role in my explanation, is a version of Fox's (1999) Scope Economy. In particular, I will assume that each step of successive cyclic QR must have an independent motivation. For example, consider ( 1 8), in which linear precedence indicates c-command, and suppose that the QNP must QR from a position inside y to a position inside a. but cannot do that in a single step because this long movement is blocked by PIC. Also suppose that the QNP is allowed by PIC to arrive to its final landing site if it uses the edge of � as an intennediate landing site, as indicated in ( 1 8): ( 1 8)
[a ... ..QNP
. ......[jl tQNP ......
[y
. .....
1QNP . . . . . I OQ
QR in the Theory of Phases
I will assume that the successive cyclic derivation in ( 1 8) is acceptable only if the movement from the base position to the edge of 13 has a motivation other than simply allowing the further movement of the QNP. The second assumption is about the specifics of the landing site of QR and I make it mostly for concreteness. Take a sentence like a technician inspected every plane under the inverse scope reading . Clearly, after QR the QNP every plane must c-command the QNP a technician. However, this leaves a certain space of liberty, depending on general hypotheses about clausal structure and reconstruction that one adopts. For example, in May's ( 1 985) theory, the landing site of QR is the adjunction position to IP. In the much more sophisticated system developed by Beghelli and Stowell ( 1 997), QNPs are divided in different types and each type is associated to a specific landing site in the IP field. In other proposals every plane would be taken to QR to the periphery of the vP and inverse scope would b e obtained through reconstruction of the subject a technician (cf. Johnson and Tomioka 1 997). Luckily, we do not need to take a precise stand on this debate. In order for my analysis to work, I just need to assume that the landing site of QR is located somewhere lower than the CP. In the representations that I will give, the landing site of QR is, as in May's (1 985) system, a position adjoined to IP (but remember that this is not essential for my argument). Having spelled my assumptions, I start explaining the pattern we have observed. 6.3 Why lo ng QR is not possible if measured by using the inverse scope diagnostics
Let us take ( 1 0) as a representative. The base configuration is indicated in (l Oa). I assume that the short occurrence of QR in ( l Ob) is legitimate, since it is short enough to obey PIC and is semantically motivated. This is so because a QNP in the object position of a transitive verb can undergo QR to allow its trace, which is of type e, to combine with the transitive verb, which is of type < e, < e.t» . However, after the short QR of ogni uscita in ( l Ob), which is licit, a problem arises. Two possible ways to derive the inverse scope reading are conceivable but are both blocked. The first conceivable derivation is in ( I Oc), in which ogni uscita moves directly from the embedded IP to the matrix IP area. PIC does not allow this long movement, though. The next highest strong phase with respect to the embedded IP is the matrix vP, so ogni uscita cannot move in a single step further than the edge of the matrix vP. The second conceivable derivation is in (l Od). In this derivation PIC is obeyed since ogni uscita uses the matrix vP as an intermediate step. However, Scope Economy is violated, because the movement from the embedded IP to the edge of the matrix vP is not independently motivated (it only occurs to allow the further movement of ogni uscita to the matrix IP). ( l Oa) [.P Almeno un poIiziotto gli ha [vp v [ vp ordinato [ep di rtp PRO INFL [vp v [vp sorvegliare ogni uscita]]]]]]] ( l Ob) ltp Almeno un poliziotto gli ha [vp v [vp ordinato [ cp di [ IP [ogni uscita]i [IP PRO INFL [vp v [vp sorvegliare til ]]]]]]]
1 10
Carlo Cecchetto
( l Oc) *llP [ogni uscita]i [IP Almeno un poliziotto gli ha [liP v [vp ordinato [cp cii lIP ti [IP PRO INFL [ P v [vp sorvegliare til ]]]]]]]] li
out because PIC is violated
( I Od) * [IP [ogni usc ita] i [ IP Almeno un poliziotto gli ha [vp tj [vp v [vP ordinato [cp di [IP tj [IP PRO INFL [vp v [vp sorvegliare lj]]]]]]]]] out because Scope Economy is violated
As the reader can easily verify, the same reasoning excludes the inverse scop� reading in the cases like (3), which are standardly used to argue that QR cannot take place out a finite clause. Also the fact that long QR is possible in restructuring contexts like (9) is expected adopting PIC. The characteristic property of restructuring contexts is that the embedded subject position is suppressed and the complement of th� embedded verb becomes the complement of the restructuring one. There are different implementations of this idea and of all of them would get the desired result in our case. Mostly for concreteness, I will assume that the restructurin� and the embedded verb sit in two distinct verb phrases but the embedded verb , phrase is a VP which is not embedded in a vP layer (this expresses the idea that the embedded verb does not take an external argument). So, I propose that (9) should receive the analysis in (9'):
(9')
[IP Almeno uno studente [vp v [vp vorrebbe [vp licenziare ogm professore]]]]
Remember that VPs are not phases (only vPs are). This means that ogm can QR in a single step up to the matrix IP (the next highest stron1s phase with respect to the one in which ogni pro/essore overtly sits is the matrix CP). So, it can take scope over almeno uno studente without violating PIC.
professore
6.4 Why long QR is possible if measured by using the ACD diagnostics
So far, so good. PIC, when combined with Scope Economy, naturally explains why QR out of fmite clauses and complements of non restructuring verbs is illicit. However, we have not yet dealt with the puzzling fact that QR out of finite clauses and complements of non restructuring verbs becomes possible whenever the length of QR is measured by using A CD as a diagnostics. To explain this, take the matrix reading in ( 1 6) as a representative (cases of ACD motivated QR out of finite clauses are structurally similar and admit a similar account). The legitimate derivation is in ( l 6a)-( 1 6c). ( 1 6a) is the base configuration. In ( 1 6b) ogni ragazzo che mi aveva chiesto di farlo undergoes short QR, which is motivated by the need to repair a type mismatch with the transitive verb. Under PIC, ogni ragazzo che mi aveva chiesto di farlo can further move as far as it does not escape the matrix vP. In ( 1 6c), the QNP adjoins to the matrix VP and this suffices to create the configuration that rules the matrix reading in.
111
QR in the Theory of Phases
( 1 6)a. [IP pro INFL [vp v [vp ammettero [cp di [IP PRO aver [vp v [vp licenziato ogni lavoratore che mi costringenl a farlo ]]]]]]] b. [IP pro INFL [yP v [vp ammettero [cp di bp [ogni lavoratore che mi costringera a far/o] j [IP PRO aver [yP v [vp licenziato tj ] ]]]]]]] c. [IP pro INFL [vp v [vp [ogni lavoratore che mi costringera a far/o ] j [vp ammettero [cp di [IP t"i [IP PRO aver [liP v [vp licenziato tli ]]]]]]]]] All in all, the matrix scope reading in ACD contexts is possible in non restructuring contexts because the required QR is slightly shorter than the QR required by the inverse scope reading in the same contexts. The longer QR necessary to establish the long distance inverse scope reading requires the occurrence of a non independently motivated instance of movement.
7.
A fine grained prediction
The long distance inverse scope reading is out in ( 1 0) because the occurrence of QR from the embedded IP to the matrix vP is not independently motivated (cf. I Od). This analysis makes a fine grained prediction. If QR from the embedded IP to the matrix vP area were motivated in a certain syntactic context, the long distance inverse scope reading should be available in that context. As it happens, this prediction can be tested. We know that the matrix reading in ACD requires adjunction to the matrix VP. So, we predict that a QNP in the complement clause of a non restructuring verb should take scope over the matrix subject whenever it QRs to resolve the infinite regress problem in ACD. Stated differently, the prediction is that the long distance inverse scope reading should be contingent on the presence of the matrix reading. I now discuss one minimal pair that confirms this prediction. In order to facilitate the judgments, I asked the informants to evaluate the sentences in the scenario that I report. ( 1 9)
(20)
Uno di noi ammettera di aver picchiato ogni manifestante One of us will admit to have beaten each demonstrator (the VJ9 reading is out) Uno di noi ammettera di aver picchiato ogni manifestante che ci obblighera a far/o One of us will admit to have beaten each demonstrator who will force us to do that (the 'v'/3 reading is OK, that is (20) can mean: "For each demonstrator, if that demonstrator forces us to admit to having beaten him, then one of us will admit to having beaten him '1 SCENARIO: a meeting of policemen after the 200 1 Genoa 0-8 summit, during which policemen committed violent actions against demonstrators. It is decided to acknowledge that at least some of these violent actions took place, since it is impossible to deny the evidence across the board.
1 12
Carlo Cecchetto
The long QR that has as an output the wide scope reading of ogni manifestan che ci obbligheriz a larlo over the matrix subject uno di noi is a successi' cyclic transformation in which each step is independently motivated. Th
derivation is illustrated in (20'). (20')
[IP [ogni manifestante che ci obblighera a farlo l j [ IP uno di n o i INFL [vp [vp tllli [vp ammettera £Cp di riP t"i [IP PRO aver [vp v [vp picchiato
]]]]]] ]]]]
The movement from the base position to the embedded IP is motivated by tl need to resolve the type mismatch, as in the previous examples. The moveme from the embedded IP to the matrix VP is motivated by the need to resolve tl infmite regress problem triggered by ACD. Finally, the movement from t1 matrix VP to the matrix IP is motivated by the need to establish inverse scop So, each link of this chain is motivated and is short enough to obey PIC. Inver. scope obtains,
as my approach p redic ts .
8. Back to Wh-movement
' One surprising result of this paper is that, if I am right, QR obeys stricter conditions than (overt) wh movement . Assuming the theory based on PIC, a wh phrase that overtly moves can use the edges of vP and CP to escape from a category that is being spelled out, even if these intermediate steps do not have an independent motivation (cf. the intermediate traces in (5) and (6» . One can say that these intermediate steps are motivated because v and C have some uninterpretable feature F that must be deleted by the wh phrase. However, saying this would be just a way to restate the facts and would not be very illuminating. It is more tempting to assume that the difference between QR and wh movement has to do with the overt/covert character of the movement operat ion. Suppose that we have a language that requires that wh features be checked before wh phrases are spelled out, otherwise the derivation crashes.
Also suppose that a certain wh phrase A is in the domain of a ph ase a (say, the VP) and needs to move to a certain position outside a (say, the CP) to check its wh feature. According to the derivations in (5) and (6), A can use the edge of a (say, the outer Spec of vP) to avoid being spelled out. Note that this does not necessarily imply any kind of "look ahead" property: immediate Spell Out of A would lead the derivation to crash, no matter what will bappen in the further steps of the derivation. So, movement of A to the outer Spec of liP is motivated ( without it, the derivation immedia tely crashes) . Remember that we have been assuming that QR is a movement of a category that has already been spelled out. So, clearly QR can never be motivated by the need to postpone Spell Out. Each step of its successive cyclic occurrence must have a different motivation. We have seen three of them: the need to create the inverse scope reading, the need to resolve the i nfin ite regress problem that arises with ACD and the need to allow the combination between the trace of a QNP and a transitive verb. Other motivations for QR are conceivable but without them Scope Economy is violated.
1 13
QR in the Theory of Phases
Note that this analysis of the source of the difference between QR and wh movement makes a prediction. If wh movement occurs covertly, it should be
like QR, namely each step of successive cyclic covert wh movement ought to
have an independent motivation. I hope to test this prediction in future research.
9. Conclusions
Given current assumptions about syntactic theory, the fact that QR has locality conditions different from those of other types of A-bar movement comes
as
surprise. I hav e looked at these locality conditions in English and Italian and
a
I
have identified a pattern that, despite some prima facie anomalies, is consistent and admits a principled explanation. A central aspect of my account, together with a principle of economy, is PIC, the principle that oversees a derivation in
which the interfaces are accessed cyclically.
References Beghelli, Filippo and Tim Stowell. (1997). Distributivity and Negation: The Syntax of Each and Every. In A. Szabolcsi (ed.), Ways o/Scope Taking, Dordrecht: Kluwer. Brody, Michael (1 995). Lexico-Logical Form, Cambridge, Mass.: MIT Press. Chomsky, Noam ( 1 9 8 1 ). Lectures on Government and Binding, Dordrecht: Foris. Chomsky, Noam (200 1 ). Derivation by Phase. In Michael Kenstowicz (ed.), Ken Hale: A Life in Language, Cambridge, Mass.: MIT Press, pp. 1 -52. Fox, Danny (1 999). Economy and Semantic Interpretation, Cambridge, Mass.: MIT Press. Fox, Danny and Jonathan Nissenbaum (1999). Extraposition and scope: a case for overt QR. In Bird S. et al. (eds.) Proceedings 0/ weeFL /8. Somervil le: Cascadilla Press. Hornstein, Norbert (1 995) Logical Form, Oxford: Basil Blackwell. Johnson, Kyle and Satoshi Tomioka (1 997). Lowering and mid-size clauses. In Graham - Katz, Shin Sook-Kim and Winhart Haike (eds.) Proceedings of the 1 997 Tiibingen workshop on reconstruction, pp. 1 85-206. Kennedy, Christopher (l 997). Antecedent Contained Deletion and the Syntax of Quantification. Linguistic Inquiry 28.4. Legate, Julie Anne (2003). Some Interface Properties of the Phase. Linguistic Inquiry 34(3); 506-5 16. May, Robert ( 1985). Logical Form, Cambridge, Mass.: MIT Press. Nissenbaum. J. (2001) Investigation ofCovert Phrase Movement, MIT Dissertation. Rizzi, Luigi (1982). Issues in Italian Syntax. Dordrecht: Foris. Wilder, Chris ( 1997). Phrasal Movement in LF: de re readings, VP-ellipsis and binding. In K. Kusumoto (ed.) Proceedings a/NELS 27. Facolta di Ps icologia Universita degli Studi di Milano-Bicocca Piazza dell' Ateneo Nuovo,
[email protected]
1 14
1 20 126 Milano (Italy)
Phases and autonomous features: a case of mixed agreement in European Portuguese JOQO Costa & Sandra Pereira Universidade Nova de Lisboa & Centro de Linguistica da Universidade de Lisboa This paper studies a case of mixed agreement
in European Portuguese, triggered by referential and
the pronoun a genIe. This pronoun has active non-agreeing
grammatical features that emerge in different local domains. In prcdicative c�ntexts, the two sets of features become visible. We analyze this fact. by proposing that phases arc the relevant domains for agreement to operate and that the two types of features arc separately represented in the featural hierarchy of pronouns.
1 . Introduction
and goals
The pronominal system of European Portuguese includes the expression a ge This expression, which literally means ..the people", is used as a first person pi' pronoun, generally occurring in free variation with the fIrst person plural pronoun J Menuzzi (2000) shows that this expression truly behaves like a pronoun, since it 01:: conditions B and C of Binding Theory like the other pronouns of the pronom paradigm, as illustrated in the examples below. In ( I ), it is shown that a genie behaves like a pronoun for the sake of principl effects. since it can only be bound non-locally:
I i
Condition-B: a.
(I)
!
b.
*[Eu e o Pedro]i vimos a gent�. I and Pedro saw A GENTE [Eu e 0 Pedro]i dissemos que a gentei adorou a festa. I and Pedro said that A GENrE loved the party
In (2) and (3). it is shown that a gente behaves like the first person plural pronoun nos the sake of compliance with principle B and C: (2)
a. b.
[Eu e 0 Pedroli dissemos que a gentei adorou a festa. I and Pedro said that A GENTE loved the party [Eu e 0 Pedro]i dissemos que n6si adoramos a festa. I and Pedro said that we loved the party
MIT Working Papers in Linguistics 49. / /5-/24 Perspectives on Phases © 2005 Joiio Costa and Sandra Pereira
Phases and autonomous fearures
( 3)
a.
h.
"'A gentei viu [a mim e ao Pedro]i. A GENTE saw me and Pedro *Nasi vimos [a mim e ao Pedro]j. We saw me and Pedro
Further evidence for the starus of this expression as a pronoun comes from the fact that it behaves like other pronouns as far as the avoid pronoun principle is concerned:
(4)
a.
b.
Nos achamos que somos felizes. We think that are happy A gente acha que e feliz. A GENTE thinks that is happy
Also, like other pronouns, a genre resists adjectival modification, as illustrated in
(5)
a.
h.
il
1
c.
*Nos felizes romos a praia. We happy went to the heach '" Eu feliz fui a praia. I happy went to the beach '" A gente feliz foi Ii praia. A GENTE happy went to the beach
Finally, apart from its referential reading. like other pronouns, a readings, as illustrated in (6); (6)
A gente tenta sempre fugir ao tisco. A GENTE tries always to run from taxes
genIe
(5):
allows arbitrary
These data, discussed in Menuzzi (2000), provide compelling evidence that a gente behaves like other pronouns of the pronominal paradigm of Porruguese. An aspect in which this fonn differs from other pronouns, making it particularly interesting, is that its referential and grammatical features do not coincide. In fact, the pronoun a genie refers to first person plural, alternating with the pronoun nos (we), but it is grammatically specified as third person singular. The grammatical specification becomes obvious in the fact that this pronoun triggers third person singular agreement on the verb. In this respect, as mentioned by Menuzzi (2000), the pronoun a gente behaves like the French pronoun on. Menuzzi (2000) shows that both the semantic and the grammatical features are active and relevant for binding purposes. This author shows that the grammatical features emerge in the establishment of local binding relations, as illustrated in (7). (7a) shows that 3w person singular agreement is triggered on the verb; in (7b), it is shown that in local binding relations, the anaphoric form selected is of the SE-type, which is specified with 3rd person singular fearures as well:
1 16
Jolio Costa and Sandra Pereira
(7)
a. b.
From
A gente (vail*vamos) it praia.
The people goesl* golst plur.l1 to the beach. A gente ve-(sel*nos) no espelho. The people sees herself/*ourselves in the mirror
(7), one might suppose
that the referential features of a genie are just a matter of
interpretation without reflexes on the syntax. However, Menuzzi (2000) shows that non locally, a genIe selects a pronominal form specified for the referential features. As shown 1 st in (8), the pronominal fonn selected in non-local environments must be specified with person plural features.
(8)
A gente disse que a Maria (nos/*a) viu. The people said that Maria us/*her saw
These data provide the necessary evidence to posit that the two sets of features play a role in understanding the distribution and behavior of this pronominal form. According to Menuzzi (2000), there is a correlation between the locality of the binding relation established with a genIe and the set of features that becomes salient. An interesting puzzle arises when predicative contexts are considered. Since adjectival and participial predicates bear agreement morphology, it is interesting to know whether these forms will agree with the grammatical features of the pronoun or with its referential features. Although, at fIrst sight, there seems to be a great deal of variation in ' the data, a quantitative study carried out in Pereira (2000, 2003) reveals that the most predominant pattern is the one illustrated in {9}:
(9)
a. b. c.
I
'"
!
d.
A gente esta cansados. A GENTE is tired-rnase-pi A gente esta cansadas. A GENTE is tired-fern-pi Minha mae que Deus haja, ia para a casa de uma tia my mother who god has, went to the house of an aunt minha, defronte ao hotel Porto Santo, e mine, in front of hotel Porto Santo, and a gente ia, pequeninas, a gente ia. A GENTE went-3 sg little-fern-pi Se ele por mais, arrebenta com a gente todos, maritimos if he puts more, burst with A GENTE all-masc-pl sailors-masc.pl
I The methodology used in Pereira (2000, 2003) for analyzing these data was as follows: at a IlI'St stage, a collection of occurrences of the relevant type of construction in different types of corpora was carried out (CORDIAL-Sin and CRPC); at a second stage. the data obtained from the corpora were completed so as to include all types of relevant syntactic environments. After the data was compiled, a grammaticality judgement task was conducted, in which informants were asked to complete participial and adjectival forms by adding the relevant agreement markers.
1 17
Phases and autonomous features
The data obtained in Pereira (2000; 2003) reveal that a geme triggers 3rd person singular agreement on the verb, but l Sl person plural agreement on the predicate . The variation in gender on the predicate depen ds on the referent for the pronoun (masculine or feminine) . The conclusion we can draw from these results is that there is a case of mixed agreement within one single IP. This is obviously problematic for Menuzzi's conclusion that the activation of grammatical or referential features was dependent on the locality of the relation between a gente and the agreeing fonn, since, in the cases illustrated in (9), the relation between the pred icates and a genIe is local. Given the facts just described, the goal of this paper is to answer the fo llow ing questions: How can the mixed agreement in predicative sentences be accounted for? a) How can the grammatica l and referential features be set apart? b) The paper is organized as follows: in section 2, we spell out some assumptions regarding the hierarchical organization of the feature specification of pronouns, and concerning the locality of agreement relations . In section 3, we present the analysis based on the idea that words are inserted at Morphological Structure (Halle and Marantz 1 993). Finally, in section 4, some predictions made by this hypothesis are explored . 2. Assumptions
The analysis we are about to propose for the pattern of mixed agreement described above is based on the assumption that the features of the pronominal fOnDS do not work as a bundle, and that the Phase is a relevant domain for syntactic agreement to be established. Let us review each of these assumptions individually. a)
The feature sp ecification of pronouns is hierarchically organized. Different sets of features group together independently of other features.
This assumption is independently motivated in work by Harley and Ritter ( 1 999), Duarte et a!. (2002), among others. According to these authors, the features of a pronoun are subject to an internal hierarchy. It is not particularly crucial for our goals what the exact internal hierarchy of the pronoun a gente is as long as the grammatical features and the referential features are set apart. This is achieved in the follOWing feature specification for the pronominal a genie, adapted from Duarte et a1. (2002):
1 18
Joao Costa and Sandra Pereira
( 1 0)
Feature specification of a gente: Root
� Individuation
Referential
---Group
Participant
� Class
/'-.... [speaker] [proximity] As can be observed in the infonnation given in (t o), the grammatical (Individuation) and referential features (Referential) are separate. We will further assume that each of the nodes in the hierarchical representation can be activated in different domains. For instances, assuming this type of hierarchy, and coming back to Menuzzi's observations regarding the binding facts, one must assume that only the referential node is active in non-local domains. Note that we are aware of the fact that, by itself, this is not an explanatory device. For the purposes of this paper, what is relevant for us is that the two sets of features are isolated, and that one does not consider that the features of pronominal fonn work as a non-structured bundle. Phases are the domains for agreement operations to take place (Chomsky 1998, 1999).
b)
The second assumption we make is inspired on ChomskY's ( 1998, 1999) proposal that Agree operates phase-internally. This assumption is independently motivated by the observation that.agreement, when it occurs, is the morphological reflex of a local relationship between two agreeing elements? c)
The small clause is a Phase.
Related to the previous point, we propose that the Small Clause is a phase-boundary, as independently proposed in Svenonius (2003). Although there is little consensus regarding the exact d�finition of a Phase, three aspects motivate this assumption:
2 It is important to note that we are not assuming that lack of morphological agreement signals the lack of establishment of an Agree relation between two elements. The assumption just states that whenever morphological agreement is visible, it reflects the existence of a local Agree relation.
1 19
Phases and
autonomous features
First. Chomsky ( 1 998, 1 999) proposes that a phase is a coherent unit in terms of co mplete domain of predication , in which theta-roles are fully saturated.
reference. As is well known, Small Clauses constitute a
Second, small cl auses
are prosodically coherent, which is another characteristic of ( 1 998, 1 999). The example in ( 1 1 ) i llu stra tes this of the coordinated small clauses may be argued to constitute an
phases, according to Chomsky property, since each independent prosodic unit:
l :
(1 1 )
Eu cons idero [a Maria bonita] , [a Olga gorda] e [0 Pedro fe io] . I co nsid er Maria pretty, Olga fat and Pedro u gly
Finally, it b etween
attests: ( 1 2)
a. b.
is well-known that small clause predicates do not scramble, as the contrast DP scramb ling and small-clause predicate scrambling illustrated in ( 1 2) -
o Pedro fala frances sempre. Pedro speaks French always. *0 Pedro esm cansado sempre. Pedro is tired always
If short-distance-scrambling must be internal to a Strong Phase, the difference between nominal obje cts and small clause predi cates follo ws . The predicate of the small clause
cannot be scrambled, because it would have to cross the strong phase boundary, not b eing
able to use the edge of the d)
phase as
an
escape hatch.
Words are inserted in the morphological component. Syntax only handles features (Halle and Marantz
1 993).
A fina l assumption we need to make is that, following H all e and Marantz ( 1 993) lexical insertion is post syntacti c This assu mption ensures that the selection of the pronominal fonn agreeing with a genie in the case of binding, or the selection of the agreement marker in the predicative contexts w ill be sensitive to the features activated in each domain, and not to the whole set of features present in the feature specificati on of a gente given in ( 1 0). Without this assumption, it becomes difficult to explain why only some features are activated in certain domains, and why not all features are checked in all steps o f a derivation, as argued in Costa, Moura and Pereira (200 1 ). -
.
3. Analysis
Given the assumptions sp elled out in the previous section, we have the ingred ients for deriving the pattern of mixed agreement observed The subject of the sentence is generated as the subj ect of the small claus e predicates Within the small clause domain, the semantic features local ly probe the predic ate for agreement, spec ifying its features for 1st person plural. It is important to .
.
1 20
loao Costa and Sandra Pereira
I ,
clarify why the referential features are the first features to be
activated. Following and Figueiredo S i lv a (2002), we assume that referential numb er agreeme n1 dissoc iated morpheme in the sense of Embick and Noyer (2001 ) and adjoins t( potential root. Costa and F igueiredo Silva (2002) present evidence for this c lai m ba a comparison of the agreement patterns in three registers of Portuguese. It is argl these authors that referential number agreement is a dissoci ated morpheme in Eu) Portuguese, and a sin gleton in Brazili an Portuguese. This proposal is based ( following facts: i) ii)
iii) iv)
v) vi) vii) viii) ix)
In European Portuguese (EP), there is full DP-intemal agreement; In the two registers of Brazilian Portuguese (BPI and BP2) studied, number wit DP is marked either on the D head or in all prenominal elements; Prenominal adjectives and possessives differ in BP, in the sense that only the 1au carry number marking in the absence of plurality on the definite article; In EP, there is full subject-verb agreement, independently of the position of the s ex.cept in the case of unaccusative verbs in which agreement is op tional with it subjects in colloquial speech; In EP, there is full number agreement with passive participles and SC predicates; In B P I , there is full subject-verb agreement, ex.cept for the case of inverted subject� In BPI , there is no number agreement with passive participles and SC predicates; In BP2, there is no number subject-verb agreement; In BP2, there is no number agreement with passive participles and SC predicates.
According to this analysis, the first step of the derivation pro ceeds as in ( 1 3): the fe of a gente are merged at the edge of the small clause, and due to the d issoc iated nat number in Portuguese. the number features of the predicate become specified and V�
(1 3 )
a. b.
rd [sc OP [[3 sg] [pi]] AP [p i] ] 3rd sg] [pi] ] AP [pi]] OP [[ [sc
Assuming that the number marking on the small clause predicate partially follows the dissociated nature of the morpheme is not crucial for the analysis, but ma: possible to d ispense with a theoretical apparatus in which it would be necessary to independ ent processes for establishing agreement relations within the same language Since the small clause is a str�ng phaSe, only its edge is access ible for fi operations. This enables movement of the subject to Spec,IP in order to check features, as illustrated in ( 1 4):
( 14 )
rd d [IP O P[[3r sg] [pl11 I [sc DP [[3 sgJ [pi]] AP [pi]]
Due to the Phase Impenetrability Condition, the only features that c� accessed within the IP domain are the unvalued ones: the grammatical features. explains why 3rd person singular verbal agreement is found on the verb. In other '" the claim is that checked features are not available for further agreement operatic other domains. Since the referential number features were valued within the small cI they will no longer be valued in the Inft domain.
Phases and autonomous features
This analysis has 'the' advantage of bringing the case of mixed agreement close to ' the generalization obtained by Menuzzi (2000) for the binding facts. For both sets o f facts, the explanation lies on the observation that there are two different domains deriving the activations of different parts of the feature specification of the pronoun a gente. The case of mixed agreement discussed in this pap er reproduces the binding p attern in th e sense that in both cases, the grammatical features emerge locally, while the referential features emerge non-locally. The assumption that the small clause is a strong phase derives the non-local facet of agreement in predic ative contexts. Agreement in this context is similar to the selection of 1 SI person plural pronouns in non-local domains, because the pronoun a genIe and the agreeing predicate surface in a non-local relationship. It is important to emphasize that without late lexical insertion, there would be a problem for checking theory, as argued in Costa, Moura and Pereira (2001 ), since the pronoun would be inserted in the derivation endowed with the complete feature specification. Since the predicate and the verb should be inserted already inflected, there would be a clash between the features of the verb and the predicate and part of the features of the pronoun. ,
4. Fu rther Predictions
The analysis put forward in the preceding section makes two additional
pre d ic tions :
a) Variation in agreement on the verb and a difference between European and Brazilian Portuguese:
Although in some dialects of European Portuguese, a gente triggers 1 st person plural verbal agreement, there is no dialect in which the pronoun trigge�s plural agreement on the verb, and singular agreement on the predicate, given the independent nature of the two types of features. ( 1 5)
a.
b.
A gente estamos contentes. A GENTE are- l st-pI happy-pi * A gente estamos c ontente A GENTE are-1 st-pi happy-sg .
As mentioned in the previous section, we contend that the plural referential agreement visible on the predicate is a consequence of the dissociated status of the referential number morphology, as argued in Costa and Figueiredo Silva (2002). According to these authors, in European Portuguese, plural dissociated morpheme attaches to all potential hosts. On the other hand, the grammatical agreement fonows only from the feature specification of the pronoun. This derives the ungrammaticality of ( 1 5b). Since the plural morphology on the verb is a consequence of the dissociated status of the number morpheme, ( I Sb) is ruled out because the number morph eme is not attaching to all roots, not surfacing on the predicate.
1 22
Joao Costa and Sandra Pereira
Costa and Figueiredo Silva (2002) hypothesize that number is a singleton morpheme in Brazilian Portuguese. This correctly predicts that sentences like ( 1 5a) are less frequent in Brazilian Portuguese than in European Portuguese. On the contrary, in European Portuguese, ( 1 5a) is becoming more and more frequent. Accepting this difference in the status of the morpheme in the two languages further predicts that in Brazilian Portuguese, a genie never triggers plural agreement on the predicate of the small clause.
b) Behavior oJa gente in null subject contexts. The behavior of this pronoun with respect to the null subject parameter is different from other pronouns. In section 1 , it was observed that normally a genie freely alternates with the first person plural pronoun nos (we ) . However, in null subject contexts, it can be observed that ( 1 6)
nos can be dropp ed, but a genie cannot, as illustrated in ( 16):
a. b.
Vamos embora.
(we) go away.
"'Vai embora.
(A GENTE) goes away.
According to the analysis outlined above, the problem with ( l 6b) is that the referential features of the subject are not present, which prevents recoverability of its reference. In other words, it may be said that the null subject in (16b) is not referential. Incidentally, this fact may indicate that Agr is not necessarily pronominal in null subject languages, casting doubt on recent views on the null subject parameter, such as Barbosa ( 1 995) and Alexiadou and Anagnostopoulou (1 998), who claim that Agr is pronominal in null subject languages. These facts indicate that the pronominal nature of Agr is not defmeable for a whole verbal paradigm, but it may vary depending on the type 3 of subject associated with it.
5. Conclusion The pattern of agreement analyzed in
this
paper was never explained.
Combining phase theory and some of the assumptions of Distributed Morphology provides a straightforward analysis for this problem. Therefore, this type of phenomenon is the kind of empirical evidence supporting these' recent developments in syntactic
theory.
References Alexiadou. Artemis & Elena Anagnostopoulou. 1998. Parametrizing Agr: Word order, verb-movement and EPP-checking. Natural Language Linguistc Theory. 1 6:3, 49 1 -539 Barbosa, Pilar. 1 995. Null Subjects. Doctoral dissertation, MIT.
3
and
See also Costa (200 1 ) and Costa and Duarte (2003) for other arguments against a in which Agr is pronominal .
characterization of null subject languages as languages
1 23
Phases and autonomous features
Chomsky� Noam. 1998. "Minimalist Inquiries", MIT Occasional Papers in Linguistics Chomsky, N. 1 999. "Derivation by Phase" . MIT Occasional Papers in Linguistics. Number 1 8. CORDIAL-SIN: Corpus of Portuguese Dialects Syntactically
Annotated
http://www.ciuJ.ul .ptiengJish/scclorcs/cordialsin/projccto cordialsin.htm l
Costa, Joao. 200 1 . ''Null vs Overt Spec,TP in European Portuguese" to appear in J. Schroten (cd) Romance Languages and Linguistic Theory 200 1 Johns Benjamins, Amsterdam Costa, Joao & M' Cristina Figueiredo Silva. 2002. "Nominal and verbal agreement in Portuguese: an argument for Distributed Morphology". Ms. Universidade Nova de Lisboa Costa, Joao.� Denilda Moura & S. Pereira. 200 1 . Concordancia com a gente: urn problema para a teoria de verificac;:ao de trac;:os. In Actas do XVI Encontro Nacional da Associacao Portuguesa de Linguistica. Coimbra. Setembro de 2000. Costa, Joao & Ines Duarte. 2002. Preverbal subjects in null subject languages are not necessarily dislocated. In Journal ojPortuguese Linguistics� 1 59-1 76 Duarte, IDeS, M. Joao Freitas, Anabela Gonc;:alves, Matilde Miguel & Celeste Rodrigues. 2002. "Geometria de trac;:os e distribuic;:ao de pronomes sujeito em PE e em PB". Paper presented at the 3 rd Workshop do Projecto PE-PB. Lisboa� September 2325. Embick, D. and Noyer, Rolf 200 1. "Movement operations after Syntax". Linguistic Inquiry 32: 555-595. Halle. M. & A. Marantz. 1993. Distnbuted Morphology. In Keyser, J. (ed) The View from Building 20. Cambridge. MIT Press: Harley, Heidi & Ritter. 2000. "Person and number in pronouns: a feature-geometric analysis", ms, University of Arizona Menuzzi, Sergio. 2000. First Person Plural Anaphora in Brazilian Portuguese: chains and constraint interaction in binding. In Joao Costa (ed) Portuguese Syntax. New Comparative Studies. Oxford University Press. Pereira, Sandra. 2000. "A gente: acordos e desacordos". Ms, Universidade de Lisboa Pereira, Sandra. 2003. Gramatica compara de a gente. Varia�ao no portugues europeu. MA dissertation, Universidade de Lisboa Svenonius, Peter. 2003. "Phases at the interface". Paper presented at MIT Workshop on Phases and EPP, January, MIT �
1 24
Towards a phase-based theory of sentential stress· Arsalan Kahnemuyipour University of Toronto The purpose of this paper is to show that sentential stress is detennincd
syntactically in a phase-based manner (Chomsky 2000, 2001 and subsequent work) and that cross-linguistic differences in this respect
follow from syntactic variations. According to the Phase theory. when the derivation reaches a phase, the complement of the phase bead is shipped
off to PF. I propose that sentential stress is assigned to the highest element in this spelled out constituent. This proposal allows for a straightforward account for a wide range of stress facts within a language and cross linguistically.
1. Introduction
The question of which word in a sentence receives sentential stress (also known as Nuclear Stress, Chomsky and Halle 1968) has received considerable attention in the linguistic literature. Two major types of approaches can be identified in this respect. 1 According to the first approach, sentential stress is determined phonologically, either by a language-specific phonological rule (Chomsky and Halle 1 968, Halle and Vergnaud 1 987, among others) or by a mapping between syntax and phonology and a rule that applies to the relevant prosodic domain (works within Prosodic Phonology such as Nespor and Vogel 1 986, Nespor 1 999, Kahnemuyipour 2003, among others). The proponents of the second approach, on the other hand, offer a purely syntactic account of sentential stress (Cinque 1993, Zubizarreta 1 998, among others). According to this view, cross linguistic variations in stress pattern follow from syntactic differences. The present paper falls within this latter approach to sentential stress but differs from the previous proposals in important ways, elaborated below. The paper is organized as follows. In section 2, I briefly review and provide arguments against the phonological approaches to sentential stress. In section 3 , I review Cinque ( 1 993), the most influential syntactic account of sentential stress, and present some theoretical and empirical problems with his theory. Section 4 deals with the main proposal of this paper and shows how it can overcome the $, would like to thank Elizabeth Cowper, Diane Massam and Keren Rice for discussions on different
aspects of this work. Thanks also to the audiences at the University of Toronto and MIT for helpful
questions and comments. All errors are mine. This work is partially funded by SSHRC Canada 75 2-
200 1 - 1 763.
1 Some linguists have argued against the existence of a sentential stress rule altogether (Bolinger 1972 , Gussenhovcn 1984, Selkirk 1995, among others). I am putting such proposals and related
controversies aside in this paper.
MIT Working Papers in Linguistics 49.
J 25-/46
Perspectives on Phases © 2005 Arsalan Kahnemuy;pour
"S""
!
Towards a phase-based theory of sentential stress problems presented in section 3 and account for some other stress facts in P ersian. We will see that the sentence-final stress pattern in English poses an
apparent problem for the proposed analysis. In section 5, 1 tackle this problem (in the case of adverbials) and show that the facts can be reconciled with the
present analysis with some reasonable syntactic assumptions. Section 6 deals with the iterative nature of sentential stress and how it is handled in the proposed system. Section 7 concludes the paper.
2. Phonological Approaches to Sentential Stress Some linguists have proposed that some language-specific phonological rule detennines sentential stress (Chomsky and Halle 1 968, Halle and Vemgaud
1 987, among others; also works within Prosodic Phonology : Nespor and Vogel 1 986, Nespor 1 999, Kahnemuyipour 2003, among others). In this section I evaluate these approaches against crossiinguistic variation in sentential stress. It
is a well known fact that unlike word stress (and perhaps even compound stress), sentential stress does not exhibit much variety cross-linguistically. Thus, for instance, as illustrated in ( 1 ), in SVO, SOY and VSO languages, it is the object
that takes stress. 2
(1)
a.
SVO languages:
English: John read a book. Spanish: Juan ley6
un libro . (Zubizarreta 1 998)
Juan bought a book b.
SOY languages: Persian: Ali ye ketaab xarid Ali a book bought , Ali bought a book.' Basque: Jonek liburu irakurri Jon-ERG book-ABS read 'Jon read the book.'
c.
ban. had (Arregi
20�3)
VSO languages: Scottish Gaelic: chuala Seonag Calum. heard Seonag Calum ' S eonag heard Calum.'
(Adger 2002)
The phonological approaches to sentential stress fail to account for the limited sentential stress patterns found in world languages. In other words, these
2 Throughout this paper, Underlining shows stress. Also, unless specified otherwise, out-of-thc-bluc context is assumed. i.e. Context Question "" What happened? '
1 26
Arsalan Kahnemuyipour
analyses are too powerful, i.e. they allow for stress patterns not observed in any language. Below 1 illustrate the problem for two accounts that fall under this general rubric of 'phonological' . First we look at Halle and Vergnaud's ( 1 987) metrical grid theory and then at Kahnemuyipour (2003) as an example of an account for sentential stress under the prosodic phonology framework. Halle and Vergnaud ( 1 987) use the metrical grid theory and a set of parameters to account for the considerable crosslinguistic variety of stress patterns of words. In the metrical grid theory, constituents are marked on every line by brackets and the head of the constituent is marked by a grid mark on the line above it. The parameters they use to account for the stress patterns are the following3 : ± BND (depending on whether a constituent on a certain line is bounded or unbounded), ± HT (depending on whether a constituent is head terminal or not), left or right (depending on the direction of headedness for + HT constituents), left-to-right or right-to-left (depending on the direction the constituents are constructed). Thus, for instance, the stress pattern of Maranungku, where the stress faUs on odd syllables with the main stress on the leftmost one, is obtained by the parameter settings in (2a). The resulting grid structure for a five-syllable word is shown in (2b).
(2)
8.
Parameter settings for Maranungku Line 0: + HI, +BND, left, left-to-right Line 1 : +HT, -BND, left
b.
Grid structure for 8 five-syllable Maranungku word
* *) * (* (* *) (* *) (*)
line 2 (word) line 1 (foot) line 0 (syllable)
Different choices of these parameters account for other word-level stress patterns observed crosslinguistically. Halle and Vergnaud extend the same stress theory to account for the stress pattern of phrases and sentences. Thus, for instance, they propose the following system for constructing the grid structure for lines 3 and above to incorporate into their theory Chomsky and Halle' s ( 1 968) Nuclear Stress Rule which essentially predicts stress on the rightmost element in English sentences.
(3)
Nuclear Stress Rule a. b. c.
Parameter settings on line N (N greater than or equal to 3) are [-BND, +HT, right]. Interpret boundaries of syntactic constituents composed of two or more stressed words as metrical boundaries. Locate the heads of line N constituents on line N + 1 .
3 This is a simplified representation of their theory. A few more parametm are in facl introduced to account for the stress pattern in some languages. Such details are irrelevant to the point being made
here.
1 27
Towards a phase-based theory of sentential s tre ss
Using the above system while observing the p rincip le of the cycle (thus starting from the innermost syntactic constituent working outwards), one can give the following grid structure for a simple SVO sentence; e.g. the English sentence John saw Mary. *
(4) *
S [John
(
* *
V [saw
*
0
) )
line 5
line 4 line 3
�]
The crucial point here is that in l ooking at the sentential stress patterns of different languages, we are not confronted w ith the same range of p ossibilities found at the word level. There is nothing in the above system that would disallow an SVO language with the same constituent structure and direction of branching as English but with the following parameter settings: BND, +HT, left. This would predict stress on V in a transitive SVO sentence as shown in (5).4 Such a language does not exist The direction of headedness seems to correl ate with the direction of branch ing, so a right-branching language like English has its head parameter set to 'right' (see Cinque 1 993 for a discussion of these issues and a potential solution). *
(5) (
*
S [John
*
(
*
V [saw
*
0
) )
line 5 line 4 line 3
Mirod]
The other phonological treattnent of sentential stress falls within the general framework known as 'ProsodiclPhrasal Phonology' (Selkirk 1980a, b , 1 98 1 , 1 984, 1986; Nespor and Vogel 1 982, 1 986). It has been long observed that some phonological rules appear to be sensitive to syntactic structure in one way or another. To account for such interactions between phonology and syntax, phrasal phonology adopts an indirect approach in which: L Mapping rules derive phrasal domains from morphosyntactic constituents; 2. Phonological rules then apply with reference to these phrasal domains. The hierarchically organized prosodic domains, the phonological word, the phonological phrase, the intonational phrase and the utterance are derived from the syntactic constituents Xo, XP, (Root) Clause and Utterance, respectively. There is a vast body of literature on the relevance of these domains for segmental phonological phenomena, and a few linguists have utilized them to serve as domains for the assignment of sentential stress (e.g. Hayes & Lahiri 199 1 , Nespor 1 999, Hsiao 2002, Kahnemuyipour (2003) . Kahnemuyipour (2003), for instance, accounts for the assignment of stress in Persian using the following s ettings : rightmost at 4
1n fact. Halle and Vergnaud's actual system would make a slightly different prediction. In addition
to the above rules, their system includes a Stress Equalization Convention. If that is added to the picture, the predicted stress in a transitive
SVO
sentence would go on S, still unattested in focus·
neutral sentences crosslinguistically. Thus, the point being made here remains valid.
1 28
Arsalan Kahnemuyipour
the phonological word level, leftmost at the phonological phrase level and rightmost at the intonational phrase level. These settings correctly predict sentential stress on the object in a Persian SOY sentence as in (6). *
Intonational Phrase
...
*
Phonological Phrase
(* )
(*
S
0
(6)
*
)
Phonological Word
V
xarid]] [Ali rye ketaab bought Ali a book 'Ali bought a book.' The above system suffers from the same problem that a Halle and Vergnaud type analysis suffers from, i.e. overgeneration. Nothing in the system would disallow a language with the same settings for the levels of phonological word and phonological phrase as Persian but a leftmost stress at the intonational phrase level. This would be an SOY language that consistently puts sentential stress on the subject in neutral focus sentences, another unattested case. Unlike variation at the word level, the cross-linguistic variation of sentential stress is much more restricted than either of the above phonological accounts would allow. In addition to the overgeneration problem, some stress facts (e.g. the behavior of subjects in passive/unaccusative sentences) seem to beg for a syntactic account. These facts will be discussed in sections 3 and 4. I will show that a syntactic account, as elaborated below, does not face the same overgeneration problem. It is worth noting, however, that providing a syntactic account of sentential stress does not entail that all other phonological phenomena accounted for within a 'Phrasal Phonology' framework can be reworked in purely syntactic terms. In fact, there are some reasons to believe that stress is distinct from other phonological processes in tenns of the way it interplays with syntax, e.g. the interaction between sentential stress and information structure, scope, etc. The debate between an indirect approach to the phonology-syntax interaction, a la phrasal phonology, and a direct one in which phonological rules can refer to syntactic structures directly is not a new one (see, for example, Cooper and Paccia-Cooper 1 980, Kaisse 1 985, Odden 1 987, 1 990, Rizzi and Savoia 1 992 for a direct approach; and Selkirk 1 980a, 198 1 , Nespor and Vogel 1 986 and Hayes 1 989 for arguments against a direct approach). With the recent advances in syntactic theory and the introduction of phases and multiple spell-out, the debate has taken a new perspective. In a theory in which syntactic structure is sent off to PF in chunks known as phases, the question is now whether phases can replace the prosodic domains used in phrasal phonology (see, for example, McGinnis 200 1 , Collins 2002). In this paper, I limit my scope to sentential stress and leave the possibility of extending this approach to other phonological phenomena for future research.
1 29
Towards a phase-based theory of sentential stress
In the fol lowing section, we look at Cinque ( 1 993), the most influential syntactic account of sentential stress, and some theoretical and empirical problems with his theory will be presented.
3. A Syntactic Approach to Phrasal Stress: Cinq ue (1993) Cinque ( 1 993) proposes that phrasal stress (as well as compound stress) can be determined syntactically. He builds a system which predicts phrasal stress on the 'most deeply embedded' element in the phrase. A detailed il lustration of how
stress on the 'most deeply embedded' element is obtained is beyond the scope of this paper. The crucial point is that an element which is embedded in more layers of structure will receive primary stress. As he notes, this accounts, for example, for the sentential stress on the object in both SVO (e.g. English) and
SOY (here, exemplified with Persian) in (7). Note that in both SVO and SOY the most deeply embedded element is the object, assuming, as Cinque does, that the head parameter is what accounts for the order difference between SVO and SOY. In other words, in both SVO and SOY, the object, being the complement of the verb, is buried in more syntactic structure than any other element in the
sentence. The bracketing in (7) shows this point more clearly. In simple terms, the element with more syntactic brackets around it is what is predicted to receive
primary stress. ( 7)
a.
[DpJohn] [vpbought [opa b ook]]
b.
[DpAli] [[vP[DPye ketaab] xarid] Ali a book bought
In the remaining of this section I will discuss some conceptual and empirical problems with Cinque's stress system.
3.1 Conceptual problems with Cinque's system
The main conceptual problem with Cinque's system is with the way he handles sentential stress in sentences with a complex subject Recall that in his system
the 'most deeply embedded' element in the sentence receives primary stress. Thus, if the subject has more layers of embedding than the predicate, Cinque's system would predict stress on an element within the subject. This is contrary to fact, as illustrated in (8). While Cinque's system would predict primary stress on 'senescence', the element with most layers of syntactic structure on top of it, it is the word ' Mary' that receives primary stress. (8)
[The author of [many popular articles on [the effects of [senescence]]]] [kissed [Mm:rll.
Cinque notes this empirical problem, but the solution he offers introduces a conceptual problem into his system.5 To solve this problem, Cinque 5 Cinque uses an example that is slightly different from (8), but the stress facts are not as uncontroversial as the one used here.
1 30
Arsalan Kahnemuyipour
resorts to 'information structure'. He argues that, depending on whether the 'subject' or the 'predicate' provides new information, the stress may fall on one or the other. Within each constituent, however, stress obeys the 'most deeply embedded' generalization. Using ' information structure', however, to account for the stress pattern in (8) is questionable. A null theory of stress should be able to account for the out-of-the-blue (neutral focus) stress in (8) without recourse to 'information structure'. There seems to be a fundamental difference between the subject and the predicate regardless of the depth of embedding which Cinque's system fails to capture. Let me end this section by pointing out another important characteristic of Cinuqe's proposal, its incompatibility with an anti-symmetric theory of syntax. As noted above, Cinque's system crucially relies on the head parameter. So, for example, the object is the complement of the verb and is thus considered 'most deeply embedded' in both head-final SOY and head-initial SVO languages. If the head parameter is dispensed with (Kayne 1 994 and subsequent work), it is not clear how this system could be made to work. Note that in Cinque's system, stress is crucially read off surface syntactic structure, making unavailable an account in which SOY starts off as SVO with the 0 receiving stress in its merge position where it is the 'most deeply embedded' element. In short, in a system which derives SOY from SVO, the object cannot be the verb's complement, the most deeply embedded element, in its final surface position, and thus not a candidate for sentential stress in Cinque's system. In the following section we look at some empirical problems with Cinque's stress system. 3.2 Empirical problems with Cinque's system
The first empirical problem with Cinque's system is that it fails to account for stress on the subject in passives and unaccusatives, as shown in (9) (see also Zubizarreta 1 998). The most deeply �mbedded element in both cases is the verb. (9)
a.
(What happened yesterday?) My bike was stolen.
b.
(What happened?)
Ali umad Ali came
(Legate 2003) Persian
Legate (2003) attempts to account for (9) by proposing that a Cinque· style sentence stress rule applies in a phase-based manner (Chomsky 2000, 200 1 ). To better understand her analysis and my proposal laid out in the following section. let us briefly review the mechanics of a phase-based theory of syntax. According to the phase theory. syntactic structure is sent off to PF and LF for interpretation in chunks. Derivation proceeds bottom-up in a phase-by phase manner. When the derivation reaches a phase, the (syntactic) complement of the phase head is shipped off for phonological interpretation. I will call this syntactic chunk the SPELLEE. This is illustrated in ( 1 0).
131
Towards a phase-based theo ry of sentential stress ( ( 0)
=
SPELLEE � PF (Phonological Interpretation)
What constitutes a phase is a matter of controversy. I follow Chomsky in this respect who takes CPs and (transitive) vPs to constimte phases. In addition, Chomsky crucially takes unaccusative and passive vPs not to induce phasal boundaries.6 Legate, on the other hand, differs with Chomsky in this respect. In her system, unaccusative and passive (just like transitive) verb phrases are phases. This is crucial to her proposal which I will return to shortly. With this background about the phase theory, let us return to Cinque' s problem with passives and unaccusatives and Legate's proposed solution. As discussed earlier, Cinque's system, which predicts stress on the 'most deeply embedded' element, fails to account for stress on the subject in passives and unaccusatives (see (9»). To solve this problem, Legate proposes that a Cinque style sentence stress rule applies in a phase-based manner. Recall that in her system, unaccusative and passive (just like transitive) verb phrases constitute phases. She also makes use of the copy theory of movement. According to this theory, every instance of movement involves two identical occurrences of the moved element, one in the position it has moved from and one where it has moved to. If there are intennediate landing sites, there will be more than two occurrences. These occurrences are called copies. Instructions to the LF and PF components detennine which copy should be interpreted or pronounced, respectively. In Legate's system, the stress rule applies to the lowest copy at the phase, and the highest copy, which is the one that is pronounced, inherits the stress from its lower counterpart. Note that, according to this system, any element which moves from a stress-bearing position within a phase to a position
6 Sometimes the crucial distinction is described in tenns of strength, i.e. CPs and transitive vPs arc called strong phases, while unaccusativclpassivc vPs are called weak phases. The important point is that syntactic structure is transferred to PF and LF only at the point of a strong phase. For the sake of simplicity, phase
I
will avoid the tenns strong/weak and use 'phase' to refer to what is known as 'strong'
in the other classification. Weak phases, which by definition do not induce transfer to PFILF,
will be considered non-phases.
1 32
Arsalan Kahnemuyi pour
outside the phase should receive stress. This is contrary to the facts in ( 1 1), i.e. the wh-word and the topicalised noun phrase do not receive stress.7,S
(I t)
a. b.
What did John hY.Y? Beans,
I like.
To summarize, we have seen so far that Cinque's stress system faces a problem in the face of passive and unaccusative sentences and that Legate's solution creates new problems. An additional problem with Cinque's system is that it predicts stress on the element immediately preceding the verb in SOV languages. This is contrary to Persian facts, as will be shown in detail in section 4. 1 . In the following section, I propose a phase-based system which overcomes these problems. 4. Sentential Stress: An Alternative Account
To overcome the problems stated above and to account for some stress facts in Persian, I propose the following mechanism for the assignment of sentential stress.
( 1 2)
Sentential stress is assigned at the phase to the highest element of the spelled out constituent (here called the SPELLEE). [ HP [H YP]]: If HP is a phase, YP SPELLEE Proposal:
(i.e. the phonological border)
=
In the English and Persian transitive examples (7), the object moves to the specifier position of the AspP (1 3) (Travis 1 992 and subsequent work) perh aps for case assignment (On movement of the object see also Johnson 1 99 1 , Jelinek and Carnie 2002, among others; on the correlation between direct object/accusative case and aspect see Tenny 1994, Ritter and Rosen 200 1 , Svenonius 2002, among others.). (I put aside any V-movement in these examples, as it does not interact with the stress system proposed here.) Thus, at the phase vP, the highest element in the SPELLEE, the object, receives stress.
7 The fact that the wh-word in (I I) does not receive stress carmot be attributed to some idiosyncracy of wh·elements. Note that wh-words receive stress in-situ (Who bought �?) and in a language
such as Persian (Ali chi xarid? Ali what bought?), a focus-fronting language (see Kahnemuyipour
2001 ).
S Given that the examples in (9) involve A-movement as opposed to A' -movement in ( I I ), one may be inclined to use this distinction to .account for the observed contrast. Note, however, the difference
cannot follow from the A-A' distinction in any natural way. One can, of course, stipulate the difference by stating. for instance, that only A-copies inherit stress from their lower counterpan.
Alternatively, if we give up the copy theory of movement and try to account for the stress facts by ordering the stress rule with respect to A- and A' ·movements in a way that it would only apply to the element that undergoes A-movement, we would have to allow A'-movement to be ordered before and A-movement after stress assignment, a conceprually implausible result.
133
�
Towards a phase-based theory of sentential stress
( 1 3) v
Stress Domain
=
SPELLEE
f �/�SPI AS� y--
C
V
Obj
In the remaining part of this section, we will see how the above proposal accounts for some stress facts in Persian and how Cinque's system fails to account for some of these facts. I will also account for the unaccusative/passive facts under the proposed system. 4.1 Interesting Persian facts
I show below how the proposed system can account for some interesting stress facts in Persian. The first case at hand is the position of sentential stress in Persian when more than one element appears to the left of the verb in the stress domain. In Persian, if a single element appears to the left of the verb within the stress domain (or the SPELLEE), it receives stress. (1 4a) shows a simple sentence with a verb with the stress on the verb. (1 4b) shows that if a mood marker is prefIxed to the verb, stress falls on the mood marker (for the status of these mood markers, see Kahnemuyipour 2003). ( 1 4c) is an example of an SOV sentence with a non-specific object. Stress fall on the object as expected. Finally, ( 1 4d) shows · that if other elements appear to the left of the object within the stress domain (e.g. Left Edge Markers like some manner adverbs), they receive stress. Ali [ AspP xordl ate Ali b.
Ali [AspP mi - xord] IND IC- ate Ali
c.
Ali [ASp P gazaa xord] Ali food ate
d.
Ali [AspP xub qazaa xord] Ali well food ate
Let us see how Cinque's stress system would fare with the above data. Recall that Cinque's system predicts sentential stress on the 'most deeply embedded' element in the sentence, which in the case of an SOV language would be the element to the immediate left of the verb (Cinque 1993 : 250). This stands in sharp contrast with the proposed system which predicts stress on the leftmost element in the 'stress domain' or the SPELLEE. Sentences such as ( 1 4d) pose a problem to Cinque's system. In this example, stress falls on the
1 34
Arsalan Kahnemuyipour
leftmost element in the AspP, rather than the element immediately to the left of the verb as predicted by Cinque. 9 The second case concerns the stress behavior of non-specific and specific objects. In Persian, there is a striking contrast between the stress pattern of a non-specific object ( 1 5a) (repeated from ( 1 4c» and a specific one ( I Sb) which receives a straightforward account under the proposed analysis (Cinque 1 993 discusses similar facts for Gennan). Following a standard assumption (see Diesing 1 992, Koopman and Sportiche 1 99 1 , Mahajan 1 990, among others), the specific object is proposed to move to a higher syntactic position outside the stress domain (e.g. SpecvP), supported by its relative position to manner adverbs ( l Sc). Again, the highest element in the SPELLEE, i.e. the non-specific object ( I Sa), the verb ( 1 Sb) and the manner adverb ( I Sc) receive stress. I 0 (15)
a.
Ali [ASPP gazaa xord] food ate Ali
b.
Ali qazaa-ro [AspP xordl ate Ali food-SPEC
c.
Ali qazaa-ro [ASPP xub xord] well ate Ali food- SPEC
4.2 Passives and unaccusatives
Returning to the passives and unaccusatives repeated in ( 1 6), as standardly assumed, passive/unaccusative verb phrases do not constitute phases, so the first SPELLEE is TP, with the highest element, the subject, receiving stress. ( 1 6)
a.
(What happened yesterday?) My bike was stolen.
b.
(What happened?) Ali umad Ali came
(Legate 2 003)
9 Whether Cinque's prediction of sententiaJ stress on the element to the immediate left ofthc verb in SOY languages holds true of Gennan and other languages he discusses requires further exploration. Some preliminaty investigation indicates that his generalization may in fact run into trouble in the face of some German data, but I leave the question open until the data bas been further verified. If, in fact. it turns out that Cinque's generalization works for some SOY languages other than Persian, one tentative solution to account for such variation under the system proposed here would be to hypothesize that languages differ on bow much they include in the SPELLEE or the stress domain. That
is to say, for instance, in one language such as Persian, manner adverbs may be part of the
SPELLEE, while in others they may be outside this domain.
1 0 It is worth noting that according to this proposal for the assignment of sentential stress, I ike that
of Cinque's, elements that undergo syntactie movement may escape the stress domain and thus avoid sentential stress, e.g. the specific object
in Persian. In other words, no clement can receive stress due
to its base position before undergoing a syntactic movement (contra Bresnan 1 97 1 , 1 972 , Legate
2003;
but see Lakoff 1 972. Bennan and Szamosi 1972 and Bolinger 1 972 for critical replies to
Bresnan).
1 35
Towards a phase- based theory of sentential stre'ss Thus, the present analysis has another advantage over Cinque's system in being able to capture the facts in ( 1 6). Moreover, it does not make the problematic predictions about wh- or topicalised elements under a system like Legate' s (2003) which allows sentential stress to be sensitive to the merge position of an element. In addition, the fundamental difference between a subject and the predicate is built into the system, thus avoiding the further stipulation Cinque had to make by resorting to ' information structure' for sentences with complex subjects. Finally, this system does not rely on the head parameter and is thus compatible with an anti-symmetric system like that of Kayne ( 1 994). There is a potential problem for the above treatment of sentential stress in passives. It has been observed that when an adverb is added to passive sentences, sentential stress no longer falls on the subject, compare ( 1 7a) and ( 1 7b) (see, e.g. Zubizarreta 1 998). We have seen that Cinque's stress system could not account for the stress on the subject in ( 1 7a). We captured ( 1 7a) by proposing that stress is assigned in a phase-based manner and that in the case of passives, which do not constitute a phasal boundary, the highest element in the SPELLEE which receives stress would be the subject in SpecTP. But why do we not observe the same pattern in ( 1 7b)? Here is a tentative solution: In ( 1 7b), the adverb may have imposed extra structure, causing the verb phrase to become phasal. This would mean that a passive/unaccusative verb phrase does not constitute a phase only in its barest form such as ( 1 7a). In fact, this may be a way of reconciling the stress facts with Legate's (2003) syntactic arguments for passives and unaccusatives constituting phases, where all tests involved extra syntactic structure. Meanwhile, as pointed out to me by a few native speakers of English, the facts in ( 1 7) may be related to an interaction between ' information structure' and stress. In other words, by adding the adverb, the verb becomes more prominent from an informational perspective. To provide a convincing account of the facts in ( 1 7), these possibilities need to be further explored. ( 1 7)
a.
b.
My bike was stolen. My bike was mysteriously stolen.
There is an immediate question which we will have to address in order to maintain the proposed system. How can we account for all the other apparent cases of sentence-final stress in English? To address this question, I start with adverbials (including adjunct prepositional phrases) and then extend ·th e proposal to (argument) prepositional phrases.
5. Adverbials In this section 1 build on Cinque's ( 1 999, 2002, 2004) work on the syntax of adverbials and show how the stress behavior of adverbials in Persian and English can be accounted for within the stress system proposed here. It is important to note at the outset that Cinque's ( 1 999, 2002, 2004) work on adverbs is based on an underlying assumption which is incompatible with his work on stress. While his work on stress relies crucially on the he�d parameter, in his work on adverbs, Cinque (following Kayne 1 994) dispenses with the head
1 36
Arsalan
Kahnemuyipour
parameter and lays out his theory in an antisymmetric framework. Recall, for instance, that the reasoning for why the object receives stress in Cinque's work on stress in both SVO and SOY l anguages is that in both cases the object is the complement of the verb (in surface structure) and receives stress as the 'most deeply embedded' element in the structure. If, as assumed in an antisymmetric framework, SOY is derived from an SVO order by the movement of the object, the object could not be the complement of the verb in surface position in SOY and thus its stress would be unaccounted for in Cinque's stress system. In what follows, 1 am departing from Cinque's work on stress and will use his work on adverbs as a framework to account for the stress behavior of adverbials within the system I have proposed in this paper. 5.1 The Syntax of Adverbials
While Cinque (1 999) provides a detailed breakdown for a universal order of adverbs, the following less refined order has been generally accepted in the literature (Jackendoff 1 972 and subsequent work). ( 1 8)
speaker-oriented adverbs > subject-oriented adverbs > manner adverbs
There is a class of adverbials of place, time. manner, etc. which are typically prepositional phrases and appear after the complement of the verb in English (e.g l gave a talk at MIT on Thursday). The structure and order of these adverbial PPs have been the topic of Cinque (2002, 2004). The relevant points are summarized below, details irrelevant. Cinque (2002, 2004) points out that there have been two different sets of evidence, each pointing to a different structure for these adverbs. While lack of Principle C effects (e.g Lakoff 1 968, Reinhart 1 983), constituency diagnostics (e.g. Pesetsky 1 995, Nilsen 2000), relative scope (e.g. Manzini 1 995, Pesetsky 1 995, Brody 1 997) point to the left-branching structure in ( 1 9a), anaphor binding (e.g. Pesetsky 1 995), pronominal binding (e.g. Stroik 1 990, Pesetsky 1 995) and licensing of negative polarity items (e.g. Stroik 1 990, Pesetsky 1995) seem to favor the right-branching structure in ( 1 9b). .
.
( 1 9)
a.
IP
/'..
DP
I'
/'..
I
VP
/'..
VP
/'..
VP
/'..
V
PPz
PP,
DP
1 37
Towards a phase-based theory of sentential stress
b.
IP
/"-....
DP
I'
/"-....
VP
�
V'
/"-....
V
VP
�
DP
V'
�
V
VP
/"-....
pp)
V'
�
V
PP2
To resolve this paradox and to account for some ordering restrictions, Cinque proposes that these adverbials are preverbal at merge and that their postverbal surface position is the result of the movement of the lower elements (the verb and the DP object) around them (see also Nilsen 2000). The merge position and the movement are shown schematically in (20). 1 1
(20)
DPsubject
. . .
T
. .
. PP2
. . , PPI
.
. . . VP
.J
I adopt the analysis in (20) and will show in the next section how the str�ss facts in Persian and English with respect to adverbials can be accounted for. 5.2. Stress facts: Persian vs. English Let us now look at the stress behavior of different types of adverbials in Persian and English in light of the syntactic analysis presented above. The three observed patterns are discussed below. 5.2. 1 Not stressed in Persian. not stressed in English
This pattern happens with the high adverbs (refer to ( 1 8» . An example for a speaker-oriented adverb 'probably' is provided in (21 ). The stress pattern is expected; the high adverbs are outside the stress domain and thus unstressed. (2 1 )
a. b.
I'll probably take the subway. man ehtemaalan esteyk mi-xor-am I probably steak INDIC-eat-l SG 'I'll probably have steak'
1 1 Cinque's structure differs in details that are irrelevant for the purposes of this proposal. 1 38
A rsalan Kahnemuyipour
5.2.2 Stressed in
Persian, stressed and sentence-final in English
This pattern occurs with manner adverbs (perhaps as well as some measure adverbs like 'a lot', etc . ) which are proposed to be very low (refer to ( I 8» . For
Persian� it was suggested above that they mark the left edge of the AspP (or the SPELLEE) and are thus stressed. Now, if as suggested by Cinque (2002, 2004), their sentence final order in Eng l ish is the result of the movement of lower elements around them (see (20» , they could in fact occupy the same position as Persian, i.e. they mark the left edge of AspP in both languages. The fact that they are stressed in Persian and English (as in (22) and (23» thus follows. The order difference has to do with some independent syntactic difference between the two languages to be explored in future. (22)
(23)
a.
John played the game well.
b.
John likes apples a lot.
a.
Ali xub baazi kard Ali well play did , Ali played welL'
b.
dust dare Ali xeyli sib Ali a lot apple friend has 'Ali likes apples a lot'
It has been noted in the literature that the same adverb can be used in different positions in English with different interpretations, indicating that each position is associated with a certain interpretation (Jackendoff 1 972, Cinque 1 999). This is shown in (24) and (25).
(24)
(25)
a.
John cleverly answered their questions.
b.
John answered their questions cleverly .
a.
He quickly ran away.
b.
He ran away quickly.
-+
-+
subject-oriented manner
(Travis 1988)
While (24a) roughly means 'It was clever of John to answer their questions', (24b) means 'John answered their questions in a clever manner' . Similarly, (25a) roughly means 'He immediately ran away', whereas (2 5b) means 'He ran away fast'. Interestingly, this type of distinction is indicated by stress in Persian, underlining the fact that in one case the adverb is inside the stress domain while in the other it is outside it. This is shown in (26).
(26)
a.
Ali tond gors xord Ali fast pill ate tAli immediately took a pill.'
1 39
Towards a phase- based theory of sentential stress b.
Ali Ali
tond fast
qazaa food
'Ali ate quickly .'
xord ate
5.2. 3 Not stressed in Persian. stressed and sentence-final in English This pattern occurs with circumstantial adverbials which are merged higher than the manner adverbials, thus outside the stress domain. The stress pattern in Persian, illustrated in
(27),
where the adverbial is unstressed. is thus expected.
The English stress pattern, shown in
(27)
(28),
on the other hand, poses a problem.
maa tu paark futbaal baazi kard-im we in park soccer play did- l pI 'We played soccer in the park . '
(28)
I walked the dog in the park.
The question arises as to how the circumstantial adverbials receive stress in
English. We have seen so far that sentential stress is assigned to the highest
element in the stress domain. i.e. the SPELLEE. Meanwhile, if we adopt
Cinque's analysis of these adverbials, that is to say, if their fmal position is the
result of the movement of lower material to a position higher than the adverbials,
at the time of stress assignment, the stress domain will be
phonologically empty. It is in these cases that I propose that stress is realized on the closest phonologically non-nu ll element, thus for example on the locative in
(28).
The last category we explored in this section involved
adjunct
prepositional phrases. Let us examine the behaviour of argument PPs. While
adjunct PPs used as circumstantial adverbs do not attract stress (see (27) above), argument PPs as shown in the ditransitive examples in (29) attract str�ss. 1 2
(29)
a.
Ali
shir-o
tu yaxchaal
milk-SPEC in fridge 'Ali put the milk in the fridge.
gozaasht put
Ali
b.
Ali
tup-o
be Hassan
'
daad
Ali ball-SPEC to Hassan gave 'Ali gave the ball to Hassan' We thus see a contrast between the stress behavior of argument PPs in (29) and adjunct PPs in (27). This follows straightforwardly in the proposed system if we
take the argument PPs to be inside the stress domain or the AspP, and thus receiving stress. In other words, the generalization is that all internal arguments,
1 2 The sentences in (29) have variants in Persian with the PP at the end of the sentence. The stress facts remain me same with the stress falling on the PP. This variation in the ordering is not crucial to the stress system proposed here and is thus ignored.
140
Arsalan Kahnemuyipour in a neutral focus context, with the exception of specific objects, stay inside the stress domain' and receive stress. Thus, non-specific objects and argument PPs receive stress in Persian. If we have both the non-specific object and the argument PP present in the sentence as in (30a), stress falls on the leftmost element, i.e. the non-specific object as predicted by this system. Moreover, if we add a manner adverb the stress falls on the manner adverb, the leftmost element in the stress domain. (30)
a.
Ali shir tu yaxchaal Ali milk in fridge
gozaasht put
'Ali put milk in the fridge.' h.
Ali xub tup be Hassan daad Ali well ball to Hassan gave 'Ali assisted Hassan in the (soccer) game well.'
(Lit. Ali gave ball to Hassan well.)
The contrast in the stress behavior of PP adjuncts and PP arguments has also been noted for German, as shown in (3 1 ) (adapted from Wagner 2003), supporting the proposed analysis. 1 3 (3 1 )
a.
PP complement
Pete/er ist in den Garten getanzt Peterlhe is into the garden danced •Peterlhe danced into the garden.' b.
PP Adjunct Peter/er hat im Garten Peterlhe has in. the garden
getanzt danced 'Peter/he danced in the garden. '
In English, however, the argument PPs are sentence-final and stressed, as shown in (32), In other words, argument PPs fall with manner adverbs
disc.ussed in 5.2.2, i.e. they are stressed in both Persian and English, while in English they are also sentence-fmal. The manner of stress assignment is thus identical to that of manner adverbs. They receive stress as the highest (or leftmost) element in the stress domain. 1 4 Their final order in English is the result of movement of lower elements around them.
(32)
a. b.
John put the milk in the fridge. John gave the ball to Bill.
1 3 In conformity with this stress only.
14
In fact. we have seen
paper, I
ahat when
have changed fhe notational conventions and marked primary llIere arc yet higher elements in the stress domain. those higher
elements receive stress. as in (30) above.
141
Towards a phase-based t h eory of sentential stress We have been able to show so far that with the proposal in ( 1 2) and some reasonab le assumptions about phrase structure we can account for a wide range of stress facts in simple clauses. Note, however, that a system like the one proposed in ( 1 2) which assigns stress in a phase-based manner predicts iterative stress assignment. This prediction will be explored in the following section. 6. Multiple Spell-Out and Iterative Stress Assignment
It has been proposed in this paper that sentential stress is assigned in a phase based manner. Phase-based multiple spell-out is an iterative operation. When derivation reaches a phase, the complement, here called the SPELLEE is sent off to PF, and as suggested here it undergoes the stress rule which assigns stress to the highest element in the SPELLEE. According to Chomsky (200 1 , 2002), CPs and (transitive) vPs constitute phases. So far we have only considered one phase, i.e. the transitive verb phrase vP. Even in a simple clause, however, there is another phasal boundary that we have to deal with, namely CPo One would expect the same stress assignment procedure to apply to CP, and in the case of more complex sentences, to higher phases, i.e. vPs and CPs of higher clauses. In this section, I explore this issue to some extent, but leave a more detailed investigation for future research. Let us start with a simple CP. When the derivation reaches C, its syntactic complement, i .e. TP, is spelled out. The SPELLEE or the stress domain in this case is therefore TP. This is shown in (33).
(33)
1 42
Stress Domain
=
SPELLEE
Arsalan Kahnemuyipour
If according to the proposed system sentential stress is assigned to the highest element in the phase, one would expect that the subject, which occupies the highest surface position in TP to receive stress. In other words, while stress is assigned to the object at the vP phase, at the CP phase stress is assigned to the subject. Thus, in a simple SVO sentence, we expect stress on both the object and the subject. This prediction is in fact borne out. In a simple SVO sentence, while the object receives primary stress, the subject is more prominent than the verb and receives secondary stress. In more general tenns, in a simple clause, the internal argument (or to be more precise the highest element in the lower SPELLEE which could be the object, the argument PP, the manner adverb, etc.) receives primary stress and the external argument (or the subject) receives secondary stress. This is shown in the examples in (34) where I am using numbers to show primary and secondary stress. The second example from Halle and Vergnaud ( 1 987) has been modified according to these notational conventions.
(34)
a.
John saw Mmy. 2 I
b.
Jesus preached to the people of Judea. 2 1
If we build more structure on top of a single clause, i.e. if we deal with more complex sentences with complement clauses, the generalization is clear. There is one stressed element for each phase (or for each SPELLEE to be more precise). Let us look at a sentence with a complement clause given in (35). Here, we are dealing with four phases, two for the lower clause and two for the higher one. The object and the subject of each clause receive stress according to the proposed system.
I have been able to show using some basic examples that.the iterativity built into the system correctly accounts for the elements that receive some level of stress in simple or complex clauses. Nothing in our system, however, predicts that one of the stressed elements has more prominence that the others (e.g. in (35), "Bill' receives primary stress and the rest secondary stress). Let us postulate at this point that it is the lowest phase that receives primary stress and all the other phases receive secondary stress. This would correctly account for the stress facts we have seen so far. The details of how the notion of 'lowest' is incorporated into a bottom-up multiple spell-out system needs to be explored further. 7. CODclusioD
In this paper we examined the nature of sentential stress and how it is assigned. It was proposed that the position of sentential stress is detennined syntactically
143
Towards a phase-based theory of sentential stress
and that cross-linguistic differences follow from syntactic variations. The idea that sentential stress is assigned in a purely syntactic manner is not a new one. A particular manner of stress assignment, however, was promoted in this paper, one that is based on the notion of 'phases' and 'multiple spell-out' . It was proposed that the sentential stress rule applies in a phase-based manner and assigns stress to the highest element in the spelled out constituent called the SPELLEE. It has been shown that both the central claim that sentential stress is assigned syntactically and the particular phase-based proposal allow for a straightforward account of a wide range of stress facts across languages without facing the conceptual and empirical problems of previous accounts. References Adger, D. (2002). Syntax and prosody in Scottish Gaelic. Paper presented at the VSO Workshop, Stuttgart. Arregi, K. (2003). Nuclear stress and syntactic structures. Paper presented at the l SI North American Syntax Conference, Concordia University. Berman, A. and M. Szamosi. ( 1 972). Observations on sentential stress. Language 48(2):
304-325. (1 972). Accent i s predictable (if you're a mind reader). Language 48:633-
Bolinger, D.
644.
Bresnan, J.
( 1 97 1). Sentence stress and syntactic transformations. Language 47(2), 25728 1 . Bresnan, 1 . ( 1972). Stress and syntax: a reply. Language 48(2), 326-342. Brody, M. (1997). Mirror theory. Ms., UC London. Chomsky, N. (2000). Minimalist inquiries: The framework. In R. Martin. D. Michaels and J. Uriagereka (eds.), Step by Step: Essays in Minimalisl Syntax in Honor of Howard Lasnik, Cambridge: MIT Press. Chomsky. N. (200 1 ). Derivation by phase. In M. Kenstowicz (ed.), Ken Hale: A Life in Language, Cambridge: MIT Press. Chpmsky, N. and M. Halle. (1 968). The Sound Pattern ofEngLish. New York: Harper and Row. Cinque. G. (1993).
A null theory of phrase and compound stress. Linguistic Inquiry 24, 239-297. Cinque, G. (1 999). Adverbs and Functional Heads: A Cross-linguistic Perspective. New York: Oxford University Press. . Cinque, G. (2002). Complement and adverbial PPs: Implications for clause structure. Paper presented at CONSOLE XI, Universita di Padova. Cinque. G. (2004). Issues in adverbial syntax. Lingua 1 14: 683-7 1 0. Collins. C. (200 1 ). Eliminating labels. In S. Epstein and D. Seely (eds.), Derivation and Explanation in the Minimalist Program, Oxford: Blackwell, 42-62. Cooper, W. E. and 1. Paccia-Cooper. ( 1 980). Syntax and Speech. Cambridge: Harvard University Press. Diesing, M. (1992). Indefinites. Cambridge: MIT Press. Gussenhoven, C. (1984). On the Grammar and Semantics of Sentence-Accents. Dordrecht: Foris. Halle, M. and J . -R Vergnaud. ( 1 987). An Essay on Stress. Cambridge: MIT Press. Hayes, B. ( 1 989). The prosodic hierarchy in meter. In P. Kiparsky and G. Youmans (eds.), Rhythm and Meter, Orlando: Academic Press, 20 1 -260. Hayes, B. and A. LahirL ( 1 991). Bengali intonational phonology. Natural Language and
Linguistic Theory 9. 47-96.
1 44
Arsalan Kahnemuyipour
Hsiao, F. (2002). Tonal domains are stress domains in Taiwanese: Evidence from focus. MITWPL 42. 1 09-140. lackendoff, R. ( 1 972). Semantic Interpretation in Generative Grammar. Cambridge: MIT Press. Jelinek, E. and A. Carnic. (2002). Argument hierarchies and the mapping principle. Ms., University of Arizona. Johnson, K. ( 1 99 1 ). Object positions. Natural Language and Linguistic Theory 9: 577636. Kahnemuyipour. A. (200 I ). On wh-questions in Persian. Canadian Journal ofLinguistics 46( 1/2): 4 1 -6 1 . Kahnemuy ipour, A . (2003). Syntactic categories and Persian stress. Natural Language and Linguistic Theo ry 21 (2): 333-379. Kaisse, E. (1 985). Connected Speech. San Diego: Academic Press. Kayne, R. S. ( 1 994). The Antisymmetry ofSyntax. Cambridge: MIT Press. Koopman, H. and D. Sportiche. ( 1 99 1 ). The position of subjects. Lingua 85, 2 1 1-258. Lakoff, G. ( 1 968). Pronouns and Reference. Bloomington: Indiana University Linguistics Club. Lakoff, G. ( 1 972). The global nature of the nuclear stress rule. Language 48(2), 285-303. Legate. J. A. (2003). Some interface properties of the phase. Linguistic InqUiry 34(3): 506-5 1 6. Mahajan, A. ( 1 990). The AlA-bar Distinction and Movement Theory. Doctoral dissertation, MIT. Manzini, M. R. ( 1 995). Adjuncts and the theory of phrase structure. Ms., University of Florence. McGinnis, M. (200 1 ). Phases and the syntax of applicatives. In M. Kim and U. Strauss
(eds.), Proceedings o/NELS 3 l , Amherst: GLSA.
Nespor, M. ( 1 999). Stress domains. In H. van der Hulst (ed.), Word Prosodies, EUROTYP, Theme Group 9, Berlin: Mouton, 1 1 7-160. Nespor, M. and I. Vogel. ( 1 982). Prosodic domains of extemaI sandhi rules. In H. van der Hulst and N. Smith (eds.), The Structure o/Phonological Representations. Part I, Foris: Dordrecht, 225-255. Nespor, M. and I. VogeL ( 1 986). Prosodic Phonology. Dordrecht: Foris. Nilsen, 0. (2000). The Syntax ofCircumstantial Adverbials. Oslu: Novus. Odden, D. (1 987). Kimatuumbi phrasal phonology. Phonology Yearbook 4, 1 3-26. Odden, D. ( 1 990). Syntax, lexical rules and postlexical rules in Kamatuumbi. In S. Inkelas and D. Zec (eds.), The Phonology-Syntax Connection, Chicago: Chicago University Press, 259-277. Pesetsky. D. ( 1 995). Zero Syntax: Experiencers and Cascades. Cambridge: MIT Press. Reinhart. T. ( 1 983). Anaphora and Semantic Interpretation. London: Croom Helm. Ritter, E. and S. T. Rosen. (2001). The interpretive value of object splits. Language Sciences 23, 425-45 1 . Rizzi, L. and L. M . Savoia. ( 1992). Conditions on lui propagation in Southern Italian dialects: A locality parameter for phonosyntactic processes. In A. Benetti (ed.), Syntactic Theory and the Dialects 0/Italy, Torino: Rosenberg and Sellier, 2523 1 8. Selkirk, E. (1980a). Prosodic domains in phonology: Sanskrit revisited. In M. Aronof and M.-L. Kean (eds.), Juncture, Anma Libri, Saratoga, California, 1 07-1 29. Selkirk, E. ( 1 980b). The role of prosodic categories in English word stress. Linguistic Inquiry 1 1 , 563-605. Selkirk, E. ( 1 98 1 ). On prosodic structure and its relation to syntactic structure. In T. Fretheim (ed.), Nordic Prosody II, TAPIR, Tondheim, 1 1 1 - 1 40. Selkirk, E. ( 1 984). Phonology and Syntax: The Relation between Sound and Structure. Cambridge: MIT Press.
145
Towards a phase-based theory of sentential stress
Selkirk, E. ( 1 986). On derived domains in sentence phonology. Phonology Yearbook 3, 3 7 1 -405.
Selkirk, E. (1 995). Sentence prosody; Intonation, stress and phrasing. In J. Goldsmith
(ed.), The Handbook of Phonological Theory. Cambridge: Blackwel1. 550-569. Svenonius, P. (2002). Case is un interpretable aspect. Proceedings of Perspectives on
Aspect, Utrecht. Stroik,
T. ( 1 990). Adverbs as V-sisters. Linguistic Inquiry 2 1 , 654-66 1 .
Ten ny, C. (1994). Aspectua l Roles and the Syntax-Semantics Inteiface. Oordrecht: Travis,
L.
Travis,
L.
Wagner,
Kluwer. ( 1 988). The syntax of adverbs. McGill Working Papers in Linguistics : Special issue on comparative Germanic syntax, 280·3 10. ( 1 992). Inner aspect and the structure of vP. Cahier Linguistique de UQAM 1 . 1 32- 1 46.
M. (2003). Linear order and phonological domains. Paper presented at the
Phases and the EPP workshop, MIT. Zubizarreta, M. L. (1998). Prosody, Focus, and
Monograph
33, MIT Press.
Department of Linguistics University of Toronto Robarts Library, 6th floor 1 30 St. George Street Toronto, ON, M5S 3HI CANADA
[email protected] www.chass.utoronto.cal-akahnemu
1 46
Word Order.
Linguistic Inquiry
Phases and Cyclic Agreement Julie Anne Legate Yale University
This paper proposes a cyclic agreement operation that establishes inler phasal agreement relationships.
1. Introduction The starting point of this paper is the demonstration in Legate (2003a,b) that all verb phrases are phases (in the sense of Chomsky 1 999, 2000, 200 I), rather than only transitive verb phrases as is standardly claimed. In that work, I formulate three syntactic tests supporting the notion of a vP phase, from binding reconstruction effects, quantifier raising, and parasitic gap licensing, as well as a phonological test based on the Nuclear Stress Rule; I further demonstrate that these tests also identify passive, unaccusative, and raising verb phrases as phases. Indeed, although the standard claim has been that, among the verb phrases, only transitive vPs are phases, a number of previous analyses have required phases smaller than vP: Bobaljik & Wunnbrand's (2003) analysis of Italmen long distance agreement, both Holmberg's (2000) and Svenonius' (2000) analysis of Scandinavian impersonal passives, and McGinnis' (2003) analysis of symmetric applicative constructions. Here I examine the consequences of passive, unaccusative, and raising vP phases for inter-phasal agreement relations, proposing that this is ·achieved through Cyclic Agreement. Section 2 presents the problem and the cyClic agreement solution. Section 3 outlines empirical evidence for cyclic agreement Section 4 considers additional consequences of the proposal. Section 5 concludes.
2. Cyclic Agreement If all verb phrases are phases, rather than only transitive verb phrases, we must confront the issue of how an agreement relationship is establisbed between a licensing head and a DP in a prior pbase. Consider for example the agreement relationship between finite T and a DP that remains within the verb phrase: (l)
There arrive ten trains into this station every day.
This agreement relationship is problematic in that the DP ten trains is within the
MrF Working Papers in Linguistics 49, 147-156 Perspectives on Phases © 2005 Julie Anne Legate
Phases and Cyclic Agreement
Vp l phase, and thus will have been spelled out before T is merged, apparently rendering the agreement relationship between T and the DP impossible. Sentences like (1) may receive a technical solution if we adopt the proposal in Chomsky (200 1) that spellout of the vP phase is triggered by the insertion of the subsequent phase-head, i.e. the matrix C. Therefore, at t h e point at which T probes, material within the vP pbase is still accessible and T may in fact enter an into agreement relationsbip with ten trains unproblematically. However, this technical solution does not carry over to examples with a further level of embedding: (2)
There seem to have arrived ten trains into this station today
.
In (2), at least two pbases intervene between ten trains and the matrix T: the vP associated with the verb arrive, and the vP associated with the verb seem . Therefore, the agreement relationship between T and ten trains should indeed be impossible. Here I would like to develop a solution to this problem that I fIrst proposed in Legate (2003a). The idea is that agreement applies in a cyc li c fashion, through the intermediary of every intervening phase-defining head? Thus, in (2), ten trains agrees with the v associated with arrive, which in tum agrees with the v associated with seem, which [mally agrees with the finite T.3 Let us consider this cyclic agreement in more detail, by first stepping back and looking at cases involving movement rather than cyclic agreement. (3)
Ten trains seem to have arrived into this station today.
Movement to a phase edge is standardly discussed as being triggered by an EPP feature added to ensure convergence. However, an EPP feature alone is insufficient; additional features are required to identify the correct element to be mQved. Consider (4):
(4)
Who did you give the book to?
In (4), a simple EPP feature would attract the closest phrase, either the VP headed by give, or the DP the book, b oth of which are closer to v than who. Yet who rather than any other phrase must be moved to the phase edge for it to be visible for subsequent movement to the specifier of CPo Therefore, the EPP feature of v must co-occur with additional features to identify the correct goal. My proposal is thus that these additional features may occur '1 use vP here on the assumption that passive. unaccusative. and raising verb phrases are selected for by a (defective) v. Nothing hinges on this assumption. "In Legate (2003a) I refered to this as Indirect Agreement. Here I will refer to it rather as Cyclic
Agreement.
3Assumedly, the embedded nonfinite T also agrees with the v associated with arrive, so the derivation mOre precisely would involve agreement of the v associated with arrive with the embedded T. agreement of the embedded T with the v associated with seems, and then agreement of the v associated with seems with the matrix finite T. This extIa step of agreement is not direclty relevant to the current discussion.
1 48
Julie Anne Legate
independently of the EPP feature. Consider in this light the derivation of ( 1 ). The v associated with arrive has unvalued phi features but no EPP feature. The unvalued phi-features on v agree with len tra ins. Adopting the conception of agreement as the establishment of a hyperlink to a shared set of feature values, akin to the un ification operation of GPSGIHPSG, (see Frampton & Gutmann 2000, Frampton, Gutmann, Legate, & Yang 2000), the v thus bears both the phi features and the unvalued case feature of ten trains. Subsequently, T is merged
with unvalued phi features, and the ability to l icense nominative case. T enters an agree relationship with v; as a result of this operation, v, T, and ten tra ins bear the phi features orginally associated with ten trains and the nominative case feature originally associated with T. A question arises as to the morphological realization of the nominative case feature on ten trains (which of course is not overtly visible in English, but can be in languages with richer case morphology). Since ten trains is spelled out before the agree relationship with T is established, how does the morphology succeed in realizing the correct case morphology? A number of possibilities arise. One is a return to a prior conception of features in which features may be 4 uninterpretable, rather than unvalued. Thus, the DP ten trains would enter the derivation with a valued but uninterpretable case feature; if this value is the same as that assigned by the case assigner (here nominative), the derivation converges. If however the wrong feature value is inserted, when the v phase head agrees with T, a feature mismatch will occur and the derivation will be cancelled. A second possibility that appears promising is that the morphology has access to prior phases. When the (lowest) vP phase in (1) and (2) is spelled out, the morphology receives a DP, ten trains, without a case feature. At this point it may plausibly insert the fonn of the DP bearing the morphological default case,
or realize the DP without case morphology. On a subsequent phase, when the finite T that licenses nominative case on ten trains is spelled out, this DP now bears nominative case features. Recall that agreement is conceived as a hyperlink to a set of feature values. Thus, the feature values of ten trains is updated as a result of the syntactic agreement relationship between v and finite
T at a later phase, because this agreement relationship alters the feature values that ten trains links to. The morphology may then repair the material on the previous phase, giving ten trains nominative case morphology as required. S Allowing the morphology access to prior phases seems independently necessary , given cases in which a single morphological word appears to contain morphemes from more than one phase (see Henderson 2003 for recent relevant discussion regarding verbs in Swahili). Cyclic agreement thus solves the apparent problem with unaccusative
4See Legate (2002) for an argument that in such a system in fact all features that are relevant in the syntax are un interpretable at LF. This not only resolves certain technical difficulties in the syntax, and seems more accurate semantically, but also eliminates the perceived problem that [:I: interpretable] is a feature of a feature and thus undesireablc. If all morpho-syntactic features are by definition uninterpretable at LF, this problem vanishes.
50n a similar point, see Legate (2003b) and Adger (2003) for analyses that require the phonology to access and modifY prior phases for the Nuclear Stress Rule in English and Sconish Gaelic
respectively. This seems unavoidable given the existence of prosodic domains larger than
the phase.
149
Phases and Cyclic Agreement
vP phases like in ( 1 ) and (2) by indirectly establishing the agreement relationship between finite T and a DP in a previous phase. However, the cyclic agreement operation has broader application than case and agreement. It has clear applications for example in A'-constructions without movement (e,g, wh in-situ), If the cyc lic agreement proposal is on the right track, we would like to see cases involv ing overt morphological relatization of this agreement on intervening ph ase-defining heads. I turn to this issue in the fo llow ing section.
3. Morpbological Agreement
If indeed long-distance agreement is achieved through cyclic agreement mediating between a licensing head and a DP in a prior phase, we expect this agreement relationship to be potentially morphologically real ized. Here I mention a couple of examples from the literature which seem amenable to such an analysis.
3.1 Celtic Relativization Recent analyses of certain Celtic relativization patterns have converged on the idea that they do not involve movement (see for example Adger & Ram chand 200 1 , Rouveret 2003), However, th ey do appear to involve cyclic agreement. Rouveret (2003) independently proposes a cyclic agreement operation for relativization strategies in Irish and Welsh that is similar to the present analysis. 6 Adger & Ramchand (200 1 ) demonstrate that a Scottish Breton relativization strategy does not involve movement. However they also show that in long distance relativization, each com,P lementizer must be morphologically realized as the relative complementizer a: (5)
An duine a thuirt e a/*gun bhuaileas e the man C.REL said he C.REU*that strike he 'The man that he said he will hit'
Furthermore, they note (ftn 2) that the verbs also appear in a special relative fonn: 8
(6)
an duine a bhuaileas e the man C.REL hit.REL he ' the man that he hit'
6An interesting aspect of Rouveret's work is the suggestion that strong island cffccts may be
accounted for through the agreement relationship, rather than movement of an abstract cl ement. On the assumption that island-inducing heads cannot participate in cyclic agreement, the relationship between an
in situ element (in this case a resumptive pronoun) and an agreeing head (the relative
complementizer) would not be established. resulting in ungrammaticality.
'Data from Adger & Ramchand (2001 :9). IIAlthough they state that this is
true of all verbs in clauses with the complementizer a. in (6) they 'said'.
gloss the verb 'hit' as relative, but not the verb
1 50
J ulie Anne Legate Thus, Scottish Gaelic re latives overtly morphologically mark both the cyclic agreement relationship with both the phase-defining heads C and v, on the complementizers and on the verb respectively.
3.2 Passamaquoddy A second example of morphologically realized cyclic agreement comes from Bruening's (200 1 ) description of focus in Passamaquoddy. Here we'll consider
examples involving associ atio n with the tehpu 'only'. Bruening demonstrates 9 that focus may eith er involve movement or in situ association with lehpu ' on ly, .
(7)
a.
b.
Nibtol
tebpu kisapem-ac-il
that.OBV only rely.on-3cONJ-PARTOBV ' He's the only one she can rely on.' Mali tebpu kesi- iyw-ac- i l Piyel-ol Mary only IcJike-have-3CONJ-PARTOBV Piyel-oB v ma=te=apc wen-it
NEG=EMPH=again someone-OBV 'Mary only likes PIYEL, no one else.'
The in situ focus construction is in terestin g for our purposes in its agreement properties. Bruening demonstrates that it involves agreement with the focused DP (when this DP is 3rd person) on every verb between the DP and tehpu. In the fo l lowing examples the agreement has been bolded.
(8)
a.
Tehpu kisi-wicuhkem-uk-il Piyel kespah l-ac- il Only PERF-help- I CONJ.PARTOBV Piyel lc.wash-3CONJ-PARTOBV '- temis-oL 3-dog-OBv
'I only helped P iyel wash HIS
b.
DOG.'
Tehpu kesiciy-uk-il wisukiluwo hehtw-ac-il
only
IC.know.TA-I CONJ·PARTOBV make .angry-3CONJ-PARTOBV
Mal iw-ol. Mary-oBv 'I only know that he made
MARY mad.'
This agreement is plausibly the morphological realization of a cyclic agreement 10 relationship between tehpu and the focused DP .
9All Passamaquoddy data from Bruening (200 1 ). pages 222-228.
'OBruening analyses in situ focus as involving covert movement. On the analysis of covert movement as movement at
LF (rather than overt movement with lower copy pronunciation), this is potentially
problematic in having an operation at LF affecting the morphology at PF.
On the cyclic agreement
analysis this issue does not arise. since cyclic agreement occurs in the overt syntax. Bruening uses
151
Phases and Cyclic Agreement
3.3 Blackfoot A third example of morphologically realized cyclic agreement comes from Blackfoot. 1 1 In Blackfoot, a topical DP in an embedded clause may trigger 1 agreement on higher verbs. This is illustrated in the following examples. 2
(9)
kits-iksstakk-a oma n-oxko-wa m-axk-itap-aapiksistaxsi 20BJ-want-3SUBJ my l -son-3 3-might-toward-throw
a.
kiistoyi omi pokon-i you DET ball-4 ' My son wants to throw the ball tolat you'
b.
nits-iksstata-wa n-axk-ssksinoa-xsi m-anist-sskonata'psspi l -want-3 I -might-know-3 3-manner-strong ' I want to know how strong he is'
Furthermore, in Blackfoot there is good reason to believe that cyclic agreement rather than covert movement is implicated. The agreement may be triggered by one DP in a conjoined structure: 13 nits-iksstata-wa n-oxko-wa ki niistowa and I l-want-3 I-son-3
( 1 0)
n-axk-a'po'takss-innaani I-might-work- ! PL 'I want my son and myself to work.' If movement were involved, ( 1 0) would be in violation of the coordinate structure constraint. On an agreement analysis no such difficulty arises.
3.4 KashmirilHindi-Urdu The fmal construction considered is found in Kasbmiri and Hindi-Urdu. This also involves agreement between an embedded DP and higher verbs. Examples from Kashmiri (1 1 a) and Hindi-Urdu (l Ib) follow.1 4 island restrictions to support the covert movement analysis, but see footnote 6. l i The Blackfoot data come from Frantz (1 978). as cited by Polinsky (2003). 11
Although the cyclic agreement is only optionally morphologicaly realized on intermediate clauses
(Frantz 1 978: 1 03-5. cited in Polinsky (2003:288). 13polinsky analyses the Blackfoot construction as involving a null R-cxpression in the matrix clause. which she assumes will not trigger a Condition C violation with the coindexcd DP in the lower clause. Unfortunately, she docs not demonstrate that Condition C effects between two R-cxprcssions arc absent in Blackfoot. Furthermore. the cyclic agreement analysis does not require positing a null R-expression. I"'Data from
Bhatt
(2003). Bhatt shows that when the long-distance agreement is present. the
embedded DP that triggers the agreement may clause (indicating the construction need not
1 52
scope either in the matrix clause or in the embedded be associated with covert movement). Without the
Julie Anne Legate
( 1 1)
a.
Raam-an che bameeSI yatshlmut [p anlnis necivis kbAAirl Ram-ERG be.PREs.F always wanted.FPL sclf.DAT son.DAT for
koori vuchini] girls
see.INF.FPL
'Ram has always wanted to see girls for his son b.
.'
Shahrukh-ne [tehnii kaaT-nii] chaah-ii thii. Shabrukh-ERG branch.F cut.INF.F want-PFV.F be.PST.F 'Shahrkh had wanted to cut the branch.'
Unlike Blackfoot. however, this construction is only al lowed when the embedded clause is nonfinite, and may involve restructuring. On the' present analysis, this indicates that although the phase-defining head v may bear the features that trigger this agreement in the syntax, the phase-defming head C may not. Thus, cyclic agreement is stopped at the CP phase in Kashmiri and Hindi Urdu.
3.5 I mpoverished Agreement In the prev ious sections we have seen diverse phenomena which seem to involve morphological realization of cyclic agreement relationships. This lends support to the cyclic agreement proposal for the simpler cases of inter-phasal agreement between fmite T and an in situ object of an unaccusative verb in English. However, long-distance agreement is often found to be more impoverished than apparent specifier-head agreement. This is potentially problematic under the analysis of movement in Chomsky ( 1 999, 2000, 200 1 ). In this theory. agreement (i.e. the Agree operation) unifonnly appli es between the head and the element in situ. The distinction between a movement versus in situ derivation lies in the presence versus absence of an EPP feature on the head. Thus there appears to be no explanation for a distinction in agreement p ossiblities between a movement versus in situ deriv ation .
is
The present proposal provides a possible approach to this contrast. This because on the present proposal, the agreement mechanisms used for
movement versus in situ derivations differ. Consider an example: ( 1 2)
a. b.
There arrive t en unicorns into the station today. Ten trains arrive into the station today.
In ( 1 2a), the OP ten unicorns remains in situ in the VP arrive; the v associated with arrive agrees with the OP ten unicorns, and the finite T agrees with v. In ( 1 2b). on the other hand, given the Phase Impenetrability Condition, ten trains must move to the edge of the vP phase in order to undergo subsequent movement to the specifier of TP. Therefore , T agrees with the OP len trains itself rather than v. We thus have a potential restatement of the generalization: agreement between a bead and an XP is richer than between a head and another agreement. the embedded
DP may only scope in the embedded clause.
153
Phases and Cyclic Agreement head that the XP has agreed with. In that the movement derivation involves a more direct syntactic relationship between the head and agreeing XP, it is
perhaps plausible for this relationship to trigger richer agreement morphology.
4. N u merations Finally, I would like to consider the infamous contrast in ( 1 3) that served as one motivation for phases and that is a potential problem for the present analysis:
( 1 3)
a. b.
*There seems a man to be in the room. There seems to be a man in the room.
On the standard assumption that nonfmite T does possess an EPP feature (but see Boskovic 2002), this contrast has been puzzling. It is unclear why the EPP feature of the embedded T could not be satisfied by raising of a man instead of insertion of there. As observed by Alec Marantz, we cannot simply claim that expletives must be merged as early in the derivation as possible due to the lack of a contrast between (1 4a) versus ( 1 4b) and ( 1 4c). ( 1 4)
a. b. c.
There was circulated a rumour that someone was in the room. There was circulated a rumour that there was someone in the room. A rumour was circulated that there was somone in the room.
The crucial distinction between the case in (1 3) and the case in ( 14) is that ( 1 3) involves a nonfinite T, whereas ( 1 4) involves a finite T. The now-standard analysis runs as follows. A numeration is chosen for each phase. If at any point during the derivation, the EPP may be satisfied either by merge of there or by movement of a DP, merge is prefered as more II economical " (hence the slogan "merge over move"). On Chomsky's assumption that only transitive vPs and CPs are phases, ( 1 3) is derived using a single numeration. Since this numeration contains there, it must be inserted to satisfy the EPP feature of the embedded TP. ( 1 4) on the other hand, consists of two pbases, and thus two numerations- one for the embedded CP, and the other for the matrix CPo In ( 1 4a), the numeration for the embedded CP does not contain there (althougb the numeration for the matrix CP does), and so merge over move is not implic;:ated. In ( l 4b) and ( l 4c), on the other hand, the numeration for the embedded CP does contain there, and so there must be inserted in the specifier of the embedded TP. On this view, the contrast between ( 1 3) and ( 1 4) motivate the selection of a distinct numerations for CP phases, but not transitive vP phases. Indeed, evidence for distinct numerations for transitive vP phases bas not been forthcoming. Thus, the merge over move analysis could be carried over straightforwardly to the present system, on the assumption that only CP phases trigger the formation of a distinct numeration.
IS
ISIt should be admitted, however, the merge-over-move analysis docs not seem satisfactory, and an alternative solution is needed. Therefore, the need to restrict distinct numeration selection to CP phases on the cunent system to maintain the merge-over. move analysis may not be a relevant
1 54
Julie Anne Legate
5. Conclusions This paper has proposed a mechanism of cyclic agreement mediating between a
licensing head and an element in a prior phase. The primary motivation was the agreement relationship between a fmite T and an in situ DP, although examples of apparent morphological realization of cyclic agreement were also provided. Further research is needed to determine the full range of constructions that involve cyclic agreement mediating between phases.
References
Adger. David. (2003). Stress and Phasal Syntax. Paper presented at GLOW, Lund. Adger David & Gillian Ramchand. (200 1 ). Phases and InterpretabiHty. In K. Megerdoomian & L.A. Barel (eds ), WCCFL 20 Proceedings. SomerviUe. MA: CascadiHa Press, 1 -14. Bhatt, R ajes h. (2003). Long Distance Agreement in Hindi-Urdu. Ms., Univ ersity of Texas. Bobaljik, Jonathan & Susi Wurmbrand. (2003). When a Phase is Not a Phase. Paper presented at GLOW. Boskovic, Zeljko. (2002). A-Movement and the EPP. Syntax 5(3). Bruening, Benjamin. (200 1). Syntax at the Edge: Cross-Clausal Phenomena and The Syntax of Passamaquoddy. Doctoral dissertation, MIT. Distributed by MITWPL. Chomsky, Noam. ( 1 999). Derivation by Phase. MIT Occasional Papers in Linguistics 1 8. Cambridge, MA: MITWPL. Chomsky, Noam. (2000). Minimalist Inquiries: The Framework. In R. Martin, D. Michaels, and J. Uriagereka (eds.), Step by Step: Essays on minimalist syntax in honor ofHoward Lasnik. Cambridge. MA : M ITPress Chomsky. Noam. (200 1). Beyond Explanatory Adequacy. MIT Occasional Papers in Linguistics 20. Cambridge. MA: MITWPL. Frampton, John and Sam Gutmann. (2000). Agreement is Feature Sharing. Ms., Northeastern University. Frampton, John. Sam Gutmann, Julie Anne Legate, & Charles Yang. (2000). Remarks on "Derviation by Phrase": Feature Valuation, Agreement. and Intervention. Ms., MIT, Northeastern University, and Yale University. Frantz. Donald G. ( 1 978). Copying from complements in Blackfoot. In Eung-Do Cook and Johnathan Kaye (cds.), Linguistic Studies of Native Canada. Vancouver: University of British Columbia Press: 89-1 10. Henderson, Brent. (2003). PF Evidence for Distributed Morphology. Poster presented at NELS 34. SUNY. Holmberg. Anders. (2000). Expletives and agreement in Scandinavian passives. Working Papers in Scandinavian Syntax 65: 35-64. Legate. Julie Anne. (2002). Phases in "Beyond Explanatory Adequacy." Ms., MIT. Legate, Julie Anne. (2003a). Identifying Phases. Paper presented at the Workshop on EPP and Phases, MIT. Legate, Julie Anne. (2003b). Some Interface Properties of the Phase. LinguistiC Inquiry 34(3):506-5 1 5 . McGinnis, Martha. (2003). Variation i n the phase structure of applicatives. In lohan .
.
.
consideration.
1 55
Phases and Cyclic Agreement Rooryck & Pierre Pica (eds.), Linguistic Variation Yearbook. Polinsky, Maria. (2003 . ) Non-Canonical Agreement is Canonical. Transactions of the Philological Society Vol 1 0 1 (2):279-3 1 2 . Rouvcret, Alain. (2003). Relativization Asymmetries i n Celtic. Paper presented a t USC. Svenonius, Peter. (2000). Impersonal Passives: A Phase-based Analysis. In Holmer,
Svantesson & Viberg (eds.). Proceedings of the 18th Scandinavian Conference ojLinguistics. Travaux de I'lnstitut de Linguistique de Lund.
1 56
Going through a phase* Ora Matushansky CNRSlUniversile Paris-8
We will show that an attempt to apply phasebood diagnostics to DPs yields contradictory results. From the point of vicw of the interfaces, LF diagnostics suggest that DPs are not phases, while PF diagnostics argue for the opposite. Other tests. such as an escape hatch at the edge or compulational complexity, arc equally inconclusive. The conclusion of the paper is that eithcr interface independence is not a diagnostic for phasehood or phases do not exhibit independence at both interfaces at once, which casts the whole notion
1.
into doubt.
Introduction
The central proposal of the Minimalist Theory as stated by Chomsky ( 1 995, 2000, and 200 1 a, b) is that language is an optimal solution of the computational system (grammar) to the constraints imposed by the two interfaces: with the articulatory-perceptual system (PF) and the conceptual-intentional system (LF). If this assumption is correct, constraints on syntactic operations follow from the fact that sentences are processed, produced and stored. The surmise that there are limitations on the quantity of information that can be kept in the workspace/short-term memory gives rise to the proposal that such computational complexity can limit the scope of syntactic ope�ations. Chomsky (2000, 200 I) forwards the idea of Multiple Spell-Out, which suggests that syntactic computation proceeds in phases:.
I.
Phase (Chomsky 2000, 2001): A phase is a unit of syntactic computation that can be sent to Spell-Out.
As is clear from the definition above, phase theory is at least partially motivated by constraints on processing. Phases are defined by non-exhaustive
*
Many thanks to Eddy Ruys, Jon
Nissenbaum and
Alain Rouverct for useful comments, to Tim
Stowell for the joke that led to the title, to Martha McGinnis or Norvin Richards for the review, and
to the audiences at Seminaire tie I 'UMR 7023 (October 2002), MIT lAP Workshop on EPP anti Phases (January 2003), and Trans Seminar on Synlevc anti Semanlics al UiL ors (February 2003). where various versions of this paper were presented
MIT Working Papers in Linguistics 49,
Perspectives on Phases
157-18/ © 2005 Ora Matushansky
Going through a phase
enumeration: vPs (ofa particular kind) and CPs are phases, VPs and TPs are not. Nissenbaum ( 1 998) argues that what is spel1ed out and is no longer accessible to the syntactic component is not the phase node itself (vP, CP), but the complement of its head (TP and VP, respectively). The reduction in the computational burden is assumed to follow from the (in)accessibility of the content of a phase from outside this phase (strict cycle):
II.
Phase Impenetrability Condition (PIC) Only the edge of a phase can be accessed from outside this phase.
The general idea of this proposal is to restrict syntactic computation to smal1er chunks. It is unclear, however, how this goal is achieved by the proposed analysis.
1.1.
Modifiers
Computational complexity is not sizably decreased by proceeding in vPs or CPs, because there is no theoretical upper limit on the number of modifiers of such vPs or CPs. If the reduction of complexity i s with respect to the number of items kept in memory (7±2, Miller 1 956), modifiers pose a serious problem. 1 Likewise, if we are primarily concerned with the size of the search space, it is easy to show that modifiers can be targeted by the syntactic operations o f movement (e.g. i n Topicalization, Focalization and question-formation) and
therefore must be viewed as eligible targets for the search. Little more can be
added to the topic, since the relevant notion of computational complexity is left tmdefined.
1.2.
Exploded IP and multiple arguments
The analysis proposed above does not consider verbal arguments beyond direct objects (which presumably appear alongside additional heads), or the fine structure of the IP (including not only TP and NegP, but also various aspectual projections) or of the CP (which may contain a TopP and a FocP. at least). ,This . .
omission is not negligible. If we restrict our consideration of the clausal structure to VPs, vPs, TPs and CPs, the actual reduction in computational complexity (from four projections to two) is virtually non-existent. If we believe that both the CP-Iayer and the IP-Iayer contain several maximal projections (whose number reaches I An additional problem with such a simplistic approach is that it might not clearly define some constirucnts (as opposed to others) as phases, since the moment of the Spell.Out would depend on the number of items and might vary depending on the nu mber of auxiliaries or the type of the verb (assuming that additional arguments increase the number of heads inside vP). This is clearly not why phase theory was first introduced.
1 58
Ora Matushansky
forty, in some estimates), then it is unclear to what extent isolating the vP phase (two to four maximal projections) reduces the computational burden.
1 .3.
Other phases
The inventory of phases isn't intended to be limited to the two XPs enumerated by Chomsky, CP and vP . Therefore, for any XP we should be able to tell whether it is a phase.
In the remainder of this paper we will consider the question of whether there is or there is not a phase boundary within the extended NP projection (xNP).2 In order to do so, we will use the phasehood diagnostics proposed by Chomsky (2000) and Legate ( 1 998, 2003). Phasehood diagnostics are based on the assumption that phases are units of computation sent to Spell-Out as independent chunks of structure. As such, they should exhibit independence at interfaces. Phasehood diagnostics test for such independence. We will show that phasehood diagnostics contradict each other when used to detennine whether DPs (or some DP-intemal XPs) are
phases, casting severe doubts either on the pertinence of both interfaces to Spell Out, or on the notion of the phase itself. We will begin with the discussion of computational complexity issues, then tum to phonological and morphological behavior of phases, then address such syntactic operations as Case assignment to phases, extraction out of phases, and parasitic gap licensing. The issue of
2.
Computational complexity (YES)
If we take the computational complexity argument to mean that there must be some limit on the number of maximal projections in the workspace (for example the famous
7±2,
discussed by Miller 1 956), then DPs must be phases or contain
phases. Since DPs can be iterated and thus arbitrarily long, at some point they must exhaust the memory resources available:
(1)
a. b.
my mother's sister's mother's sister's . . . son a friend of a friend of a friend . . .
I n the examples that w e have chosen, the DP that i s iterated is an argument of a noun rather than a modifier. If the notion of computational complexity is defined with respect to the number of heads, maximal projections or lexical items in the workspace, then DPs must be phases.
2 We will
use the xNP notation to emphasize our agnosticism as to whether the DP layer is projected in all the extended NPs that we consider.
1 59
Going through a phase
3.
Morpho-phonology (YES)
The behavior of DPs with respect to morphology and phrasal phonology suggests that they have independence at this interface. The syntactic PF diagnostics for phasehood (Legate 1 998) test for phonological independence from the point of view of syntax and phrasal phonology.
III.
PF tests for phasehood phases can be isolated phases can be moved and targeted by movement like operations phases are assigned phrasal stress through the
Nuclear Stress Rule We will show that nearly all PF diagnostics of phases suggested by Legate ( 1 998) apply to DPs, even those that don't apply to CP or vP phases. Exceptions can be shown to be ruled out by independent factors. We will also consider the theoretical status of these diagnostics. We will not address the application of the Nuclear Stress Rule to DPs (see Legate 1 998, 2003 for the discussion of how it influences movement within and out of phases)� since there are no clear cases of movement from one DP internal position to another.
3.1.
Isolation
If syntactic derivation proceeds in phases, then a minimal convergent derivation must he a phase. CPs are clearly phases. To test PF- and LF-convergence for smaller constituents, we will follow Legate ( 1 998) and Use the so-called "Mad Magazine" sentences:
(2)
a. b.
(Can you teach lexical semantics?) - Me teach lexical semantics? (How about arriving early for once?) - Me arrive early?
DPs (any semantic type thereof) can clearly be isolated in the same contexts and under the same conditions as vPs:
(3)
a. h. c.
(Hillary was elected president) - President? Hillary? A woman?
(e, t) (e) «e, t) , t)
The problem with this test is that it equally well applies to a non maximal projection of't (assumed not to be a phase);
(4)
1 60
Alice didn't leave. - Didn't leave? What do you mean, didn't leave?
T'
Ora Matushansky Likewise. it is unclear what this test really shows for VPs, because it is hard to say whether what is isolated in a " Mad Magazine" sentence in (5) is a VP rather than a vP with a PRO subject or the trace of the subject raised to [Spec, TPJ: (5)
(Can you teach lexical semantics?) - Teach lexical semantics? Why? VP?
It is not impossible that only a few XPs cannot be isolated, TP being one of them. 3.2.
Movement-like operatio ns
The aim of this diagnostic is to show that phases behave as a unit in movement like structures that may not involve purely syntactic movement. My intuition is that the basis ofthis diagnostic is that whether moved or not, the ability to appear at the clausal periphery is regarded as shedding light on the isolability of the peripheral constituent. If a particular construction renders an XP peripheral by a movement operation, then the diagnostic reduces to the ability of this XP to be moved, while if movement is not involved, the test would verify whether the XP can be isolated. In other words. we are dealing with versions of isolability (section 3 . 1 ) and movement (section 3.3). which are given a special status in virtue of insufficient information on the nature of the operation. I will show that this collection of tests does not properly distinguish vPs and CPs (phases) from VPs and TPs (non-phases). However, for DPs, most of them yield a positive result. 3.2.1.
Extraposition
The ability to undergo extraposition distinguishes CPs from TPs. However, neither vPs nor VPs can be extraposed (for the simple reason that they cannot be easily found in the subject position), so the fact that DPs cannot either is not surprising: (6)
a. It surprised Ron [cp that Hermione was interested in someone else] . b. * It surprised Ron [Hermione (to) be interested in someone else]. c. * It surprised Ron [DP Hermione' s interest in someone elseJ
This test is inconclusive: it may be that DPs can undergo extraposition, since there is an independent syntactic reason for the ungrammaticality of (6c). Consider how Case-assignment functions in extraposition: the fact that the expletive it is assigned Case by the matrix verb, means that the DP associate of it is not (because Case cannot be assigned twice). The fact that CPs do not need Case makes them possible associates of it, while DPs, which require Case. are ruled out. 3.2.2.
Clefting
Like extraposition, c1efting also doesn't target vPs, but does, CPs and DPs: 161
r Going through a phase
(7)
a. b. c.
*
It's [ep that Desdemona was faithful] that Othello doubted. It's [vp doubt that Desdemona was faithful] that Othello did. It's [oP Desdemona's faithfulness1 that Othello doubted.
The fact that c1efting doesn't apply to all (uncontroversial) phases casts some doubts as to its utility as a diagnostic for phasehood. 3.2.3.
Pseudo-clefling
Pseudo-clefting clearly can target DPs, CPs and vPs:
(8)
What King Lear said was [cp that Cordelia was no longer his favorite daughter1. What Goneril did was [vp blind Gloster]. What Regan listened to was [oP Goneril's suggestions] .
a.
b.
c.
The problem here is that pseudo-clefts also target what doesn't seem to be a phase, though this ability seems to depend on what main verb is chosen: (9)
a. b.
*
What Goneri l did was hp to blind Gloster]. What Goneril seemed was [TP to fear King Lear].
The ungrammaticality of (9b) su ggests that either the clefted constituent in (9a) is not a TP or the constraints on pseudo-clefting are not (just) about phase s. But if the clefted constituent in (9a) is not a TP, what is it? Furthermore, it can be shown that the choice of the matrix verb can influence movement of CPs, suggesting that the contrast in (9) is due to the same factor: 3 ( 1 0)
a. b.
[cp That Cordelia was no longer his favorite daughter] King Lear certainly said t i. * [cp That Cordelia was no longer his favorite daughter] it certainly seemed t i.
We conclude that pseudo-clefting is not a very reliable test when it comes to distinguishing between standardly assumed phases. However, it yields positive results for DPs. 3.2.4.
Though-constructions
Since though-fronting targets predicates, CPs cannot be tested for phasehood in this way. DPs, however, can and yield a positive result: ( I I)
a.
*
[ep That she would many her lover] though Juliet said, the results were disastrous.
b.
[vp Marry her lover] though Juliet did, the results were disastrous.
3 This fact can also explain some differences in the behavior of infinitival complcmcnlS of control and raising verbs, used to claim to be CPs and TPs, respectively.
1 62
Ora Matushansky
C.
[01' Th e villain of the plaY]i though Regan is t i, I still like her the best.
It is unclear what is shown by this diagnostic. The most likely theory is that though-constructions ta rget predicates (semantic type (e, t» . Th i s would explain why CPs are not p ossibl e targets, but then the grammaticality of ( 1 1 b, c) has n othing to do with phases: OPs and vPs can be predicates, and TPs are not. 3.2.5.
(Predicate) fronting
Predicate fronting should not be applicab le to CPs, since CPs are not predicates. However, ( 1 3), whi ch is superficially similar to (12) is not ungramm ati cal:
( 1 2) a. pl aY] i b. ( 1 3)
Regan is called the villain of the play and [opthe villain of the she is t i. Goneril said she would pluck out Gloster s eyes, and [vp pl uck out h is eyes] she did. '
Juliet promi sed that she would marry Romeo, and [ep that she would marry Romeo] her parents didn't thinklknow.
It is unclear whether ( 1 2) and ( 13 ) diagnose for the same thing. On the surface, (predicate) fronting is the only PF-diagnostic that unambiguously treats CPs and vPs identically. 3.2.6.
Summary
The ability ofOPs to undergo movement-like operati ons (argued to diagnose for phasehood by Legate 2003 and this volume) is if anythin g greater than that of vPs and CPs. The apparent negative result in section 3.2. 1 can be explained by the interaction with (some version of) Case Fi lter The probl em with this class of diagnostics is that they are internally inconsistent, which makes it not at al1 c lear what they really test for. It may well be that, once independent factors, such as type, are taken into consideration, the contrasts between p hases and non-phases with respect to these tests will be accounted for. [n absence of such evidence, however, we conclude that DPs do di splay PF-isolability. ,
3.3.
Real movement
Since the complement of the head ofa phase is what is spelled out, its extraction i s impos sib l e under the assumption that Comp-to-Spec movement is not allowed. It can therefore be argued that phases (CP, vP) can move, while XP complements of heads of phases (TP. VP) cannot. For the sake of completeness it should be mentioned that OPs can be moved while no intennediate xNP projections can, which, following the l ogi c of the argument, suggests that they are phases. ,
1 63
Go in g through a phase
( 1 4)
a. b.
Alice seems [t i happy] . Which person did you see t i?
A-movement (raising)
A'-movement ( wh-mo vement)
In this respect also, DPs seem to have more PF-i ndependence than vPs or CPs. 3.4.
Concord
The morphosyntactic operation of concord (DP-internal agreement) appli es strictly within the DP-domain, suggesting that DPs have morpho-syntac tic isolab i l ity. This possible test does not extend straightforwardly to other XPs ( unless we take it to parallel the assertion that phases are agreement domains). 3.5.
Summary
PF di agnostics rather favor a phase status of DPs. DPs can be isolated and moved by a variety of syntactic operations, including the standard A- and A' movement. The intennediate conclusion seems to be that DPs are as PF-isolable a s standardly assumed phases (CPs and vPs) if no t more so. However, we have discovered multiple problems with PF-diagnostics proposed here. Conversely, from the point of view of the morpho-syntactic operation of concord (a novel test), DPs are phases . 4.
Syntax (NO)
In this section we will examine what various syntactic facts can tell us about the phasehood of DPs. The implementation of Case a ssignment as valuation of an unvalued feature necessitates DPs not to be ph ases. Attempts to discover whether DPs have escape hatches yield indeterminate results (no evidence for either conclusion) . The criterion of ql-completeness (Chomsky 2000) d isting uishes between two types of DPs : arguments �n the one hand and generalized quanti fiers and p redicates on the other, which will be shown to result in incorrect predictions. Finally, the unavailability of parasitic gaps on the edge of DPs also seems to argue against their status as phases. 4.1 .
Case
Concord means that Case-marking can be spread throughout a DP. This makes it impossible for DPs to b e phases or to contain an xNP phase, in flat contradiction to the conclusions previously reached. In order to see why it is so, we will begin with supposing that there exists a DP-internal FP (possibly OP or NumP) that is a phase. When FP is spelled out, what becomes inaccessible is the compl ement of FO, and the complement of F' is minimally an NP. Thus the head noun is not accessible from outside the FP (by PIC):
1 64
Ora Matushansky
( 1 5)
Suppose that when DPs are merged into their base position. they bear an unvalued Case feature ([uT], following Pesetsky and Torrego 200 1 , 2002). When [uT] on DP is assigned a value (in [Spec, TP], for subjects), the complement of the DP-intemal � has already been spelled out and should be inaccessible to the syntactic computation. Since NP is contained within (or is) the complement of fO. NP and modifying APs should be inaccessible for Case marking. This is obviously untrue. There are two possible ways out: I. Case marking inside the xNP is a result of a special operation applying after the Spell-Out (concord is a good candidate for a PF branch operation) 2. There is no Case valuation/assignment, only Case checking. A DP is arbitrarily assigned some Case, which is then checked against a feature of the functional head that attracts it (the previous approach). Case is a problem for any theory assuming cyclic Spell-Out, unless Spell-Out means being removed from the short-term memory/workspace rather than being sent off to the interfaces. One possible way of doing so is returning phases to the numeration (Johnson 2002). We conclude that in the case-valuation framework, DPs can't be phases or contain phases.
4.2.
Escape hatch
The introduction of phases and PIC entails the existence of escape hatches: an item cannot move out of a phase unless it {"lTst moves to the periphery. For example, an object that has not moved to [Spec, vP] will be spelled out in its base position, along with the rest of the VP. Since it will thus be rendered inaccessible to syntactic computation, it will not be able to raise any further. Do DPs have escape hatches? It has been suggested that [Spec, DP] is an escape hatch for A' movement out of xNP (Cinque 1980, Giorgi and Longobardi 1 99 1 , Stowell 1 989, Szabolcsi 1 983, 1994, Campbell 1996, Gavruseva 2000, among others). One argument for this proposal is the fact that extraction out of a DP is blocked O by a filled [Spec, DP] or by an overt D ;
( 1 6)
a. b. c.
Who did you buy a portrait of? * Who did you buy the portrait of? * Who did you buy Michelangelo's portrait of?
1 65
G oing through a phase
However, the reason for th is restriction may be semantic: specific (existence-presupp osing) DPs are islands for extraction (Fiengo and Hig ginbotham 1 98 1 , Stowell 1 989, Em; 199 1 , Diesing 1 992, Mahajan 1 992):
( 1 7)
*
Who did you buy a specific/particular/certain portrait of?
That the constraint is not syntactic is supported by the fact that mere presence of a definite article is not enough to block extraction, as it is the case in comparative superlatives (Szabolcsi 1 986, Heim 2000, Sharvit and Stateva 2002, etc.). Comparative superlatives cannot presuppose existence: ( 1 8)
a. b.
*
(Of those present) Who did you take the picture of? (Of those present) Who did you take the best picture of?
The assumption that existence presupposition blocks extraction accounts for the data in both ( 1 6) and (1 8). Suppose, however, that a fi lled D O is not relevant for our purposes Only if there is a maximal projection in [Spec, DP] is movement out of a DP blocked: the escape hatch is filled. Two objections to this hypothesis can be raised. The first one has to do with the number of specifiers that a maximal projection must have. Consider multiple extraction out of a vP: .
( 1 9)
What did Sue ask where Alice will [yp wftM where A:flee- [vp buy whef where]?
As ( 19) shows, the vP periphe ry can contain more than one candidate for extraction out of the vP phase. Therefore vP has more than one specifier. Why not DP? That the "Genitive" Spec is filled should not have any effect on the possibility of projecting others. The second objection is theoretical. The proposal that specific (existence-presupposing) DPs are islands is sufficient to explain possessed DPs as well. All possessed DPs are necessarily familiar (Barker 2000 and references cited there; see also Woisetschlaeger 1 983 for the argument that all Saxon possessives are definite) and thus have the existence presupposition (which can be shown , among other things, by their inability to appear in existentials and with have). Why have two explanations for one effect? A final argument for the claim that [Spec, DP] is an escape hatch out of DP comes from the fact, noted by Cinque ( 1 980) and Giorgi and Longobar:di ( 1 99 1 ), that in Romance, only those arguments that can be possessivized can be extracted. While this fact can be explained by the assumption that an argument needs to pass through [Spec, DP] in order to be extracted out of the DP, an alternative hypothesis would be that, since possessivization is a movement to [Spec, DP]. arguments that cannot be moved cannot be possessivized either. 4.3.
cp-completeness
Chomsky (2000) suggests that phases mu st be qJ-complete, i.e. have all their internal arguments saturated. From the semantic point of view, this corresponds to a simplex type: (e), (t), etc. Thus vPs, which in non-intensional semantics
166
Ora Matushansky
denote a truth value (type (t» are -complete by definition. The status of quantified DPs { everyo ne, at least sixteen monkeys . } is not at all clear, since they are assumed. in standard semantics, to take one-place predicates as their arguments (semantic type «e, t), t» . Finally, the worst problem is probably that of predicate DPs, which do require an argument, provided in a small clause structure: .
. .
(20)
a. b.
Rissa is [sc �a princess]. Gar considers [sc Spido an idiot).
small clause, simplified
Since a predicate is defined as something that requires an argument to give a truth value, predicate DPs cannot be cp-complete, which means that at least some DPs are not fhases {though see section 5. 1 .3 for the alternative point of view, assuming Pred as the head of the small clause (Bowers 1 993». Theoretically. this entails that argument DPs (being cp-compJete) should behave differently from predicate and quantified DPs (which are cp-incomplete) with respect to extraction out of them. Not only have no such distinction ever been attested, the prediction is nearly the exact opposite of the facts. Indeed, suppose that argument DPs (type (e), q>-complete) are phases, while generalized quantifiers {type « e, t), t» and predicates (type (e, t}), both cp incomplete, are not. If this is true, then extraction out of argument DPs must proceed through the escape hatch, while extraction out of non-argument DPs is unconstrained. As we have noted in section 4.2 , extraction out of specific (type (e» DPs is impossible (Fiengo and Higginbotham 1 98 1 , Stowell 1 989, Ent;: 1 99 1 , Diesing 1 992, Mahajan 1 992) or difficult (Tellier 1 99 1 , Davies and Dubinsky 2003 ) . This means that we cannot check whether extraction out of' them proceeds through [Spec, DP]. On the other hand, the standard argument that [Spec, DP] is an escape hatch for At-movement out of xNP (Cinque 1 980, Giorgi and Longobardi 1 99 1 , Stowell 1 989, Szabolcsi 1 983, 1 994, Campbell 1 996, Gavruseva 2000, among others) has been made for non-specific DPs, i.e. exactly for those DPs for which q>-completeness predicts no restriction. Finally. quantified non-indefinite DPs allow no extraction out of them. We conclude that the cp-completeness criterion makes a wrong prediction about extraction out of various kinds of DPs. This casts serious doubts on its validity as a phasehood diagnostic. The only possibility is that I have not analyzed q>-completeness correctly and it has nothing to do with the semantic type and the ability to take arguments.
4.4.
Parasitic gaps
Nissenbaum (2000) links parasitic gaps to edges of phases. Since edges of phases serve as escape hatches, they are obligatory landing sites for anything moving out of a phase. Therefore, parasitic gaps can be used to diagnose the edges of phases.
167
Going through a phase
The first thing we need to establish is that parasitic gaps are not limited to extended verbal projections (adjunct participles or CPs). That this is not so is shown by the fact that subject DPs, like adjuncts, can contain parasitic gaps: (2 1 )
Mary's the kind of woman j that [people who meel _PG.j ] usually end up inviting _REL-i into their homes.
However, subj�ct-intemal parasitic gaps are not licensed at the DP periphery. Nissenbaum (2000) shows that subjects containing parasitic gaps are interpreted inside the vP and that the parasitic gap inside them is licensed at the vP-periphery. Given that we have been unable to find a movement that unambiguously uses the presumed escape hatch on the DP periphery, it is hard to know what to test for. However, we can try to use the fact that parasitic gaps appear only inside islands (subject and adjunct islands). If we place a parasitic gap inside a relative clause island, can wh-movement out of a DP license a parasitic gap inside a relative clause? The answer is no: (22)
a. b.
Mary told a story about Claire i that _REL really impressed her i ·
* Whoj did Mary tell [a story about _ ilj that _REL.j really impressed _POoj?
We don't know why (22b) is bad. Maybe relative clauses are attached too low for parasitic gaps to be licensed. Maybe relative clause islands are different from adjunct and subject islands. But a priori, DPs don' t seem to license parasitic gaps. This failure of parasitic gaps to appear at DP-edges does not mean that DPs are not phases. Rather it leaves the question open. While supporting the claim that there is no escape hatch at the DP-periphery, it does not unambiguously prove it. 4.5.
Summary
The syntactic tests presented above mostly agree in their results: DPs cannot be phases. We have shown that those �f these tests that are not based on type denotation (like cp-completeness is) all yield either a negative result (Case) or no clear result (parasitic gaps, escape hatches). We will return to escape hatches in the next section, but the major conclu�ion of this one is that DPs are not phases, . in contradiction with the conclusion that we have reached in sections 2 and 3. We will now show that LF diagnostics yield the same result as syntactic diagnostics. 5.
LF diagnostics (NO)
The fact that certain maximal projections (phases) are spelled out while others are not entai ls that phases exhibit certain independence at LF. This independence has been assumed to be the ability to be interpreted in isolation.
1 68
Ora Matushansky
Obviously, simple interpretability in isolation is not enough - almost any non-affixal head, being a lexical entry, has this property. Chomsky (2000) and Legate ( 1 998) propose two slightly more complex criteria:
IV.
LF tests for phasehood phases have the status of a "proposition" QR, reconstruction and successive cyclic wh movement can target edges of phases
In this section we will show that when applied to DPs, the LF diagnostics above contradict the results obtained from the PF diagnostics (section 4): while PF diagnostics (see section 3) suggest that DPs are phases (full phonological isoiability), LF diagnostics argue against this conclusion. On the other hand, just like the syntactic tests discussed in the previous section, they argue against the hypothesis that DPs are phases. We will first show that the two diagnostics test for· the same thing: the semantic type ( t). Since DPs never have this type, the two LF diagnostics agree that DPs cannot be phases. However, there are good reasons to believe in the existence of a DP-internal (t) -type mode and therefore. a DP-internal phase (which clearly does not show any PF independence). Overall, we will show that LF diagnostics contradict PF diagnostics. 5.1.
Semantic status
Chomsky (2000) suggests that phases must have the status of a proposition. It is unclear what this means from the syntactic point of view, since no independent definition is provided. However, semantically, it would mean that phases must have the semantic type (t) (abstracting away from event, world, time, etc., variables, which is not an innocent assumption). CPs and vPs are then assumed to have the semantic type (t) (fully compatible with model-theoretic semantics), while TPs and VPs are assumed to have some other type (undefined).
5. 1.1.
Type (t)
The fact that DPs never have the semantic type (t) shows that they cannot be phases, according to this diagnostic. However, Heim and Kratzer ( 1 998:22 1 ) demonstrate that a (t)-type node (which this diagnostic would argue t o b e a phase) is present inside an extended NP projection (xNP). (23)
No student from aJany foreign country was admitted.
....3. > 3
The quantified complement of the preposition (a/any country) cannot be interpreted in-situ, because to be interpreted, quantifiers must adjoin to propositions �type ( t» . On the other hand, the meaning of (23) cannot be obtained by adjunction to the IP or to any node higher than the xNP in which the two quantifiers are contained (see Heim and Kratzer 1998 for the discussion).
169
Going through a phase Moreover. if the PP from any foreign country were moved to a position above the article, the NPI any would have .raised out of the scope of negation. Heim and Kratzer ( 1 998) propose that there is a (t)-type node ins i de the OP, to which a O P-internal generalized quantifier can adjoin. This proposal is implemented by postu lating that the xNP has a semantically vacuous and phonologically null subject, which Heim and Kratzer identify with PRO. When PRO moves. it leaves behind an (e)-type trace, which is i n terpreted as a free variable and abstracted over at the adjunction site Gust b elow the determiner). We will assume that the xNP-internal subject is in [Spec, FP], where � cou ld be nO, Numo, etc. (as long as it is not DP). After PRO -mo vement, the FP has type (t) and allows QR to it. In the tree below, FP is assumed to be NP, but nothing hinges on this assumption: (24)
OP «Col).l)
no
/
� X E Dc�
�
poss ibl e quantifier adjunction site
(e, t)
PR
•
I i
-
DP �
i aforeign country � A i . .. ...... .. ... ... .... .
.
.
_.!
.
!
I
1
__
_._
(e, t) � .J E Dc NP
�
..
�PP(e. 1}
. . _.�
_...._... . .
NP
;Q;;zt
s
0:
.. .... .... .......... ...... fr.om
... -............... .................-.
.
..
.
.
i
-_ ._............."
If the DP contains what is diagnosed as a phase (has type (t» , how does it act with respect to other phasehood diagnostics? From the point of view of PF, n o DP-internal projection can move stranding the article or appear in isolation:
(25)
a. * It's [xp best known play] that Romeo and Juliet is (the) ti. b. * What Romeo and Juliet is (the) is [xp best known play]. c. * [xp best known play] Romeo andJuliet certai nly is (the) ti.
We conclude that if being a phase means having the type (t), DPs cannot be phases. Conversely. there is a DP-internal node that is a phase for this diagnostic and which has no PF�isolabiljty. -
5. 1 . 2.
A
simplex semantic type
Perhaps the primary LF-diagnostic for phases (interpretability in isolation) should be understood in the fo llowing sense: only constituents having simplex semantic types (e). (t) and (d ) (and possibly some others, depending on the semantic theory assumed) can be phases. This assumption is not unreasonable, since constituents of simplex semantic types do not take any arguments and thus 1 70
Ora Matushansky
can b e interpreted in isolation. Among those, only the constituents that exhibit PF-independence are phases. This approach still does not give us a reliable result given that DPs can have a variety of semantic types ( i ndividual (e), predicate (e, t) or quantifier «e, t), t» : ,
(26)
a.
b. c.
The Atlantic/the ocean is huge. The Pacific is an ocean Every ocean is huge. .
(e)
(e, t) «e, t), t)
It seems unlikely that we can avoid the necessity of having the types (e) and «e, t), t), since individual-denoting and quantified DPs do not give rise to the same entailments. But what about predicates? 5. 1.3.
PredP
Bowers ( 1 993 ) suggests that non-verbal predication is mediated by the o functional head Pred , which is the actual head of small clauses. I will only briefly discuss this proposal, since it has no major impact on the issues under discussion: DPs (or xNPs, to take no stand on their status here) will still have at least two possible semantic types. The major problem with Bowers ' proposal is the semantics associated o with Pred . Let us consider what Predo might do. The first ob servation is that the meanings of APs (usually viewed as denoting properties in theories in which they are not taken to denote predicates) and xNPs (usually considered to be o entities in these theories) are clearly different, which means that Pred would have to have two different functions and thus two different meanings.4 This is the problem with the approach (based on Chierchia 1 985) that Bowers' himself proposes. However, since APs (or PPs, for that matter) do not interest us here, let us tum to the semantics of predicate xNPs. The first possibility to consider is that predicate xNPs denote entities (Le. an ocean denotes a particular ocean) and Predo associates the subject with this entity (an identity function). This i s untenable, since (27) does not require existence of a single unicorn. (27)
Firenze is not a unicorn.
We can therefore tum to the next hypothesis, which is that predicate DPs denote kinds (i.e. an ocean denotes the kind "ocean") and PreJ asserts that th� subject ([Spec, PredP]) is an instantiation of this kind. The problem with this approach is that xNP predicates can be predicates over kinds, as in (28).
(28)
T. Rex is a dinosaur.
4 This problem docs not arise for type-shifting, because type-shifting is taken to be a more general
operation, ttiggercd by a type-mismatch:
171
Going through a phase
The fact that the predicate
a dinosaur in (28) is interpreted as a kind of instantiation hypothesis difficult if not impossible to defend.5 One standard implementation of the fact that an xNP is underlyingly ambiguous between a kind-denoting reading and a token-denoting reading is to assume that a meaning shift occurs. However, in a framework pennitting type shifting the role of Predo would once again require clarification. A final possibility is that DPs are generalized quantifiers (Montague's original proposal) and Prec:f converts them into predicates. But then we expect other generalized quantifiers, including universally quantified DPs, to appear in the predicate position, which is clearly impossible. a dinosaur renders the kind
_
(29) * Mary is every linguist. To conclude, I believe that valence-changing operations (non-predicate to predicate among them) should be given explicit definitions. It is certainly not o impossible to suggest other variants of a meaning for Pred , but in absence of such suggestions I will consider the theory irrelevant for any discussion involving semantic types or the ability of xNPs to take arguments.
5.2.
QR-13nding site
The second LF-diagnostic for phases is that phases are possible landing sites for A/-movement and QR. If such landing sites have to be possible final landing sites (i.e. moved elements should be able to be interpreted there), then from the semantic point of view this amounts to saying that phases have type (t) (strong fonnulation). But
this formulation, applied to xNPs, leads to a contradiction with PF-diagnostics, since no internal part of an xNP that is not a PP or an xNP itself can be extracted
(at least, not in English). We will therefore now test the weak formulation of the QR-diagnostic: a phase can be an intermediate landing site for QR (i.e. an escape hatch). For overt movement we have already shown that there is no reason to believe that there is an escape hatch at the DP-edge: overt extractability out of a DP depends on how it is interpreted (no extraction out of specific or strongly quantificational DPs). What about covert movement? Matushansky (2002) argues that adjective fronting in (30a) is actually QR of the degree operator out of DP, accompanied by AP pied-piping. In (30b), the adjective is not pied-piped: (30)
a. b.
how clean a he n eles8 glass what a whet clean glass
Since degree fronting accompanied by AP-pied-piping is sensitive to o the phonological overtness of the indefinite article in Num (or in D°, this assumption is irrelevant here),6 one can conclude that it targets [Spec, NumP): D
5 IfPred means 'pan of a kind'. one can't help wondering why it doesn't surface as of
6 That the overtness of the article is a phonological rather than semantic constraint is shown by what happens in this constructions in Scandinavian, where an overt indefinite artiele can be used with 1 72
Ora Matushansky
(3 1 )
a. b.
Bresnan ( 1 973)
* how clean glasses
* how clean water
Matushansky also argues that degree fronting is an obligatory step of successive-cyclic QR of degree operators out of the xNP. In other words, [Spec, NumP] is the escape hatch out of the indefinite xNP, at least for the A' movement of degree operators in (32) (shading indicates unpronounced copies at PF and traces at LF). (32)
NumP
a.
Num'
AP -------
� so
--0----Num NP
A'
10
A capable I
a PF of degree fronting (so)
�nr�--·· ���� :
i
--
DegP
'"'--... . fV
:�
i assassin
l � bl �
capable i
""..........._.....--...,,".............".......
While at PF, movement of the degree operator so triggers the pied-piping of the entire AP containing it, at LF only the degree operator moves: 7 NumP
b.
the LF of degree fronting
� so
In (24) we assumed that the OP-internal (t}-type node is the NP. What if it was wrong? Is it possible that the subject position of xNP is [Spec, NumP] in ( 32), and it is NumP rather than NP that ends up with the semantic type ( t), which becomes (e, t) once the PRO subject is remerged? In other words, can the landing site of so in (32) be the same as the landing site of Q foreign country in (24), i.e. can the xNP-internal QR-landing site and the escape hatch for degree QR out ofxNP coincide?
plurals and mass nouns. In Scandinavian, such degree fronting is possible for all indefinites (Dclsing 1993). If the degree operator is modified (a far too interesting problem), grammatical with plurals and mass nouns (Bolinger 1972, Matushansky 2002).
the in-situ position is
7 The fonnal implementation of the semantics of degree-fronting will be discussed in the following
subsections 5.2.1 and 5.2.2. The basics are the same as in (24).
1 73
Going through a phase
5.2. 1 .
Hypothesis J: [Spec, NumP] cannot serve as the final QR-landing site
Suppose that PRO in (24) mo ves Qut of [Spec, NP] (making the NP type (t» and is merged again in [Spec, NumP] (making the NumP type (et t) by virtue of A abstraction).8 Then, in order to be interpreted DP-intemally, quantifiers have be merged exactly between these two positions. In this subsection we will examine the possibility that the DP-internal landing site for QR is [Spec, NP] (as in (24» and the external [Spec. NumP] in {33} is the escape hatch (a position higher than the final landing site of the PRO subject). (33 )
� ..
/
�� �
t� L.��%h
Ad E
��
mp (d, (e, l»
PROx l
not possible final QR-landing sites
�
Ax E Dc
possible final QR-Ianding site
Num' (I>
� Num NP(I> �
a
I
� � P C, I)
D
�try � ..
"J..y E Dc x
(I>
NP (c. 1)
Nr
pp(c. 1)
d-nice ftudent
P<'y
-c===::===-.
I
from
I
If [Spec, NumP] is not the final QR landing site, then a NumP is a phase only by v irtue of satisfying the weak formulation of the QR-diagnostic (i.e . a phase can be an intermediate landing site for QR). Howevert the truth of the matter is that any node can be an intennediate landing site for QR:
(34)
..�
(0) � .
".po.m,
--
i
(x, 0)
..,-----.
Ax E DK
(0)
�
� . . .x . . . i 1..._._......._•••__......_----_._.......:
!
8
possible intermediate QR landing site
-
1f the PRO subject is base-generated in [Spec. NPJ. then its landing site (once the generalized
quantifier a foreign country is merged) cannot be [Spec. NPJ on general syntactic grounds (merging
with the same head twice. checking a movement-triggering feature in what is arguably a a-position, etc.). Therefore, we need to make the necessary assumption that the indefinite article
semantically vacuous (i.e. that it is an identity function of merged in [Spec. NumP).
type ( 1, t» and that the
Ignoring this issue and assuming that the PRO subject in (33) does not change the nature of the semantic argumenL
1 74
a is
PRO subject is
is remerged in [Spec, NP]
Ora Matushansky
At an intermediate landing site, the variable left behind by the QRing element is abstracted over. It is easy to see from (34) that whatever the type of the adjunction node, once the QRing element is moved further (leaving behind a variable), the resulting type does not change. In other words, the weak formulation of the QR-diagnostic does not diagnose for anything.
5.2.2.
Hypothesis 2: [Spec, NumP] can serve as thefinal QR-Janding site
If NumP is the escape hatch for QR out of xNP, then the strong fonnulation of the QR-diagnostic means that NumP is a phase. From the point of view of type, this assumption means that the landing site of the OP-internal PRO subject (whose adjunction turns a (t)-type node into an (e, t) one) has to"'be higher than the landing site of the degree operator (which must adjoin to a (t)-type node):
�
(35)
i
not possible final QR-Ianding sites ....-----..., . ----NumP (c. t) PRO"
_
/
possible final QR-Ianding site
.
___
__
�ME�umPP}
.
_
.
(e. I)
-n:-Num/(l)
"A.y e
�NP{t)
Num
I
�
a x NP
-===-=
NP {e. I)
d-nice student
Whereas the final type of the tree in (35) is (e, t), it can still be argued that [Spec, NumP] serves as a final landing site for QR, and therefore, NumP is a phase, from the point of view of LF. However, this conclusion brings us into contradiction with the PF diagnostics for pbasebood. Does NumP have phonological independencelisolability? The answer to this question depends on our assumptions about the functional projections in indefinite DPs. Recall that in section 3 we have shown that DPs have full phonological independence. Then we have tacitly assumed that all noun phrases under consideration (definite or indefinite) contain a DP-layer. This assumption may be revised. If indefinite noun phrases are NumPs, unlike definite noun phrases. which are DPs, the phasehood of indefinite xNPs is confinned by PF-tests: the indefinite noun phrase. just like the definite one, exhibits full phonological independence. This would seem to be a welcome result until we consider definite DPs, which must be assumed to also have a NumP layer. As we have
1 75
Going through a phase
noted above, no sub-constituent of an xNP can be moved strandin g the article. In other words, NumPs have no phonological independence jf they are embedded inside a DP, which argues against their phasehood: an XP can hardly be a phase in one environment and not be a phase in another.9 It would seem, therefore, that the PF independence of
5.2.3.
Summary
We have presented our reasons to believe in the existence of a (t)-type node inside the DP, which functions as a QR-Ianding site, and of an escape hatch for degree QR to the clausal level. Are the two in fact the same? Neither of the two possible answers (and the corresponding fonnulations of the QR diagnostic) shed any light on whether DP is a phase. In other words, there is no evidence
that [Spec, DP] is an escape hatch, but there is also no evidence that [Spec, DP] is not one.
5.3.
Reconstruction to the edge
Since the edge of the phase is a possible landing site for QR, it is also a possible
reconstruction site. In other words, if we can show that the edge of the DP is a possible site for reconstruction, then it is a possible landing site for QR. Unfortunately reconstruction properties of the DP-edge cannot be discovered. Indeed. in order to check them, we need to have a quantifier raised out of the DP (overtly or covertly) and another quantifier, which the first quantifier would be able to scope under. Since the reconstruction position is on the DP periphery, the second quantifier can only be a higher argument of the verb:
(36)
Which senator did Sue introduce every journalist to a boyfriend of _?
Can the interrogative in (36) scope below the universal quantifier? Whether the answer is yes or no, we have no way' of proving that the reconstruction site of the interrogative is inside the existential DP. The problem is that there are multiple available landing sites that would allow the universal quantifier to outscope the interrogative: The universal quantifier may adjoin to CP, with the interrogative adjoining to vP The universal quantifier may adjoin to vP, with the interrogative adjoining to VP, or even to vP, below the universal quantifier. In other words, testing quantifier reconstruction to the DP-periphery appears to be difficult if not impossible.
5.4.
Summary
In this section we have considered the LF-diagnostics for phases. We have shown that there is a (t)-type node inside xNP (suggested to be NP) which may 9 It is not impossible that vPs can have different phaschood characteristics. depending on whether they are unaceusativc. raising and passive Or nOl (Legale 1998. 2003, and this volume). But it is necessary then to show that NumP inside a DP behaves differently from a "free-standing" NumP.
1 76
,/
Ora Matushansky
or may not coincide with the escape hatch out of the
xNP ([Spec, NumP]). We
have also argued that neither the QR-Ianding site nor the escape hatch can be [Spec, DP]. [Spec, DP] could b e an intermediate landing site for QR (a successive cyclic position). but there is no evidence for it. However, even if the LF-diagnostics for phases suggest that there exists a phase boundary inside the xNP, no intermediate projection inside the xNP shows independence at PF (isolation), at least not when it is contained within a DP. In other words, we now have a serious problem:
1.
()
If only nodes of type t are phases, DP i s not a phase (though it has PF-independence)
2.
()
If any nodes o f type t are phases. a n xNP-internal projection
must be a phase (no PF-isolabiIity) We conclude that PF and LF diagnostics produce contradictory results.
6.
Conclusion
There doesn't seem to be a one-to-one correlation between semantic properties and phonological isolability. Therefore. the major conclusion of this paper is that LF-independence (corresponding to several semantic diagnostics assumed to verify semantic isolabiIity) does not coincide with PF-independence. This should not be a possible result jf phases are spelled out at both interfaces at once. Does this mean that the successive-cyclic Spell-Out model is incorrect? The answer is no. We have assumed that the derivation is sent to the Spell-Out at particular nodes (CP. vP . . . ) and not at particular other nodes (TP, VP . . . . If so, then the nodes at which the derivation is sent to the Spell-Out (phases) should have some characteristics that non-phases do not have. We have tested for isolability at the two interfaces as a means of determining whether DPs are phases or not, and discovered that our diagnostics are contradictory on the matter. If we reject the conclusion that phases cannot be spelled out at both interfaces at once. two possible courses of action suggest themselves. The first one is to question the status of isolability as the correct test for phasehood. The natural question to ask then is what are the right tests for it - a question well beyond the scope of this paper. The alternative is to disregard either the LF diagnostics or the PF ones (which is equivalent to assuming that phase are only spelled out at one of the interfaces). Suppose we take the second course of action. Will the assumption that only PF or only LF diagnostics are the relevant ones allow us to correctly assign the phase status to vPs and CPs and the non-phase status to VPs and TPs? The answer is no.
)
6.1.
PF Spell-Out
What if we suppose that Spell-Out means stripping off phonological features of an XP? In this case. phases would only have PF-isolability and there is no reason for them to have any special properties at LF.
177
Going through a phase
(section
From the point of view of PF-isolability, DPs are certainly phases
3). This means that extraction out of them should proceed via an escape
hatch on the DP-periphery. The evidence we have seen above suggests that there is an escape hatch on the NumP periphery (for degree QR). However, we have no evidence for or against escape hatches on the DP periphery. In absence of such evidence. we can claim that if Spell-Out consists of sending the complement of the head of the phase to PF, then DPs are phases. However, it is not clear to what extent PF-independence is a reliable set of diagnostics. As shown above, they are not free from internal contradictions. To take one example, TPs can be targeted by pseudo-clefting (see (9a» , but not by other movement operations. On the other hand, extraposition, which is used to maintain that CPs but not TPs are phases. doesn' t target either vPs or VPs, which means that neither of them should be a phase, according to this diagnostic. Suppose now that we make an assumption that additional factors are at work. Then any degree of phonological independence (the ability to be isolated or moved in at least one environment) would suffice to argue that a particular constituent is a phase. This is clearly not enough, since, as we have shown, each PF-diagnostic applies to at least one of the set {CP, TP, vP, VP} . W e conclude that PF-diagnostics alone do not yield the desired results.
6.2.
LF Spell-Out
Suppose Spell-Out only occurs on the LF-branch of the computation. This hypothesis was never entertained by any of the existing proposals, for the good reason that its empirical coverage is difficult if not impossible to assess. The diagnostics suggested for LF-independence (status of a "proposition", reconstruction, successive cyclic QR and wh-movement to the edge) amount to having the semantic type (t). Whether we call (t) nodes phases or not does not seem to make any new predictions. Worse still, semantic type alone does not, properly speaking, allow us to distinguish CPs and vPs (phases) from TPs and VPs (non-phases). Suppose, as we have so far, that CPs and vPs have the semantic type (t) (abstracting away from times, possible worlds and events). If this is true, then TPs also have the semantic type (t), and therefore should be phases. On the other hand, if we allow for the time and possible world variables, then vPs and CPs will no longer have the same semantic type, assuming that the time argument e (t}), while the time argument slot ofvPs and TPs �lot of a CP .is saturated ( . IS not (type (1, t), for both). I . ' Finally, the semantic type of a VP is not altogether clear. If we disregard event semantics completely, then a VP headed by an unaccusative verb has the semantic type (t), which would make it a phase, in contradiction to the accepted assumption that even unaccusative vPs are not phases.
�
l O An a lternative hypothesis (Matushansky and Spector in progress) would ass ume that embedded CPs are no different from TPs and their time/possible world argument slots are not saturated when they arc combined with the selecting verb. This approach also results in no type difference between CPs, TPs and vPs.
1 78
Ora Matushansky
We conclude that the assumption that phases undergo only LF Spell Out does not provide us with the desired results, either.
6.3.
Results
We have tested on DPs the hypothesis that certain nodes (but not others) are loci for LF and PF Spell-Out (phases). We have adopted the standard assumption
that spelled out nodes have certain special properties stemming from the independence at the interfaces. We have shown that theoretical considerations (computational complexity) seem to require DPs to be phases. but that this assumption raises serious problems for the valuation-based theory of Case. We have also shown that DPs answer to some but not all diagnostics for independence at the two interfaces. The problem with such "partial independence" is that it is unclear what each diagnostic means. The minimal result that we believe must be retained from the analysis above is that PF and LF Spell-Outs cannot be simultaneous, if the tests suggested for the two are valid. Indeed, we have shown that for DPs, a t -type node (a possible landing site for QR and reconstruction) can only be DP internal, while no extended NP that is not a DP shows phonological independence. On the other hand, if only PF-diagnostics are retained, DPs would appear to be phases - a theory fully consistent with the fact that they can be infinitely iterated and thus must, under the assumptions we have made, at some point exhaust the workspace available. However, this would seem to require certain adjustments to Case theory, since Case-feature valuation would not seem to be a possibility. We conclude that the behavior ofDPs does not bode well for phases.
()
7.
References
Barker, C. (2000). Definite Possessives and Discourse Novelty. Theoretical Linguistics 26. pp. 2 1 1 -227. Bolinger, D. ( 1 972): Degree words. Mouton, die Hague. Bowers, J. ( 1 993). The Syntax of Predication. Linguistic Inquiry 24, pp. 59 1 -656. Bresnan. J. ( 1 973): Syntax of the Comparative Clause Construction in English, Linguistic Inquiry 4, pp. 275-343. Campbell, R. ( 1 996). Specificity operators in SpecDP. Studia Linguistica 5012, pp. 1 6 1 1 88.
Chierchia, G. ( 1 985). Fonnal Semantics and the Grammar of Predication. Linguistic Inquiry 1 6, pp. 4 1 7-443. Chomsky, N. ( 1 995). The Minimalist Program. MIT Press, Cambridge, MA. Chomsky, N. (2000). Minimalist Inquiries: The Framework. In: R. Martin, D. Michaels, and 1. Uriagereka, eds., Step by Step: Essays on minimalist syntax in honor of Howard Lasnik. MIT Press, Cambridge, MA. Chomsky, N. (200 1 a) : Derivation by Phase. In: M. Kenstowicz. ed., Ken Hale: a life in language. MIT Press, Cambridge, MA. Chomsky, N. (200I b): Beyond Explanatory Adequacy. In: MITOPL 20. MITWPL, Cambridge, MA.
·179
r
Going through a phase Cinque, G. ( 1 980). On Extraction from NP in Italian. Journal o/ltalian Linguistics 5, pp. 47-99. Davies, W. and S. Dubinsky (2003). On extraction out of NP. Natural Language and Linguistic Theory 2 1 , pp. 1 -37. Delsing. L.-O. ( 1 993). The internal structure of noun phrases in the Scandinavian languages. Ph . D. thesis, University of Lund. Diesing, M. ( 1992). Indefinites. MIT Press, Cambridge, MA. En�, M. ( 1 99 1 ). The semantics of specificity. Linguistic Inquiry 22, pp. 1 -26. Fiengo, R., and J. Higginbotham ( 1 9 8 1 ). Opacity in NP. Linguistic Analysis 7, pp. 39442 1 . Gavruseva, E . (2000). On the syntax o f possessor extraction. Lingua 1 1 0, pp. 743-772 .
Giorgi, A., and G. Longobardi ( 1 99 1 ). The syntax of noun phrases: Configurations. parameters and empty categories. Cambridge University Press, New York. Heim, I. (2000): Degree Operators and Scope. In: Proceedings of SAL T X, Cornell University, Ithaca, NY, Cornell Linguistics Club, pp. 40-64. Heim, I., and A. Kmtzer (1 998). Semantics in Generative Grammar. Blackwell, Oxford, UK. Johnson, K. (2002). Towards an Etiology of Adjunct Islands. Ms. UMass, Amherst. Legate, A. J. ( 1 998). Verb Phrase Types and the Notion of a Phase. Ms. MIT. Legate, A. J. (2003). Some Interface Properties of the Phase. Linguistic Inquiry 34/3, pp. 506-5 16. Mahajan, A. (1 992). The specificity pondition and the CEO. Linguistic Inquiry 23, pp. 5 1 0-5 15. Matushansky, O. (2002). Movement ofDegree/Degree ofMovement. Ph.D. thesis, MIT. Matushansky, 0., and B. Spector (in progress). Intensions or propositional functions? Ms. CNRSIENS. Miller, G. A. ( 1 956). The magical number seven, plus or minus two: Some limits on our capacity for processing information. Psychological Review 63, pp. 8 1 ·97. Nissenbaum. J. ( 1 998). Derived predicates and the interpretation of parasitic gaps. In: O. Percus and U. Sauerland, eds., The Interpretive Tract: MIT Working Papers in Linguistics 25. Nissenbaum, 1. (2000). Investigations of Covert Phrase Movement. Ph.D. thesis, MIT. Pesetsky, D., and E. Torrego (2001 ). T-to-C Movement: Causes and Consequences. In: M. Kenstowicz, ed., Ken Hale: a Life in Language. MIT Press, Cambridge. Pesetsky, D. • and E. Torrego (2002). Tense, Case, and.the Nature of Syntactic Categories. Ms. MIT, To appear in J. Gueron and 1. Lecanne, eds., The Syntax of Time, MIT Press. Sharvit, Y. and P. Stateva (2002). Superlative Expressions, Context, and Focus. Linguistics and Philosophy 25, pp. 453-505. Stowell, T. ( 1989). Subjects, specifiers and X-bar theory. In: M. Baltin and A. Kroch, eds., Alternative Conceptions ofPhrase Structure. Academic Press, New York. Szabolcsi, A. ( 1 983). The possessor that ran away from home. The Linguistic Review 3, . pp. 89-102. Szabolcsi, A. ( 1 986). Comparative superlatives. In: N. Fukui, T. R. Rapoport, and E. Sagey, eds., Papers in Theoretical Linguistics, MITWPL 8, pp. 245-265 . MITWPL, Cambridge, MA. Szabolcsi. A. ( 1994). The noun phrase. In: F. Kiefer and K. E. Kiss, eds., Syntax and Semantics 27: The Syntactic Structure of Hungarian Academic Press, New York. Tellier, C. ( 1 99 1). Licensing Theory and French Parasitic Gaps. Dordrecht, Kluwer. Woisetschlaeger, E. ( 1 983). On the question of definiteness in "an old man's book". Linguistic Inquiry 14, pp. 1 37-1 54.
1 80
Ora
Matushansky
UMR 7023 1 5, rue Catuliennc 9 3200 Saint Denis Frdnce
matushan@noosJr http://mapage.noosJr/matushansky/main.html
181
r
r
UTAH at
Merge: Evidence from multiple applicatives·
Martha McGinnis University of Calgary It is argued that UTAH is a property of Merge, not a condition on representations. It is argued that both upward and downward Merge is
needed to capture the word order and
A-movement
properties of Bantu
applicatives, along with their selectional properties. The different A movement properties of applieativcs arc attributed largely to their phasal l nonphasal status and to cross-linguistic variation in the distribution of EPP features.
L
Introduction
In the generative linguistic tradition, the language faculty is regarded as a biological system. Like any other biological system. it is subject to resource limitations. For example, it has been recently argued that derivations are computed in phases, which constitute small, manageable domains for syntactic operations (Chomsky 2000, 200 1 ). There has also been an attempt to eliminate global conditions on representations in favour of local conditions on operations. For example, what was previously formulated as a condition on D-structure, the Uniformity of Theta Assignment Hypothesis (UTAH; Baker 1 988), is most straightforwardly recast as a property of the operation Merge. This approach makes new empirical predictions which have not yet been explored in great detail. I argue here that these predictions are confirmed by a set of observations about Bantu applicatives (APPLs). Bantu APPLs can be classified into two typest which Pylkldinen (2002) calls High and Low APPLs. High APPLs establish a semantic relation between an individual and an event; for example, a benefactive argument can benefit from an eventt or a locative argument can specify where it takes place. Low APPLs instead establish a semantic relation between two individuals, such as a recipient and a theme, or a source and a theme. Compositionalityt then. implies that a High APPL has a structure like that in (l a), with an individual-denoting specifier and an event-denoting complement; and that a Low APPL has a structure like ( l b), with an individual-denoting specifier and an individual denoting complement. · Thanks to Teal Bissell Doggett, Donna Gerdts. Alec Marantz, and Norvin Richards for thought
in which phases and EPP now play minor supporting Kinyarwanda speakers Alexandre Kimenyi and Alexandre Rutayoberana. APPL:
provoking discussions that inspired this paper, roles. Thanks to
applicative; ASP: aspect; B EN : benefactive;
cs :
case; FOC: focus; f V : final vowel; INST:
PASS: passive; PR : prescnt; PST: past; DB: Duranri and 8yarushengo 1 977; K: Kimenyi 1980. Some interlinear glosses are trivially altered. instrumental; LOC: locative; NOM: nominative; OBl: oblique;
MIT Working Papers in Linguistics 49. /83-200
Perspectives on Phases @ 2005 Martha McGinnis
UTAH at Merge
(1)
a.
APPLH
b.
�PLH
D
�
V
�
APPLH
V
�
v
v
APPLL
�
DP
APPLL
�
APPLL
(DP)
DP
As a result of their selectional properties, Low APPLs can appear only with verbs that take an internal argument, while High APPLs can appear with both transitive and intransitive (including unergative) verbs. 1 For example, English has a Low recipient APPL which establishes a direct semantic relation between two DPs, such as Mary and a letter in (2a). As predicted, the Low APPL is incompatible with an unergative verb. (2)
a. b.
*
John sent Mary a letter. John ran Mary. CJohn ran for Mary.')
On the other hand, the Kichaga benefactive is a High APPL, which relates a OP individual to a verb phrase event. In this case, an internal argument is not required: (3)
N·a-i-zric·i·a mbuya. FOC-he-PR-run-APPL-FV friend 'He is running for a friend.'
Bresnan and Mosh; J 990: J 49
This analysis predicts certain restrictions on the possible combinations of APPLs. A High APPL head could, of course, merge with a VP containing a Low APPL, as with any other VP. Likewise, a High APPL head could merge with a High APPLP, which also denotes an event. However, a Low APPL head should not be able to merge with a High APPLP, both because the High APPLP does not denote an individual, and because the High APPL would then have no event-denoting argument.2 A review of the Bantu literature reveals that the expected combinatorial restrictions do not apply (McGinnis and Gerdts 2003). For example, in Haya, a Low dative APPL can take a High instrurnental APPLP as its complement (4). (4)
1
Kat' a-ka-5iig-is' 6mwaan' amajilt' ekitambila. Kato he..pST-smear-INST child oil handkerchief 'Kato smeared oil on the child with the handkerchief.'
DB: 63
Pylkkiinen (2002) also argues that a depictive secondary predicate can modify a High applied
argument, but not a Low one, except in languages where other internal arguments can be so modified. I have not yet been able to tcst this prediction for the cases discussed here.
2 Not discussed here, but also of interest. are combinations of Low
APPLs.
For example.
Kinyarwanda allows the combination of dative and alienable possessor APPLs (Kimcnyi 1 980: 101-105). See Legate (2003 : 184) for discussion of a similar case in Choctaw.
1 84
Martha McGinnis
I will argue that such apparent violations of compositionality are predicted if semantic composition obtains at the moment of Merge, rather than at a subsequent level of representation. I will argue that in the cases described above, relations established by Merge are subsequently altered, yielding a structure that no longer directly reflects the compositional interpretation of the derivation. 2. UTAH as a condition on external Merge
Baker ( 1 988:46) formulates UTAH as follows, within the framework of Government and Binding Theory:
(5 )
Identical thematic relationships between items are represented by identical structural relationships between those items at the level of D structure.
Minimalist theory seeks to eliminate such global conditions on representations. Within this framework, UTAH can be most straightforwardly recast as a condition on external Merge - that is, Merge of an item not already merged into the derivation. For example. it could be postulated that identical thematic relationships between items are represented by identical Merge o erations applying to them. The condition may be more clearly stated as follows:
f
(6)
Thematic relationships are established by external Merge.
The two formulations in (5) and (6) have different empirical consequences. For example, (5) in principle allows non-local thematic relationships, while (6) allows such relationships only between two elements that merge. There is some support for the more local version. On the basis of an asymmetry between subject and object interpretations, Marantz (1 984) argues that the external argument of a trans,itive c lause must receive its theta-role compositionally from the verb phrase, not directly from the main verb. This requirement must be stipulated under (5), but it follows naturally from the formulation in (6), on the conventional assumption that the external argument merges with the verb phrase after the direct object has merged. The formulation in (6) also opens up the possibility of non compositional structures, since it is the derivation that is compositionalJy interpreted, not the resulting structure. Under standard assumptions, each Merge operation is reflected in a structural relation of sisterhood, which persists throughout the derivation; thus, structures will always be compositional. However, if the sisterhood relation created by Merge could be subsequently altered, or if two elements could merge without becoming sisters, then (6), but not (5), would allow the derivation to be interpreted compositionally, even if this interpretation is not accessible from the resulting syntactic structure. There is indeed evidence that a sisterhood relation created by Merge can be subsequently altered (Phillips 1 996, 2003; see also Richards 2002). Constituency conflicts suggest that the standard assumptions about constituency 3 Hornstein (200 I) argues that internal Merge can also establish thematic relationships.
1 85
UTAH at Merge
are incorrect (Ernst 1 994, Pesetsky 1 995). For example, in (7), [visit them] appears to be a constituent, since it undergoes movement. On the other hand, [them [on each other 's birthdays]] also appears to be a constituent, since the pronoun c-commands the anaphor. On the traditional view, the two constituent structures are incompatible: the PP either is, or is not, part of the VP constituent. Nevertheless. both seem to co-exist in a single derivation. ( 7)
Bill said he would visit themi, and. [visit them;! he did [tj [on each other'si birthdays}] .
Phill ips argues that the apparent conflict arises because a constituent created by Merge is destroyed by a subsequent application of Merge. First the object DP merges with V, as in (8a); then P merges, as in (8b), followed by its complement DP.4 When P merges, the object op i s reanalyzed as its specifier. (8)
a.
VP
�
V
�
Vi it
DP
,Q
b.
VP
�
V
I
visit
PP
�
P
DP
�; I� � P
on
P2
e. o. 's; birthdays
Under this view, the moved VP constituent in (7) is interpretable because it establishes a dependency with a copy VP like the one in (8a), even though this copy is subsequently altered. After the complement of P merges, the "cascade" structure in (8b) allows the pronoun to c-command and bind the anaphor. This procedure results in a representational opacity: when P merges,. the sisterhood relation between V and OP, in (8a) ceases to exist. Nevertheless, this OP is still interpreted as the object of V. Phillips argues that the entire syntactic derivation proceeds from the top down - or, more accurately, from left to right. However, if the formulation of UTAH in (6) is to be preserved, it is not clear that a purely left-to-right derivation can be maintained. As noted above, there is evidence that the external argument receives a compositional theta-role from the verb phrase. If this thematic relationship is established by external Merge, then the verb phrase must be constructed before the external argument is merged with it. even though (hi most languages) the subject precedes the verb phrase. I propose that the derivation actually proceeds outwards from the verb - either upwards or downwards, depending on what undergoes Merge. If the structural reflection of a thematic relationship can be eliminated before all Merge operations are complete, then UTAH must be treated as a condition on Merge, not as a condition on representations. I will also argue that two elements can merge without becoming sisters. The resulting structure
4 For simplicity. I will assume that DPs arc completed in a parallel set of Mcrge operations before merging with other constituents.
1 86
Martha McGinnis
therefore does not provide sufficient information to identify the thematic relationships involved. Consider, for example. the portion of a derivation shown in (9). In (9b),
it seems clear that v has merged with v. and VP becomes the sister of v.
VP: VP is completed just prior to merging
(9)
b.
a.
VP
�
V
DP
vP
.......
v
VP
�
V
DP
On the other hand, in (8b), it is not clear what P has merged with: VP is completed just prior to merging P, but DP) becomes the sister of P. In the next section I argue that a structural configuration like (8b) can arise from merging a
head with a constituent just completed. Thus, a head selecting an event-denoting argument, like a High APPL, can merge downward with VP.
In such a case, the thematic relationship created by Merge is not reflected structurally by a sisterhood relation. However, this thematic relationship can arguably be identical to one that is reflected by sisterhood. For example, a passive by-phrase has a compositional thematic role identical to that of the external argument in the corresponding active clause (Marantz 1 984: 1 29).
Nevertheless, an external argument is merged upward. where i� can bind internal arguments ( l Oa), while a by-phrase is merged downward, where it can be bound by internal arguments (1 Ob). ( 1 0)
a. b.
Each professofi taught heri students heri favourite subject. The students were taught every subjecti by ilsi foremost expert.
This alternation is something of a puzzle under the representational version of UTAH. Under the Merge version, however, the external argument and the by phrase can have the same thematic relation to vP if they both merge wi th vP one upwards, the other downwards. The resulting structural relationships are distinct, but the Merge relation is identical.
3. Upward
and downward Merge in Bantu appUcatives
In this section I present a more extended argument that UTAH is a condition on Merge. In 3. 1 , I present cases expected under either fonnulation of UTAH. in which a High APPL merges above a theme, a Low APPL, or another High APPL. In 3.2, I argue that event-selecting APPL heads can also merge downward (see also McGinnis and Gerdts 2003). As a result, the thematic relationship created by Merge is not reflected structurally by sisterhood. Some remarks are in order regarding the syntactic analysis I will
assume for applicatives. As. noted above, it has been argued that syntactic derivations proceed in phases (Chomsky 2000, 200 1). According to this view, merging certain functional heads (for example, v or C) results in the completion of a phase. The domain of the phase (the complement of the phase head) is then
1 87
UTAH at Merge
sent off for phonological and semantic interpretation. This results in the Phase Impenetrability Condition: elements in the domain of a phase are inaccessible to further syntactic operations. However, the head and edge of the phase (specifiers and adjuncts of the phase head) remain accessible. Thus, for a constituent in the domain to escape the phase, it must first move to the edge. Chomsky proposes that a phase head can trigger movement to the edge of the phase by means of a EPP feature. For example, wh-subjects can escape an embedded clause by first moving to the edge of the embedded CPo There is some debate in the literature regarding which functional categories constitute phase heads. McGinnis (200 1 , 2002, 2004) argues that High APPLs are phases, while Low APPLs are not. Under this view, an EPP feature on the High APPL head allows the direct object to leapfrog over the applied argument to spec-APPL in a passive ( 1 I a). From there, it can move again to spec-v, and on to the subject position. If no EPP feature is added to the High APPL head, the applied argument can move to spec-v instead ( l i b). and thence to the subject position. Thus, the typical High APPL is symmetrical, allowing either object to move to subject position in a passive. .
( 1 1)
a.
vP
b.
�
10
t �
DO
L
v
t
vP
�
A PPLH
� 10 APPLH �VP APPLH
v
[�
APPLH
t
AP
�P �O
�
V
If a Low APPL is not a phase head, it has no EPP feature to allow movement of the lower argument. Moreover, only the closest eligible constituent can be attracted to the edge of the vP phase ( 1 2). As a result, the typic al Low APPL is asymmetrical, allowing only one object to move to subject position in a passi ve s .
5 McGinnis (2004) argues that in some Low APPLs, the applied argument has inherent Case. which
makes it ineligible for A-movement. The result is also asymmetrical. but in this case only the theme can raise to subject position.
1 88
Martha McGinnis
vP
( 1 2)
�
10
v
� v VP � V APPLLP � t APPLL �DO APPLL
However, the High/Low distinction alone does not
fully
account for the
symmetrical/asymmetrical distinction (Doggett 2004). This is partly because the availability of EPP features is subject to cross-linguistic variation. Thus, some High APPLs allow only the applied argument to raise to subject position, because the APPL head has no EPP feature to allow the lower object to move. Doggett also argues that both objects can move out of a Low APPL if v has two available EPP features. This argument is based on an intriguing set of facts from Haya, a Bantu language spoken in Tanzania (Duranti and Byarushengo 1 977). Haya has a Low dative APPL, expressing a direct semantic relation between the theme and the indirect object ( 1 3).
( 1 3)
Kat' a-k-66Iek' 6mwaan' epica. Kato he-PST-show child picture 'Kato showed the child a picture. •
DB:58
The Haya "passive" is more like an inverse derivation: the thematic role of the external argument is expressed, not by a prepositional phrase, but by an optional DP. In a double-object passive, the dative argument can raise to subject position ( 1 4a). More interestingly, the theme also can raise to subject
under certain circumstances. For example, it can raise if the external argument is not expressed ( 1 4b). ( 14)
a.
Omwaanj a-k-66Iek-w-a
kat'
tj epica.
child he-PST-show-PASS-FV Kato picture 'The child was shown the picture by Kato. • b.
Epic' e-k-66Iek-w-a (*kat') 6mwaana. picture he-PST-show-PASS-FV Kato child 'The picture was shown to the child by Kato. '
DB: 59
DB:59-60
Doggett proposes that in a Haya passive, v has two EPP features. One can be checked by merging an external argument in spec-v, while the other is checked by an internal argument that raises first to spec-v, then to the subject position. However, if the external argument is not merged, then both EPP features of v can be checked by internal arguments ( 1 5). Thus, the theme can
move over the dative argument, and onwards to the subject position.
1 89
r·
UTAH at Merge
( 1 5)
v
o0v
� �V v � V APPLL � APPLL t �t APPLL
10
v
If the external argument is merged, the theme can still raise to subject if the dative argument incorporates into the verb ( 1 6). Doggett argues that in this case, the dative argument is a clitic that undergoes head-movement to v. The dative clitic does not check the second EPP feature of v, so the theme can do so instead.
( 1 6)
Ba-ka-mw-66Iek-w-a kato. they-PST-him-show-PASS-FV Kato 'They were shown to him by Kato.'
Since the DO can also incorporate into the verb, ( 1 6) can shown to them by Kato.
DB: 60
also mean 'He
was
'
4.1 Merging High applicatives upwards
Both the representational and Merge versions of UTAH predict that High APPLs can merge with either a VP or another High APPLP. Such derivations can be illustrated with Kinyarwanda, a Bantu language spoken mainly in Rwanda (Kimenyi 1 980). Benefactive ( 1 7a) and locative APPL5 ( 1 7b) are High APPLs, as demonstrated by their ability to merge with unergative VPs: 6 ( 1 7)
a.
umugabo. U mug6re a-ra-kor-er-a she-PR-work-BEN-ASP man woman 'The woman is working for the man. '
6 In Kinyarwanda it is important to distinguish between uncrgative with a Low
K:32
verbs, which cannot combine
APPL, and transitive verbs with a null thematic object, which can (i). Object suppression
as in (i) is possible for most arguments with objcct properties (sec footnote 8), though not for datives or benefactives (Kimcnyi 1 980: 6 1 ). (i)
Umwaalimu teacher
y-eerets-e
abanyceshuuri.
he -show -ASP students
'The teacher showed the students.'
1 90
K:61
Martha McGinnis
b.
ameeza. Abaana b-iica-y6-ho chi ldren they-sit-ASP-LOC table 'The children are sitting on the table.'
K:92
Both benefactive and locative A PPLs can also combine with a verb phrase containing a Low APPL. For example, Kinyarwanda has a Low dative
APPL. A VP containing a dative APPL can merge with a benefactive ( 1 8a) or locative APPL ( I 8b). ( 1 8)
a.
b.
Umugore a-ra-he-er-a
umugabo
imbwa ibiryo.
woman she-PR-give-APPL-ASP man dog 'The woman is giving food to the dog for the man.'
food
K:65
Umugore a-ra-he-er-�m07 ishuiiri umuhuungu ibitabo. books woman she-PR-give-APPL-ASP-LOC school boy 'The woman is giving the books to the boy in the school. ' K:96
The neutral word order suggests that the (underlined) benefactive and locative arguments c-comrnand the dative and theme arguments, as is expected if a High APPL merges with VP, while a Low APPLP merges with V. As noted above, High APPLs are typically symmetrical, allowing either the applied argument or the next lowest argument to move to the subject position of the passive. This type of alternation can be seen with benefactive+dative APPLs, where the benefactive ( 1 9a) or the dative ( 1 9b) argument can move to the subject position of a passive. 8 ( 1 9)
a.
ti
UmugabOi a-ni-h6-er-w-a
man
he-PR-give-BEN-PASS-ASP
imbwa ibiryo n'umugore. dog food by-woman
'For the man is given food to the dog by the woman.'
b.
K:66
umugabo ti ibiryo n 'umugore. imbwai i-ni-he-er-w-a dog it-PR-give-BEN-PASS-ASP man food by woman 'The dog is given food for the man by the woman. •
K:65
The Low dative APPL i s also symmetrical, as it is in Haya: both the
dative ( 1 9b) and the theme
(20)
(20) can
move to subject position in a passive.
IbiryOi bi-ra-he-er-w-a umugabo imbwa ti n 'Umug6re. food it-PR-give-BEN-PASS-ASP man dog by woman 'The food is given to the dog for the man by the woman. ' K:65
7 Many. though not all. locative applicatives in Kimenyi ( 1 980) also have the applicative suffix
-irl-er.
This suggests that the locative suffixes �o
clities doubling the locative argument.
and
-mo are not themselves
APPL
heads. but
S In addition to movement to subj�t, Kimenyi (1 980) also discusses other object properties:
pronominalization. reflexivization. object-subject reversal. and extraction in relative clauses. clefts
and wh-questions. I have corrected an apparent error in the translation of ( 1 9a) to reflect Kimenyi's description of the facts; the verbatim version is 'The man is given food for the dog by the woman.'
191
UTAH at Merge
Following Doggett's analysis of Haya, we might propose that the head of the High benefactive
APPL
has EPP features that attract both the theme and the
dative. However, the word order of postverbal arguments appears to be the same in the active (18a) and the passive in (20). This is not expected if movement of the theme is contingent on prior movement of the dative to spec-APPL. Another possibility. which Doggett proposes for symmetrical Low
APPLs in British English, is that the theme can merge either above or below the dative argument. She notes that variable ordering of themes and datives is observed in a number of languages, including Croatian, Greek, French, and Italian. In British English, however, a dative argument always c-commands a (non-pronominal) theme in an active clause. Nevertheless. Doggett proposes that the theme can be generated above the dative argument here as well, provided that it raises to the subject position. Cases of this sort have been observed elsewhere; for example, Rackowski
(2002)
discusses a High instrumental in
Tagalog that must move to the subject position. as indicated by another
argument
cannot
move
to
subject
instead
(21 b).
ang (21 a); Marantz
(1984:245-246) notes a similar case in Chichewa. (21)
a.
I-pinang-lakad
ng
OBL-ASP-walk
CS man
lalaki ang ANG
'The man walked with a stick. b.
* Nag-Iakad
ng
tungkod. stick
•
tungkod ang
NOM.ASP-walk CS stick
ANG
lalaki. man
Rackowski 2002:52
'The man walked with a stick.'
I
will assume that arguments that must move to subject position are
generated with an uninterpretable Tense feature. Another plausible case of this
kind is PRO : 9
(22)
a.
* Mary believed [him to have fired PRO].
b.
We forced him
c.
We forced Mary
[PRO to fire Mary]. [PRO. to be fired ti
by him].
The example in (20) is open to a similar analysis: the theme can be generated above the dative argument as long as it is generated with an uninterpretable Tense feature. An EPP feature on the High benefactive APPL head will then allow the highest argument of the Low APPL - dative or theme - to move over the benefactive argument, then continue on to the subject position. In short, it is reasonable to conclude that a benefactive APPL merges above VP. Such a result is expected under both representational and Merge versions of UTAH, given that the High benefactive APPL head selects an event argument. In the High locative APPL, the locative argument can move to subject position in a passive (23a), but lower arguments, such as the theme, cannot
(23b). These observations can be captured if the locative APPL head lacks an
9 A similar case is observed in Italian,
that is, it must agree with fin ite T
192
where one type ofimpcrsonal $; is obligatorily nominative (Cinque 1988). PRO must instead agree with non-finite T.
Martha McGinnis
EPP feature to attract a lower argument, so that the theme is trapped within the domain of the APPLHP phase.
(23 )
a.
IshuUrij ry-oohere+w-e-ho tj igitabo n'uumwaalimu. school it-send-ASP-PASS-ASP-LOC book by-teacher K:94 'The school was sent the book by the teacher.'
b.
... Igitaboj cy-oohere-j-w-e-ho ishufui ti n 'uumwaalfmu. by-teacher book it-send-ASP-PASS-ASP-LOC school 'The book was sent to the school by the teacher.' K:95
Likewise, when a locative APPL merges with a VP containing a dative APPL, neither the dative (24a) nor the theme (24b) can become the subject of the passive: (24)
a.
*
Umuhuunguj a-ra-he-er-w-a-mo boy he-PR-give-APPL-PASS-ASP-LOC
ishuUri tj school
ibitabo n 'umugore. books by-woman 'The boy is given the books in the school by the woman.' b.
* IbitabOi
books
K:96
ishufui bi-ra-he-er-w-a-mo they-PR-give-APPL-PASS-ASP-LOC school
umuhuungu ti boy
n 'umug6re. by-woman
'The books are given to the boy in the school by the woman.' K:96 Since datives and themes can normally raise to subject, the locative. evidently blocks them from doing so in (23b) and (24). However, the locative argument itself can raise to subject position, as shown in (23a). This supports the view that the locative c-commands both arguments: as the closest argument, it is attracted to spec-T in a passive. Again, this result is expected under both a representational version and a Merge version of UTAH, on the assumption that the High Locative APPL merges with VP, while the Low dative APPL phrase merges with V. Locative and benefactive High APPLs can also combine, as shown in (25). This is also predicted by both versions of UTAH. Like VPs, High APPLPs denote events. Thus an event-selecting High APPL can merge with a High APPLP. (25)
Umwaana y-iicar-i-ye-ho intebe umugabo. child he-sit-APPL-ASP-LOC chair man 'The child is sitting on the chair for the man. '
K:JJ3
193
UTAH at Merge
Which APPL head merges with which APPLP in (25) is not entirely clear. Locality considerations suggest that the benefactive merges above the locative. As noted above, the locative generally blocks lower arguments from moving to subject position in a passive. However, the locative doe"s not block movement of the benefactive, as shown in (26a). The benefactive generally does not block movement of a lower argument, and indeed the locative argument can also move to the subject position (26b). (26)
a.
Umugabo y-iicar-i-w-e-ho intebe n ·uumwaana. man he-sit-BEN-PASS-ASP-LOC chair by-child K:114 'For the man was sat on the chair by the child. '
h.
umugabo n 'uumwaana. intebe y-iicar-i-w-e-ho chair it-sit-BEN-PASS-ASP-LOC man by-child K:114 'The chair was sat on for the man by the child. '
On the other hand, the neutral word order in (25) has the locative preceding the benefactive. If left-to-right word order mirrors c-command, as has been reported for other Bantu languages (Marantz 1993, Ngonyani 1996), then this suggests that the locative c-commands the benefactive. Kimenyi (1980:114) reports that it is possible to reverse the linear order of the two arguments, but then the animate argument is interpreted as a possessor ('The child is sitting on the man's chair.'). McGinnis and Gerdts (2003) suggest that the order in (25) is derived by stylistic movement to avoid ambiguity. Another possibility is that the Benefactive can merge above the locative only if it is generated with an uninterpretable Tense feature, and thus must move to subject position. Although a number of details remain unresolved, the cases described above are broadly compatible with either a representational version or a Merge version of UTAH. High APPLs merge with VPs or with other High APPLPs. and High applied arguments c-commands themes and Low applied arguments. In the following section, however, I will present cases in which a High APPL is c commanded by a theme or by a Low applied argument. Such cases support a cascade analysis along the lines of Phillips (1996, 2003), with downward Merge. This analysis is compatible with the Merge version of UTAH, but not with the representational version. 4.2 Merging High applicatives low
We have already seen that benefactive and locative APPLs in Kinyarwanda are High. Instrumental APPLs also appear to be High, since they can combine with unergatives: (27)
Umuhiinzi a-kor-eesh-a isuka. farmer he-work-INST-ASP hoe 'The fanner is working with the hoe. •
Overdulve 1975:209
They can also combine with transitive VPs (28a). Instrumental passives are symmetrical: either the instrument (28b) or the theme (28c) can raise to subject position.
194
Martha McGinnis
(28)
a.
b.
Umugabo a-ra-andik-iish-a iba.rUwa ikammu. man he-PR-write-INST-ASP letter pen 'The man is writing a letter with the pen.' ikaramu; i-ra-andik-iish-w-a ibarUwa li n 'umugabo. pen he-PR-write-INST-ASP letter by-man 'The pen is used to write a letter by the man .•
c.
ibaruw3i i-ra-andik-iish-w-a ti ikammu n 'umugabo. letter he-PR-write-INST-ASP pen by-man 'The letter is being written with a pen by the man. '
K:81
K:81
K:83
However, unlike benefactive and locative arguments, instrumental arguments in a transitive clause follow the theme, suggesting that the theme c commands the instrument. Quantifier-pronoun binding supports this view: (29a) shows that a quantificational theme can bind the instrument, while (29b) shows that a quantificational instrument cannot bind the theme (McGinnis and Gerdts 2003).10 (29) a.
N-a-rungul-ish-ije buri muryango irrufunguzo rwawo. I-PST-open-INST- ASP each door key its A. Rutayoberana. p.c. 'I opened each doon with its. key.'
b.
N-a-fUngul-ish-ije umuryango wayo buri rufunguzo. I-PST-open-INST-ASP door its each key A. Rutayoberana, p.c. 'I opened itsj door with each keYjl.j.'
On a bottom-up approach to Merge, an instrumental APPL that merges with VP should c-conunand an object contained in VP. However, this does not seem to be the case in (29). Although the instrumental APPL has the semantics of a High APPL, it seems to merge below the theme. Like benefactives, instrumental APPLs can combine with locative APPLs: (30)
ikibaho imibare ingwa. Umwaalimu y-a-andik-iis&ije-ho teacher he-PST-write-INST-ASP-Loe board math chalk �The teacher wrote math on the blackboard with chalk. K: J 07 '
As noted above, the locative APPL blocks lower arguments from raising to subject position in a passive. If the instrumental argument merges below the theme, a locative should block it from raising to subject. This prediction is confirmed. In a passive locative+instrumental APPL, the locative argument can raise to the subject position (31 a). The theme cannot, as usual in locative APPLs (31 b). The instrumental also cannot move to subject position (3Ic). This observation crucially distinguishes High instrumentals from High benefactives. 10 The data in (29) should not be used without confirmation, as the tones have not been checked.
Alexandre Rutayobcrana describes these applicativcs as "old-fashioned" Kinyarwanda. Many speakers prefer a non-applicative construction with a PP instrument instead
195
UTAH at Merge
whose movement is not blocked by the locative, as shown above (Ki rnenyi 1980, Gerdts and Whaley 1993. McGinnis and Gerdts 2003). (31)
a.
Ikiba.hoi cy-a-andik.iish.ij·w-e-ho ti imibare ingwa.11 blackboard U-PST-wnte-INST-ASP-PASS-ASP-LOC math chalk K:107 'On the blackboard was written math with chalk.'
b.
* Imibarei y-a-andik-iislrij-w-e-ho
ikibaho ti ingwa. math It-psT-wnte-lNST-ASP-PASS-ASP-LOC blackboard chalk K:108 'Math was written on the blackboard with chalk.'
c.
* ingwai y-a-andik-iislrij-w-e-bo ikibaho imibare ti. chalk it-PST-write-INST-ASP-PASS-ASP-LOC blackboard math K: 1 08 'Chalk was used to write math on the blackboard. '
The differences between instrumental and benefactive APPLs can be captured if the benefactive merges upward, while the instrumental merges downward. Further evidence that a High APPL can merge downward comes from Haya.12 As seen above (in (14», a passive with a syntactically expressed external argument allows the next lowest argument to move to subject position, but blocks movement of an argument merged lower. On the other hand, if the external argument is not syntactically expressed, either of the two next lowest arguments can raise to subject position. Haya also has an instrumental A P PL, which appears to be High. According to Trithart (1977), it can occur with intransitive verbs, including inherent reflexives (32), which are usually unergative (Takehisa 2003). (32)
Kat' a-k-6og-es' ebyombo. Kato he-PST-clean-INST soap 'Kato washed (himself) with soap. '
Trithart 1977:78
The instrumental APPL can also combine with transitive verb phrases (33a). As in Kinyarwanda, instrumental APPLs are symmetrical: in the passive, either the instrument (33b) or the theme (33c) can raise to subject position, even when the external argument is syntactically expressed. (33)
a.
b.
Kat' a-ka-teez' omwaan' ekiti. Kato he-PST-hit.INST child wood 'Kato hit the child with a piece of wood. '
DB:61
Ekitii ki-ka-teez-i-bw-a kat' omwaana ti. wood it-PST -hit-INST-PASS-FV Kato child 'The piece of wood was used by Kato to hit the child'
DB:67
11 I have omitted the phrase n 'uuwaalimu 'by the teacher' from these exam ples. 12 Many thanks to Teal Bissell Doggett for bringing these faets to my attention.
1 96
Martha McGinnis
Omwaan\ a-ka-teez-i-bw-a kat' ti ekiti. child he-PST-hiHNST-PASS-FV Kato wood 'The child was hit with a piece of wood by Kato. '
c.
DBp6J
Word order in (33 a ) suggests that the theme c-commands the as in Kinyarwanda. This conc l usion is supported by the observation that when the instrumental APPL combines with a Low dativ e APPL (34), the dative appears to c-command the instrumental.
instrumental argument.
(34)
am ajut e kitambdla oil handkerchief 'Kato smeared oil on the child with the handkerchief.'
Kat'
a-ka-siig-is'
omwaan'
Kato he-PST smear-INST child
'
.
-
DB:63
Again, the word order in (34) suggests that the dative is the highest argument. Stronger evidence comes from locality effects. In a passive the dative argument can raise to subject position even if the external argument is syntactically expressed (35a). As usual in a dative APPL, the theme can raise to subject only if the external argument is not syntactically expressed (35b). Intriguingly, the same is true of the instrument (35c). ,
(35)
Omwaan'j a-ka-siig-is-i-bw-a 13 kat' t. amajuf 6kitambila. child he-PST-smear-INsT-PASS-FV Kato oil hanky 'The child was smeared with oil with the handkerchief by Kato.
a.
'
DB:63 b.
AmajUlai ga-ka-siig-is-i-bw-a (*kat') omwaan' ti 6kitambila. oil it- PST -smear-JNST-PASS-FV Kato child hanky 'The oil was smeared on the child with the handkerchief (*by DB:63 Kato).'
c.
Ekitambalai ki-ka-siig-is-i-bw-a (*kat') omwaan' amajuta ti. hanky it-PST-smear-INsT-PASS-FV Kato child oil 'The handkerchief was used to smear oil on the child (*by Kato).'
DB:63 The contrast between (35a) and (35c) suggests that the High instrumental APPL merges below the Low dative APPL. . In both Haya and Kinyarwanda, then, there is evidence that a High APPL merges low, even though it selects an event argument. The apparent contradiction can be resolved if constituents can merge downward. Suppose that once the VP is completed, a High APPL can merge either above it (36a) or below it (36b). The Merge operation will establish a thematic relationship between the High APPL and VP in both cases. A High applied argument will c c omm and a Low applied argument and the theme if merged upwards, and will .
13 Trithart (1 9 71) argues that the instrumental APPL involves two morphemes, -5 (-is) and -i. The-i is generally not seen in surface forms, but conditions the insertion of the -bw allomorph of the
passive suffix. which occurs only post-vocalically.
197
UT AH
at Merge
be c-commanded by a Low applied argument and the theme if merged downwards . (36)
a.
b.
APPLH
.......-.
10
VP
.......-.
V
APPLH
�
VP
APPLH
DO
.......-.
V
APPLH
.......-.
DO
APPLH
.......-.
APPLH
10
Note that instrumental APPLs in both languages are symmetrical, even when they merge downwards. This suggests, intriguingly, that the EPP feature on the instrumental APPL head can attract its own complement into its specifier, as shown in (37a) for an APPLP merged below VP, and in (37b) for an APPLP merged below a Low dative APPL. (37)
a.
VP
b.
�
V
APPLH
� Inst APPLH � Theme ApPLH � APPLH
VP
� V APPLL � Oat APPLL � APPLL APPLH � Inst APPLH � Theme APPLH � APPLH
Kinyarwanda and Haya applicatives provide evidence that UTAH is a constraint on Merge, not on representations. If a High APPL merges downward, as in (36b), the thematic relationship between the High APPL head and VP is not expr�ssed by a sisterhood relation. After the APPL head merges, the thematic relationship between V and the theme is also no longer expressed by a sisterhood relation, since the theme is reanalyzed as a specifier of APPLH. Thus, there is no transparent mapping from representations to thematic relationships. On the other hand, if UTAH is a property of Merge, then each thematic relationship is determined at the particular point in the derivation when the two items in question merge. Under this view, UTAH determines thematic relationships even when two elements do not merge as sisters (downward Merge), or when their sisterhood relation is destroyed by a subsequent application of Merge. The argument for downward Merge sketched here is somewhat preliminary, and many properties of Bantu applicatives remain elusive. Nevertheless, there is convincing evidence of a conflict between selectional requirements and structural positions. The downward Merge analysis constitutes
198
Martha McGinnis
a promising step towards resolving these conflicts. Thi s analysis provides support for the Minimalist program of reformulating conditions on representations, such as VTAH, as conditions on operations. I f successful, this program will illuminate the ways in which resource limitations shape and constrain the language faculty. References Baker, Mark C. (1988). Incorporation: A theory oj grammatical function changing. Chicago: University of Chicago Press. Bresnan, Joan, and Lioba Moshi (1990). Object asymmetries in comparative Bantu syntax. Linguistic Inquiry 21, 147-185. Chomsky. Noam (2000). Minimalist inquiries: The framework. In Roger Martin. David Michaels, and Juan Uriagereka (eds.), Step by step: Essays on Minimalist syntax in honor of Howard Lasnik. Cambridge, Mass.: MIT Press, pp. 89-155. Chomsky, Noam (2001). Derivation by phase. In Michael Kenstowicz (ed.), Ken Hale: A life in language. Cambridge, Mass.: MIT Press, pp. 1-52. Cinque, Guglielmo (1988). On si constructions and the theory of arb. Linguistic Inquiry 19,521-581. Doggett, Teal Bissell (2004). All things being unequal: Locality in movement. Doctoral dissertation, MIT. Available from MITWPL. Duranti, Alessandro, and Byarushengo, Ernest Rugwa (1977). On the Notion of "Direct Object". In Ernest Rugwa Byarushengo, Alessandro Duranti and Larry M. Hyman (eds.), Haya Grammatical Structure (Southern California Occasional Papers in Linguistics 6. Los Angeles, CA: Department of Linguistics, USC), pp. 45-7J. Ernst, Thomas (1994). M-command and precedence. Linguistic Inquiry 25: 327-335. Gerdts. Donna B .• and Lindsay Whaley (1993). Kinyarwanda multiple applicatives and the 2-AEX. CLS 28: 186-205. Hornstein, Norbert (2001). Move! A Minimalist theory a/construal. Oxford: Blackwell. Kimenyi, Alexandre (1980). A Relational Grammar of Kinyarwanda. Berkeley: University of California Press. Legate, Julie Anne (2003). Warlpiri: Theoretical implications. Doctoral dissertation, MIT. Available from MITWPL. Marantz, Alec (1984). On the nature oj grammatical relations. Cambridge, Mass.: MIT Press. Marantz, Alec (1993). Implications of asymmetries in double object constructions. In Sam Mchombo (ed.), Theoretical aspects 0/ Bantu grammar. Stanford: CSLl Publications. pp. 113-151. McGinnis, Martha (2001). Phases and the syntax of applicatives. In Minjoo Kim and Uri Strauss (eds.), Proceedings a/NELS 31. Amherst, Mass.: GLSA. McGinnis, Martha (2004). Lethal ambiguity. Linguistic Inquiry 35: 47-95. McGinnis, Martha, and Donna B. Gerdts (2003). A Phase-Theoretic Analysis of Kinyarwanda Multiple Applicatives. In Sophie Burelle and Stanca Somesfalean (eds.), Proceedings oJlhe 2003 Annual Conference a/the Canadian Linguistic
Ass ociation . Department de linguistique et de didactique des langues. Universite du Quebec a Montreal. pp. 154-165. Ngonyani, Deogratias (1996). The morphosyntax of applicatives. Doctoral dissertation, University of Califomia, Los Angeles. Overdulve, C. M. (1975). Apprendre la langue rwanda. The Hague: Mouton. Pesetsky. David (1995). Zero syntax: Exper;encers and cascades. Cambridge, Mass.: MIT Press.
199
UTAH at Merge
Phillips, Colin (1996). MITWPL.
Order and structure. Doctoral dissertation, MIT. Available from
Phillips, Colin (2003). Linear order and constituency. Linguistic Inquiry 34: 37-90. Pylkkiinen, Liina (2002). Introducing arguments. Doctoral dissertation, MIT. Available from MITWPL.
Rackowski, Andrea (2002). The structure of Tagalog: Specificity, voice. and the
distribution of arguments. Doctoral dissertation, MIT. Available from MITWPL. Richards, Norvin (2002). Very local A-bar movement in a Root-First derivation. In Samuel D. Epstein and T. Daniel Seely (eds.), Derivation and Explanation in the Minimalist Program. Malden , Mass.: Blackwell, pp. 227-248. Takehisa, Tomokazu (2003). Against the syntactically unifonn analysis of reflexive intransitives. In Sophie Burelle and Stanca Somesfalean (eds.), Proceedings of the 2003 Annual Conference of the Canadian Linguistic Association. Department de linguistique et de didactique des langues, Universite du Quebec
Ii Montreal. pp. 224-235. Trithart, Le e (1977). Causatives and instrumentals. In Ernest Rugwa Byarushengo, Alessandro Duranti and Larry M. Hyman (eds.) , Haya Grammatical Structure (Southern California Occasional Papers in Linguistics 6). Los Angeles. CA: Department of Linguistics, USC, pp. 73-88.
Department of Linguistics SS 820, University of Calgary 2500 University Drive NW Calgary, AB TIN IN4 Canada
[email protected] http;llwww.ucalgary.cal-mcg;nnis
200
On the EPP*
Shigeru Miyagawa Massachusetts Institute of Technology The EPP is shown to interact with either agreement or focus. I suggest that there is a structural reason why agreement and focus are parameterized relative to the EPP: both agreement and focus are merged very high in the structure, in the region of C. While languages vary as to whether they are "agreement prominent" (e.g .• Indo-European) or "focus prominent" (e.g .• Japanese, Kinande. Turkish). I will show that, with
close inspection, all languages instantiate both features in some fashion, although in some cases a feature does not always get expressed the same way.
1. Introduction
I will be concerned with issues of movement, particularly those operations that move a category to the specifier of TP or the specifier of a projection higher than the TP. The following exemplify the kinds of movement I will be concerned with.
(1) Movements to TP and higher wh-movement focus movement "agreement" movement (e.g., thematic subject) scrambling My analysis of these movement operations is that they are fundamentally the same: they are all triggered by the EPP feature on the relevant head (Chomsky 2000). Whether it is movement of the subject to the Spec ofTP to meet the EPP requirement of T, or the movement of a wh-phrase to the Spec of CP, the movement is triggered by the EPP. The EPP was originally suggested by
•
I am grateful to Noam Chomsky for discussions of the ideas
in this paper
stages in their analysis. Thanks
at various
development, including providing some key suggestions that belped to shape the
also to Ken Hiraiwa. Jordi Fortuny, Jaklin Komfilt. Sabine latridou. Martha McGinnis, David
pesetsky. and Norvin Richards for helpful suggestions. This paper is not the
paper I presented at the
EPP/Phase Conference in Janua})' 2003. In that presentation I gave arguments that QR is a covert
version of scrambling. In attempting to put this idea down on paper, it became apparent that I had to · first clarify what the EPP is, since 1 depend on this notion for accounting for scrambling (Miyagawa
2001. 2003). The present paper arose
as 8 result, and I thank the editors. Martha and Norvin. for
allowing me to contribute it to this volume instead of the original paper.
MiT Working Papers in LinguistiCS 49. 201-235 Perspectives on Phases © 2005 Shigeru Miyagawa
On the
EPP
Chomsky (1981) because. of the appearance of the expletive in existential constructions
(There stands a statue in the town center). The agreement is
between the verbal inflection and the postverbal nominal, and the expletive there fills the Spec of TP. The expletive makes it possible for the existential construction with this long-distance agreement to have a subject. The
EPP
is in
fact informally referred to as the requirement that a clause must have a subject. I will generalize this to all movements of a category, focusing my attention in this article on movement to the Spec ofTP and to positions higher than the TP. I will not deal with movement to the Spec of vP, which presumably also involves the EPP (Chomsky 2001). The
EPP
is merely a stipulation. We hope that it will follow from some
deeper properties of the grammar. For example, it may be related to the requirement identified by Alexiadou and Anagnostopoulou (200 1 ) that something must vacate the VP. Despite this stipulatory nature of EPP,
I will
continue to assume it. My concern in this article is to discover the ways in which the
EPP
interacts with other formal features, with the hope that this and other
properties noted in the literature (e.g., Alexiadou and Anagnostopoulou 1998, Baker 2003) will eventually lead us to the real identity of the EPP. In so doing, there are three questions I wish to address about (l). First, if all four movements in (1) are the same "EPP" movement, why do they involve such disparate heads? Wh-movement and focus movement are typically thought to be in the region of C. This leaves agreement, which is normally thought to be on T (for subject-verb agreement). I will suggest, using data from recent work by Carstens (2003) and Komfilt (2004), that we can postulate the agreement feature at C instead ofT. If this turns out to be correct, it has the desirable result that much - possibly all - of the inflectional features (agreement, focus) are in the region of the two phase heads, C and v (cf. Chomsky 2000, 2001, for relevant comments on v). I will. show that scrambling, too, involves a feature on a head in the region of C. . Second, why is it that a given language typically only has a subset of the four movements? English has wh-movement, while Turkish doesn't. Japanese has scrambling, but not French. The answer I will give is that these differences are only apparent. I will give evidence that all languages share the same features relevant to all four movements. The variations emerge because of different ways in which the shared features interact with each other in a given language. This view of language is consistent with the Uniformity Principle (Chomsky 200 I: 2).
(2)
Uniformity Principle In the absence of compelling evidence to the contrary, assume languages to be uniform. with variety restricted to easily detectable properties of utterances.
The approach I am going take is very much in line with Sigurdsson (2003). who assumes this uniformity principle literally for all languages, and suggests the "Silent Principle," by which he means that a given language, which shares all of
202
Shigeru Miyagawa
the features with other languages, has features that are not pronounced. This is, at least in part, the reason for the differences among languages. I am going to adopt a strong version of the Uniformity Principle along these lines, and assume that, at least for inflectional features, such as agreement and focus, every language not only shares a unifonn set, but, contra Sigurdsson, every language overtly manifests these features in some fashion. Thus, for example, Japanese, which shows no agreement inflection, is predicted to have something that reflects '·agreement," something I will attempt to show. This strong version of the Uniformity Principle cannot be right for all features of a language. After all, languages do vary. In adopting the strong version at least for inflectional features, I intend to explore some of the outer bounds of the Uniformity Principle. Third, although I assume that all four movements are the same "EPP" movement, they clearly are triggered by different features, e.g., agreement, wh, etc. How many features do we need to account for the four movements? Is it four (wh, focus, agreement, scrambling)? In the literature we find all four proposed. I But it would considerably weaken the explanatory power of the analysis to postulate one-for-one; it would be describing the problem. I will suggest that the four movements can be reduced to two features that interact with the EPP: agreement and focus. 2. Agreement, Focus, aod the Uniformity
Principle
My analysis of the EPP is based on a parametric variation between agreement and focus, whereby a language is either agreement prominent or focus prominent. The idea is that in a given language, the EPP on T, which I assume to be universal, works in tandem with either agreement or focus. This is similar, although not the same, as an idea of Baker (2003), who argues on the basis of Kinande and pro-drop languages of Indo-European (cf. Alexiadou and Anagnospoulou 1998), that there are three relevant features, the EPP, agreement, and (nominative) Case. In his system agreement is central, and it works in tandem with the EPP, or with Case. Although I take a slightly different approach, I inherit the basic notion of variability in the way features interact with each other, including the EPP I also utilize Baker's (2003) analysis of Kinande, a Bantu language, which plays a key role in the analysis. An immediate problem arises with the idea of a focus-agreement parameter. Focus and agreement are usually thought to be on fundamentally different heads. Focus is commonly postulated on the FOCUS head that is higher than T and in the region ofC (e.g.,Culicover and Rochemont 1983, Rizzi 1997). In contrast, agreement in, for example, subject·verb agreement, is normally construed as being on T. While it is not entirely implausible for two features on fundamentally different heads to vary parametrically, it would be more plausible if they aren't on such vastly different heads. There is sufficient .
I
For focus movement, see, for example, Brody 1990, Ord a fifth feature, 'iopic."
6iiez 1997, and Rizzi 1997, and Yang
2004. Rizzi in fact suggests
203
On the EPP
evidence to
associate
focus with a head higher than T, so if we are going to do
anything about "head parity," we need to look at agreement. To get right to the
point, I suggest that agreement in, for example. subject-verb agreement, is
principally associated with a head higher than suggesting this possibility;
cf.
T (I
thank Noam Chomsky for
Chomsky, to appear. and Boeckx
2003,
for a
similar idea). I will assume that it is C, which will put it in the same region as focus. Below, I give two pieces of evidence that the agreement in subject-verb
agreement may show up on C. Carstens
Haegeman
(3)
a.
(2003)
notes the following West Flemish examples from
(1992).
I-think
Kpeinzen
dan-k (ik) morgen go3O. that-I (I) tomorrow go
Kpeinzen
da-j
'I think that I'll go tomorrow.'
b.
I-think
that-you
(gie)
morgen
goat.
(you) tomorrow go
'J think that you'll go tomorrow.'
c.
Kvinden
I-find
dan
die boeken te
that-PL the books
diere
zyn.
too expensive are
'I find those books too expensive.'
While a number of linguists have proposed that the complementizer-subject agreement is an instance of the agreement on T raising to C, Carstens argues that the agreement originates on C (see Carstens
references for and against this idea).
(2003)
(2003)
for additional
In these examples, the embedded verb also
inflects for agreement, suggesting that the agreement also shows up on T. Under
our analysis. this suggests that the agreement on C may percolate down from C to T. What is the reason for this feature percolation to T? It may be to allow a
category to move to a position that is not at the edge, so that it does not automatically need to take on discourse properties of definiteness/specificity.
Alternatively, it may be to create an A-position to which a category can
internally merge without having to occupy an operator position. There are other equally plausible scenarios, but I will not pursue them here. Another instance of complementizer agreement is given in Komfilt
(2004).
Jn Turkish there is a difference between subject relativization and non
1 The data
for her analysis. but not the analysis itself. appeared earlier in Komfilt (2000).
subject relativization.2 In subject relativization, the embedded subject does not carry agreement, while in non-subject relativization it does.
204
Shigeru Miyagawa
(4)
a.
subject as the target of relativization [ [ej ge�en yaz ada-da ben-i gor-en] ki!ii-Ieri] [[ last summer island-Loe I-Acc see-(y)An person-PI 'the people who saw me on the isalnd last summer' (No $-feature morphology; special nominalization fonn on predicate)
b.
a non-subject as the target of relativization [[pro ge�en yaz ada-da ej gor-dug-um] ki�i-Ieri] [[ last summer island-Loc see-DIK-l.sg person-PI 'the people who(m) I saw on the island last summer' (cjl-feature morphology; general indicative nominalization form on predicate).
In (a) the subject is relativized and there is no agreement on the embedded verb. Kornfilt argues that this lack of agreement has to do with the fact that C in the relative clause agrees with the subject (e) . and this suppresses the emergence of agreement on the embedded verb. In (b) what is relativized is the object; no agreement with C is triggered so that the embedded verb is free to agree with the subject. These Turkish examples suggest the following. Assuming that the agreement originally appears on C, it gets copied onto T, as in the case of West Flemish and as seen in (4b) above. If the subject appears in Spec of CP, however, the agreement on C picks up this subject, and the agreement does not percolate down to T. This is what we see in (4a).3 Maintaining the view that the EPP is on T, we have the following representations for focus and agreement (I will use the head-final order).
(5) Focus
CP ------- C'
� CAGREEMENT
TP
�
FOCUS
TEPP
---,
__
E"(------J
percolate down
) Another work that is relevant here is Watanabe ( 1993). which postulates the sort of agreement I am proposing.
205
On the
EPP
(6) Agreement CP
------ C' ------TP
�
TEPP
(
CFOCUS
AGREEMENT
--,
percolate down
The FOCUS/AGREEMENT feature is matched with a feature on a category in syntax, for example, the thematic subject. In most cases this category is brought
up to the Spec of TP, to satisfy the EPP on T. I say "in most cases" because there are known cases of long-distance agreement, which lead to something other than the agreed-with category ending up on the Spec of TP. We will see many cases of such long-distance agreement. When the agreed-with category does move for the purpose of satisfying the EPP, there is no need for it to move any higher than the Spec ofTP since T is the head that has the EPp.4 By placing agreement as well as focus on a "high" head in the region of
C,
we make it plausible for these two features to be the two polarities of the same parameter. Because agreement and focus constitute the primary
inflectional features in our system, this means that we isolate all such features on phase head, C, leaving T with the EPP feature and, if relevant, Case (cf. Pesetsky and Torrego 2001). But do focus and agreement constitute some sort of a natural class? If so, there is a plausibility argument to add to our empirical and theoretical arguments. In fact, Simpson and Wu
(2001) give a number of
arguments to show that historically, agreement in a variety of languages developed from some sort of a focus structure. Although mucb of what they deal .with is what we would nonnally call concordance, as in. the case of the French ne . . .pas, their argument that dependencies such as concordance/agreement find historical source in focus structure is highly suggestive of why we see focus and agreement as constituting the two polarities of a parametric variation.
Before turning to the actual analysis, I wish to touch on an issue which will drive much of the analysis presented in this article. As mentioned above, the view of language I will present is that a language is either focus prominent or agreement prominent. But it isn�t the case that� for example, an agreement prominent language does not also have focus. In fact, as we will see, some of the most interesting cases are those in which an agreement prominent language also evidences focus, and, in parallel, those cases in which a focus prominent
� I am glossing over the problem of infinitives. Presumably the infinitive
-
T also has the EPP feature,
but, for example. in the ECM construction, the embedded subject moves higher (according to the traditional analysis). This is an issue for the analysis of ECM
does me embedded subject actually
move (Koizumi 1995), or does it not have to (Lasnik 1999)? Ifit does move beyond the lower Spec
of TP. this means that Case is also pertinent to the kinds of derivations we are looking at. something
not
at
all surprising, and. in
fact, expected.
due to limitation of space and time.
206
There are interesting issues which I cannot pursue here
Shigcru
Miyagawa
evidences agreement. What these cases suggest is th at l anguages do not have a subset of a uniform set of features but rather , they appear to have the complete set always. Some re cen t work by C i nque (1999) and the cart o graphy linguistics of Rizzi and othe rs hint at this idea that all languages have essentially the same universal features/structures. The most explicit proposal along this line is by S igurdsson (2003). He notes Cho msky s (2001) ideas of L-Un ifo rm ity ("language is uniform in the relevant sense") and L Se le ction ("languages make different selections of features from a univer sa l feature set"), and argues that we only nee d L-Uniformity, which means that all languages contain the same universal set of features. Sigurdsson then assumes that, since we don't apparently always see some of these features overtly manifested in a given language, there are many such features without phonetic instantiation. He calls th is the "Silent Principle:' Although I will differ on -some p oints, Iwill adopt the spirit of his idea that all languages share a uniform set of features. Where I depart from his proposal is in taking the stronger position that at least for inflectional features, all features in every language are not only uniformly shared, but that they are all somehow pronounced. As a way to demonstrate the gene ral approach, let us contrast it with two prop o s als in the lite ra ture. In highly influential series of works, Fukui (1986, 1995) and Kuroda (1988) look at some of the same types of movement for English and Japanese. In the case of Fukui, differences suc h as th e presence of wh-movement in English and its absence in Jap an ese and the presence of scrambling in Japanese an d its absence in English, among many parametric variations, are due to the fact that Japanese does not have functional categories (or that functional heads are "inerttt). Fukui, t her e fore identifies a deep difference between English and Japanese - that of lack of functional categories for Japanese. Although there are some s imil arities to the theory I am pursuing, we see immediately that by the strong version of the Uniformity Principle I am assuming, Fukui's overall approach is unten abl e s Our approach predicts that there should be no su ch "deep" differences between any two languages. For Kuroda, who looks at a similar range of par ametric variations, it is the lac k of "forced agreement" that gives Japanese the properties it has in contrast to English. Kuroda points to this l ack of forced agreement for the absence of overt wh-movement in Japanese. The lack of forced agreement is also r esponsible for what he considers to be optional movement typ ically called scrambling, which, for him, may freely move a category into any specifier position because there is no a gre em ent Although I differ on both views, as I will sho w the analysis in this article inherits Kuroda's basic premise that Japanese is not an agreement langu age and it makes partially the same prediction with regard to the absence ofwh-movement, a point shared by Fukui's approach as well. In this sense, my analysis is an extension of Kuroda' swork. Below, I will tum to the discussion of focus prominence using data from Japanese. language a lso
,
'
"
"
,
,
.
.
,
,
S See Shibatani (1990) for counterexamples to each of Fukui's (and to some extent also Kuroda's)
works.
207
On the EPP
3. A
Focus-prominent Language: Japanese
In speaking about "focus", I will adopt E . Kiss's ( 1 998) distinction. According to E. Kiss there are two tyj es of focus, which she calls "informational focus" and "identificational focus: (7)
E . Kiss ( 1 998) (i) Informational focus: what is not presupposed in a topic-focus (theme-rheme) structure. (ii) Identificational focus: expresses exhaustive identification.
"Informational focus" in (7i) is the familiar " topic-focus" structure, in which the relevant category at the left edge refers to what the sentence is about, and the remainder, or some portion of the remainder, provides new information about this "topic." As we will see later, there is a specific prosodic pattern associated with sentences with informational focus. Briefly, sentences with informational focus have the nuclear stress on the syntactically lowest element in the structure (Cinque 1993). (8)
John bought a BOOK.
In (8) the direct object, along with the verb, are lowest on the syntactic tree; setting aside the verb (see later), the head of the direct object receives nuclear stress. A typical way to understand this sentence is that John here is the "topic," and some or all of the VP gives new information about this topic of John. Identificational focus is different from informational focus in that, first of all, the stress falls on whatever element is associated with identificationa! focus. In contrast to (8), it is possible to create an identificationally focused sentence by, for example, placing the primary stress on John.
(9)
JOHN bought a book.
Here the primary stress is not on the lowest element in the tree, but rather on the subject. What distinguishes identificational focus from informational focus is that identificational focus expresses "exhaustive identification" (E. Kiss 1 998). This is illustrated in the cleft construction in English. ( l 0)
It was a hat and a coat that Mary picked for he�elf.
In this example, the identificationally focused phrase, "a hat and a coat," expresses exhaustivity. Consequently, as Szabolcsi ( 1 98 1 ) notes, the following is not a logical consequence of the sentence above. 6
A similar, but perhaps not identical, distinction was drawn by Kuroda (1972. 1 992), which he calls "thctic" and "categorical" judgments based on the work of the philosopher Anton Marty. While I will use E. Kiss's terminology of infonnational focus and identificational focus, it should be noted that a similar distinction was already suggested earlier by Kuroda.
208
Shigeru Miyagawa
( I 1)
I t was a hat that Mary picked for herself.
in ( 1 0) , the set, "a hat and a coat," e x haustively and e xclusiv ely identifies the entities that Mary picked for herself. Hence one cannot pick out a subset, as in ( I I ), wi th the notion of exhausti vity . It is logically wrong. As E. Kiss and Szabolcsi note, this notion of exh austivi ty carries over to the noti o n of contrastiveness as well. The reason is tha t
3.1 Japanese J will now turn to the study of identifi cational focus in Japanese. Japanese does not have any overt agreement, h ence, on the focus-agreement parameter, it is a focus prominent language. As already noted, there are two types of "focus," informational and identificationa!. The construction I will look at in Japanese to illustrate its focus prominence is the indetenninate pronoun expression. We will see that the indetenninate pronoun is associated with identificational focus. In Japanese a wh-phrase can be interpreted as an indetenninate pronoun in the context of th e universal quantificational particle mo. Thi s combination of wh-mo is an NPL
( 1 2)
Taroo-ga
nani-mo
kawa-nakat-ta.
Taro-Nom what-MO 'Taro didn't
buy-Neg-Past buy anything.'
As is well known, the wh-phrase portion and mo can be separated (Kuroda 1 965, Nishigauchi 1990).
( 1 3)
nani-o kai-mo Taroo-ga Taro-Nom what-A ce buy-MO 'Taro didn't buy anything.'
sina-kat-ta. do-Neg-Past
Here the wh-phrase
as an indetenninate pronoun occurs in the object position the accusative case marker 0, and the universal quanti fi cat ional particle mo occurs on the verb stem. One distinct property of the indetenninate pronoun expression is that it is always associated with a sense of exhaustive interpretation - something like "absolutely nothing/no one." I will assume that the indetenninate pronoun is associated with the focus feature which gives it this "identificational focus" interpretation. As we will see below, the pattern of grammaticality associated with the indeterminate pronoun demonstrates the focus-prominent nature of Japanese, and certain properties that this entails. I will make use of Kishimoto 's (2001) analysis of the indeterminate pronoun. As we have observed, the indetenninate pronoun is a wh-phrase; it is interpreted as an indetenriinate pronoun in the context of the universal particle mo. K ishimoto proposes that in order for the wh -phrase to be interpreted as an indeterminate pronoun, the wh-phrase must be dominated by the same
with
209
On the EPP
immediate maximum projection that dominates m o ; that is, mo an d the 7 i ndete nninate pronoun must occupy the same minimal domain. As part of his analysis, Kishimoto argues that the verb raises to in Japanese and it takes with it mo that attaches to it.
v
,
vP �v, � VP V-mo-v
(14)
.......
tv
In this structure, mo can go with any indeterminate pronoun in its local vP. In Kishimoto's analysis the object is assumed to move to the Spec of vP This is why i t is fine to have an object indeterminate pronoun as we saw above in (1 3). As a pi e ce of evidence for his analysis, Kishimoto observes that an indeterminate pronoun cannot occur in the subject position. .
( 1 5)
•
Dare-ga
tabe-rno sina-kat-ta. do-Neg-Past ,Anyone didn't eat pizza.
who-Nom
piza-o
pizza-Ace eat-MO •
Kishimoto assumes the EPP here. and argues that the subject indeterminate pronoun dare 'who' raises to the Spec of TP to satisfy the EPP of T, and this takes it outside the scope of mo. whi ch is on v. According to the analysis I'm pursuing, the relevant portion of the structure of (15) is the fol lowing.
1 Kishimoto' s licensing occurs for the most part at LF, although he himself mentions the
EPP
requirement. Hiraiwa (2002) makes explicit the idea tbat mo must license the indeterminate pronoun
at overt syntax.
210
Shigeru M iyagawa
( 1 7)
CP
-------
c �
C' � -----r
dare-ga [focus] vP
percolate
�
TEPP
.......
tSUB
FOC US
V'
A
VP
V-mo-v
The focus feature percolates down to T and enters into an AGREE relation with the featu re on the indeterminate pronominal subject, and the indeterminate pronoun is raised to the Spec ofTP to satisfy the EPP on T. An interesting fact which Kishimoto does not observe is that this sentence is equally ungrammatical even if the object "pizza" is scrambled to the left of the indeterminate pronoun subject (cf. Miyagawa 2003). ( 1 8) * Piza-oi dare-ga ti tabe-mo pizza-Acci who-Nom ti eat-MO 'Pizza, anyone didn't eat.'
sina-kat-ta. do-Neg-Past
We will see later that in the informational focus structure (as opposed to the identificational focus structure above), it is fine for the object to move in this way and satisfy the EPP on T. Why is it not possible in an identificational focus structure as in ( 1 8)? Before going further, let us consider another piece of data from Kishimoto. First, recall that the indeterminate pronoun is fine in the object position. The example is repeated below as ( 1 9). ( 1 9)
Taro-ga nani-o kai-mo sina-kat-ta. Taro-Nom what-Acc buy-MO do-Neg-Past 'Taro didn't buy anything.'
Now observe what happens if we scramble the object to the head of the sentence (cf. Kishimoto 2001 ). (20)
*Nani-oi Taroo-ga Taro-Nom what-Accj 'Taro didn't buy anything.'
kai-mo buy-MO
sina-kat-ta. do-Neg-Past
As shown, if the object indeterminate pronoun is scrambled to the left of the subject, the sentence becomes ungrammatical. What does this indicate? It
21 1
On the EPP
obviously indicates that this scrambled object indetcnninate pronoun cannot reconstruct. This, in tum, signifies that the movement of the indeterminate object must be solely A-movement. That is, it must exclusively be the EPP triggered movement. Crucially, it cannot be A' -scrambling, which would allow it to reconstruct (cf. Mahajan 1 990, Tada 1 993) . The pattern of grammatieality we just observed gives credence to the idea that Japanese is a focus prominent language, and, in particular, that the EPP on T in Japanese interacts with focus. To see this, let us first look at the existential construction in English. (2 1)
a. b.
There appeared a boy in the room. A bOYi appeared ti in the room.
(2 1 a) demonstrates that agreement and EPP may apply separately. (2 1 b) indicates that if the agreed-with phrase does move, it must be by the EPP and it moves into the Spec ofTP as expected. Returning to the indeterminate pronoun construction in Japanese, we can see that it is similar to the English existential. If the focus feature agrees with the postverbal nominal, and there is no movement, something else moves into the Spee ofTP. The example ( 1 9) is repeated below as (22). (22)
Taro-ga nani-o kai-mo sina-kat-la. Taro-Nom what-Ace buy-MO do-Neg-Past 'Taro didn't buy anything.'
We can see immediately that there is a difference between English and Japanese: while English uses the expletive in this context, Japanese apparently does not have any expletive, and allows a "close" DP - the subject in this ease - to raise to the Spec of TP and satisfy the EPP. If the agreed-with phrase moves, the movement must be triggered by the EPP on T, leading to the postverbal indeterminate pronoun to have to move to the Spec of TP. This leads to ungrammaticality as we saw, since this movement takes the indeterminate pronoun out of the licensing domain of mo. The example is repeated below. (23)
Taroo-ga *Nani-oi Taro-Nom what-Acci 'Taro didn't buy anything.'
kai-mo
buy-MO
sina-kat-ta. do:.Neg-Past
Finally, there is strict locality with FOCUS agreement, just as in the agreement phenomenon in English. If the agreed-with phrase is closest to T, then it must be the one to satisfy the EPP on T. Another, non-identificationally focused DP, cannot be moved across it and into the Spec of TP. The example is repeated below.
212
Shigeru Miyagawa
(24) * Piza-oj dare-ga ti tabe-mo pizza-Acci who-Nom tj eat-MO 'Pizza, anyone didn't eat. '
sina-kat-ta. do-Neg-Past
The scrambled object, "pizza-Ace," cannot satisfy the EPP on T because the agreed-with phrase, the subject indetenninate pronoun, is closest to T. What we have observed is that focus "agreement" in Japanese involving identificational focus works in the same way as agreement in Indo European. This gives credence to the idea that focus and agreement are two polarities of a single parameter. The one difference we found was that when there is long-distance agreement, English uses the expletive to satisfy the EPP, while Japanese allows a nonnal DP the subject, for example - to meet the EPP. This may reduce to the parametric variation at hand. English, being an agreement language, requires any item that can meet the EPP to have some sort of agreement feature compatible with the T; Chomsky (200 I) in fact speculates that the expletive has a person feature. But in Japanese, a focus-prominent language, no agreement feature plays a role at T, and a full DP in the right location is free to meet the EPP. It may be that this DP has some sort of a "focus" feature - the "topic" feature, or some such feature. It is, in fact, probable, but I will not pursue it further. -
4. Focus Prominent Languages and Informational Focus
In the previous section I demonstrated how focus-prominence manifests itself in Japanese - it has properties similar to agreement in Indo-European. The evidence involved the indetenninate pronoun construction, which is associated with the "exhaustive" interpretation of identificational focus structure. The indetenninate pronoun has the focus feature associated with it, and this feature is matched with the same feature on T that percolated from C. As a result the focused phrase moves to Spec of TP to satisfy the EPP on T. But not every Japanese sentence has a phrase associated with identificational focus. In the following discourse, Taroo in (b) does not entail exhaustivity. (25 )
a.
kaimasita ka? hon-o Taroo-ga book-Ace bought Q Taro-Nom
b.
Hai, Taroo-ga kono hon-o kaimasita. yes Taro-Nom this book-Ace bought 'Yes, Taro bought this book. '
We understand Taroo in (b) as what the sentence is about (''theme'' in theme rheme, or Utopic" in a broad sense), and the rest of the sentence, or some part thereof, provides new information ("focus"). This sort of Utopic" is not limited to thematic subjects. Under the right context, the same kind of characterization can be given to the object that has moved to the head of the sentence.
213
On the EPP
(26)
Hon-oi Taroo-ga tj kaimasita. book-Acci Taroo-Nom tj bought Alfhe book, Taro bought.' •
In this sentence "book" is what the sentence is about, and the remainder is understood to give some new infonnation about this ·'book." What we observed is the typical structure for what E. Kiss ( 1 998) calls informati onal focus." Informational focus is the portion that is no t the topic; what corresponds to the topic is, in the examp les in (25b) and (26), on the left periphe ry . The phrase on the left periphery does not receive stress because it is not focused. The main stress falls on the object in (25b) and the subject Taroo in (26). Later we will see that this stress assignment is due to the universal nuclear stress rule (Cinque 1993) . The question about infonnational focus in Japanese is this. Given that there is a focus feature that must seek out an appropriate feature on a category within TP, what does this feature "agree" with? Remember that it is mandatol)' that the focus feature find a comparable feature on a category within the TP, because T has the EPP feature, and this EPP can only be satisfied in conjunction with focus. If there is an identificational focused element, such as an indeterminate pronoun, we saw that the focus feature on FOe agrees with the focus feature on the identificationally focused category. But what about informational focus? I will simply assume that, in the absence of an i dentification ally focused category, some sort of agre ement takes place between focus and a categol)' within TP. The agreed-with phrase rai ses to the Spec of TP to satis fy the EPP on T. The agreed-with phrase does not carry a focus" feature; rather it is simply probed by the feature on T. This much is due to syntax; syntax does not care about informational structure. It is up to the interface system responsible for assigning informational structures to · inteIpret the syntactic structure. This interface system would use the input from syntax, which has the structure below, and impose the informational structure of topic-focus "
"
.
(27) [TP
' n ";-r-' lOCUS
. . [lIP
� tOpIC
.
"
Some category (e.g.� subject, object) raises to the Spec of TP, and everything else stays in-situ. What is important to point out is that the raised category is not focused; it is the unfocused portion of the infonnatio nal focus structure. The interface system therefore needs to know that the agreement here is anti focus sort of speak; it is the topic of informational focus. The crucial test given in Miyagawa (200 1 ; cf. also 2003) to show the effects of the EPP on T involves the universal quanti fie r zen 'in ' all' and its inteIpretation relative to sentential negation. As shown below, zen 'in 'all' in the object position may have the partial negation inteIpretati on of "not all." "
"
214
"
-
,"
Shige ru M iya gawa (28)
Taroo-ga zen 'in-o
sikari-masen-desita. Taro-Nom all-Ace scold-Neg-Past 'Taro didn't scold alL' not > all (all > not) 9
8
To do the test, it is necessary to set up a situation where it is plausible to choose "all" and negation, and have it be the same mea ning as "no X." Suppose that a product is going to be introduced, and you are trying to determine which of the three colors, red, yellow, and green, the customers would least like, and also the color that they would most like. A test is done with ten people. The following are reported by the tester. First, the example that is irrelevant to our concern. Zen'in-ga kiiro-o erabi-masita. all-Nom yellow-Ace choose-Past 'All chose yellow.'
(29)
The tester also informed you the following. Zen 'in-ga aka-o erabi-masen-desita. all-nom red-Ace ehoose-Neg-Past 'All did not choose red. ' *not > all, all > not
(30)
(3 I ) Midori-oj zen 'in-ga t j erabi-masen-desita. green-Ace j ail-Nom t i choose-Neg-Past 'All didn't choose green. not > all, all > not •
As shown in (30), when "all" is in the subject position in the SOY ord�r, the preferred reading is "all > not" (cf. Kato 1 988). In (3 1 ). we can see that, by scrambling the object to the left edge, partial negation becomes possible 10 {Miyagawa 200 1 ).
a I am using the formal style with
desuldesita to ensure that the sentence is interpreted as a root
cl ause and not embedded. Embedding the sentence leads to possibility of ambiguity in examples that arc not ambiguous in the root clause. See Miyagawa (200 I) for discussion. 9 There
is al so the possibility that the quantifier zen 'in 'aU' may be intetpreted outside the scope of
negation, as indicated by "all > nOl" In Miyagawa (2001 ). I suggest that this is due to the fact that zen 'in can be associated with a group reading, which does not have distributivity. 10
In Miyagawa (200 1) I used examples of the following sort.
(i) Zen'in-ga all-Nom •
tesuto-o
uke-masen-desita.
test-Ace
take-Neg-Past
All did not take the test.'
·not > all, all > not
215
On the EPP
How does negation take scope over "all" to achieve the partial negation interpretation? Following a long tradition starting with Klima ( 1 964), I assume the following.
(3 2)
A quantifier is in the scope of negation iff it is c-commanded by negation (cf. Klima 1964)
In (30). in which the subject "all" is outside the scope of negation. "all" begins in the Spec of vP, but moves to a position outside the c-command domain of negation. A reasonable assumption is that it moves to the Spec of TP as shown below. (The position of negation is roughly as proposed by Laka 1 990, Pollock 1989.)
(ii) Tesuto-oj test-Ace,
zen'in-ga all-Nom
t;
tj
uke-masen-desita. take-Neg-Past
'All didn't take the test' not > all, all > not Most people I consulted shared the judgment given, that
is,
(i) is solely "all > not" while, the OSV
order in (ii) made it possible to get the partial negation. Call this speaker A. There were two other speaker types. Speaker B does not get the partial negation ev�n in (ji). interpret scrambling solely as negation. Speaker type C
A'-scrambling;
is able to get
These speakers prefer to
or, they simply prefer a narrow scope reading of
the partial negation even with the SOY order in (i). I suspect
that the Speaker type C's judgment bas to do, at least in part, with the quality of the data I presented.
Given the situation of giving tests, it is highly implausible to imagine a situation where no student
took the test In addition, if such a situation did exist, it would be more natural to express it with "no student took the test" using the NPI:
daremo tesulo-o uke-masen-desita.
Hence Speaker type C may
simply be compensating for either the implausibility of the situation, Or the unnaturalness of the expression "aU did not take the test" to describe the intended situation. In the examples above in which examples are given in the context of test marketing, I hope to have overcome some of these problems with the data.
216
Shigcru Miyagawa
TP
(33)=(30)
T'
Neg
t. I
� · � v
VP
6
...Object...
In (3 I) in which the subject "aU" occurs in the "scrambled" order of OSV, it is able to be interpreted inside the scope of negation. The simplest assumption to make here is that this subject "all" stays in-situ in the Spec of vP. which is made possible by the movement of the object to the Spec ofTP. (34)=(3 1 )
TP
�
that tes
'
217
On the EPP Ifwe compare the two structures above, there is a simple generalization: (35)
Something must move to the Spec ofTP (e.g., subject, object).
If the subject moves to the Spec ofTP, the object stays in·situ, as in (33). On the other hand, if the object moves to the Spec of TP, this allows the subject to stay in.situ, as shown in (34). As I argue in Miyagawa (200 1 ), the simplest account of what we just observed is that the T in Japanese is associated with the EPP feature. The Spec of TP is filled by something (subject, object, etc.), and this meets the EPP. 1 1 How does the object "escape" vP in (34) to raise to the Spec of TP to satisfy the EPP? One possibility i s that it undergoes Object Shift ("short scrambling") to the edge of vP. It then raises to the Spec of TP to satisfy the EPP on T. In Indo·European languages, if the object were to undergo the same Object Shift to the edge of vP, it would get stuck there, because in IE the EPP can only utilize agreement, not focus, and the agreement would be with the thematic subject. The thematic subject, too, would be stuck, because now it is too far from T owing to the intervening object. As a result the EPP on T goes unfulfilled, and the derivation does not go through. The subject presumably also has an unchecked Case feature, which will get checked only if it moves to the Spec of TP. This, then, excludes the possibility of inserting an expletive to save the derivation. 1 2 4.1 Informational focus and nuclear stress
In this subsection I will make explicit the principal notions related to focus in an informational focus structure, which we just discussed for Japanese. Infonnational focus is signaled by sentential stress. This stress is autonomously assigned by the nuclear stress rule, and the sentence, once assigned the stress, is used by interface systems to relate the sentence to its context (Chomsky 1 97 1 . 1 976; lackendoff 1 972; Neeleman and Reinhart 1 998). Nuclear stress is the central prosodic stress assigned to sentences in neutral contexts (cf. Chomsky and Halle 1 968, Halle and Vergnaud 1 987, Cinque 1993, Selkirk 1 995, Ishihara 2000). In the neutral cases, the interface system uses the focus stress to partition the sentence into topic and focus, creating a theme-rheme infonnational structure. The nuclear stress rule, which is responsible for assigning the. main stress of the sentence that defines the focus in neutral contexts, is assigned by an independent rule. Cinque ( 1993), building on the work of Chomsky and Halle ( 1 968) and Halle and Vergnaud ( 1 987), argues that nuclear stress falls on the phrase located lowest on the syntactic tree. I I Kuroda (1988) was the first to propose that, for elC.ample. the object in Japanese can move to the Spec of TP. For him this is purely an optional movement, but in our approach it is an obligatory fulfillment of the EPP requirement, which can be achieved by moving the object or the subject or some other category into the Spec of TP. See Kitahara (2002) for an analysis of scrambling that also utilizes the EPP feature on T.
II I thank Noam Chomsky for this suggestion aboul lndo.European.
218
Shigeru Miyagawa
(3 6)
Mary read a book yesterday.
In this example the object phrase a book bears the nuclear stress. Note that there are two elements that are lowest in the structure, the object and its verb. The object and the verb are sisters. Cinque suggests that in such a case, what is selected by the other is the lowest, so that, in this example, the obj ect is the lowest because it is selected by the verb. In English the word order is fixed, so that if one wants to focus something other than the lowest element, nuclear stress must be supplanted by a special focus assignment. For example, in the above example, if the speaker wants to indicate that "John" is the focus (what is not presupposed), one has to forego nuclear stress (or it is masked) and invoke a special focus assignment rule that stresses the thematic subject. (37 )
John read a book yesterday.
In the so-called scrambling languages we see a different phenomenon for focus. Scrambling changes word order. By virtue of this, scrambling changes the focus structure of the sentence. Reinhart ( 1 995) and Neeleman and Reinhart ( 1 998) point out that scrambling allows a given phrase to move away from the position of nuclear stress. In the following Dutch example, the phrase that bears the nuclear stress is in bold. (38)
Oat that
Jan langzaam het boek las John slowly the book read
In this example both the object "book" and the verb ''read'' are lowest on the tree since they are sisters. As already noted, Cinque argues that in this situati�n the element that is selected is lower; in the above example, the object "the book" is selected by the verb, hence it is, by definition, lower. and it receives the stress. I ) Dutch allows some flexibility in word order; it is possible to "scramble" the object to the left of the adverb.
11 Gussenhoven ( 1984) (as quoted in Neeleman and Reinhart 1 998) gives confirmation for Cinque's idea from Dutch.
(i) a.
b.
Oat ik op een bankje wacbt that I on a bench wait 'that I am waiting on a bench' ik op een bankje wacht 'that I am waiting for a bench'
Oat
In (a) the locative " on a bench" is an adjunct not selected by the verb, hence, it does not receive nuclear stress, but instead, the stress falls on the verb, which is the lowest element. In (b) �the bench" is the object of "wait," hence selected by the verb, so that the stress falls on the object instead of the verb.
2 19
On the EPP (39)
Oat that
Jan John
het boek langzaam las the book slowly read
Note that now, the nuclear stress falls on the verb because the verb is the lowest in the structure. Reinhart and Neeleman ( 1 998) bring an interesting perspective on this word order flexibility. They note the well-known fact that Dutch scrambling of the type shown in (39) is possible only with definite phrases, and it is linked to some entity in the discourse, i.e., something that is presupposed in the discourse. But what is responsible for the word ordcr permutation in Dutch? Neeleman and Reinhart ( 1 998) argue that both the adverb-object and the object-adverb orders are base generated. However, there is much literature that argues that this type of object shift is movement. Chomsky (200 1 ) suggests that object shift as observed in examples such as (39) is due to the object moving to the edge of the v phase, where it is attracted by the EPP feature on v. This EPP feature is optional, so that object shift itself is optional. The movement here appears to be an agreement prominent one. Only the object may move, which means that the only phrase that can move is the one that agrees with the head v. The EPP is parasitic on this agreement, and if it occurs, it raises the object to vP. But why is it optional? Following Reinhart ( 1 995) and particularly Fox (2000), Chomsky (2001 ) suggests that this kind of optionality is allowed only when the movement has an effect on the output. The effect for object shift is on the focus structure of the sentence as clearly demonstrated by Reinhart and Neeleman. Returning to the Japanese cases, there are several points that distinguish the Japanese case from Dutch. First, in Japanese the "shift" is of the object, or a locative phrase, to a position above the subject. Dutch does not allow the object to shift above the subject in normal cases. Second, there does not appear to be a definiteness effect in Japanese since an indefinite phrase can move above the subject. Third and finally, as we saw earlier, the Spec of TP must be filled, which means that the EPP feature on T is always present. There is no optionality. The Spec of TP may be filled by moving the subject into it, or the object, or some other phrase such as a locative, but something mUst occupy the Spec ofTP. What Japanese shares with Dutch is that, in neutral contexts, what moves into the Spec of TP does not bear the nuclear stress, so that the focus falls on something lower in the structure. The following are taken from Ishihara (2000) (Ishihara assumes V-raising to T). 14
(40)
a.
Taroo-ga Taro-Nom
[vp ho n-o tv] katta. [vp book-Ace tv] bought
b. Hon-oj Taroo-ga book-Accj Taro-Nom
[vp lj [vp lj
tv] katta tv] bought
14 See Bailyn (200 I) for extensive discussion of these sort of issues for scrambling in Japanese and Slavic.
220
Shigeru Miyagawa
(4 1 )
a. Taroo-ga Taro-Nom b.
kyoo today
Taroo-ga Hon-oj book-Ace Taro-Nom
katta. ho n o book-Acc bought -
kyoo
lj
today tj
katta. bought
In (40a), the obj ect hon 'book' predictably receives nuclear stress. In (40b) the object has moved to the Spec of TP, and it is the subject, Taroo, that receives the nuclear stress. As Ishihara notes, this is different from Dutch; in Dutch when the obj ect shifts the verb receives stress. Ishihara argues that this is due to the fact that in Japanese the verb undergoes V-to-I raising (cf. Koizumi 1 995), so that, in the OSV order in (40b), the subject and the verb are the same height, and because the verb (more precisely, v) selects the subject, the subject receives the stress. In (4 I b), the object has moved again to the Spec of TP, and this time, an adverb, kyoo 'today', receives the nuclear stress since it is presumably lower than the subject or the raised verb. Just as in Dutch, nuclear stress gives the possible focus domains of the sentence; these are the domains that, informally speaking, are identified as "new" information. See Ishihara (2000) for a detailed discussion of how Neeleman and Reinhart's system applies to Japanese. B elow I summarize his work. Neeleman and Reinhart (1 998) propose to assign focus domains by their Focus Role. (42)
Focus Rule The focus of IP is a(ny) constituent containing the main stress of IP, as determined by the stress rule (=nuclear stress rule).
Let us look at (40a), repeated below. (40)
a.
Taroo-ga Taro-Nom
[vp hon-o tv] katta [vp book-Ace tv] bought
The focus here is on the object hon 'book', which is the phrase that bears the nuclear stress. According to the Focus Rule, the focus domain of this sentence may be hon, the VP that contains it, or the entire IP. Thus, (40a) can be used as an answer to the following three questions: (43)
a.
h.
c.
What happened? (focus on IP) What did Taro do? (focus on VP) What did Taro buy? (focus on object)
(40b) has a different focus domain set due to the scrambling of the object.
22 1
On the
(40)
Hon-oi
b.
Ta roo ga -
book-Accj Taro-Nom
EPP
ti
t;
tv) katta
tv] bought
The focus domains are the subject NP Taroo and the TP, but the VP cannot be a focus domain because it does not contain the stress. Therefore (40b) cannot be used to answer "What did Taro do?" If we compare (40a) and (40b), we can see that Japanese is a " topic p rominent language in the true sen se of the term. What occurs in the initial position of the sentence is ultimately interpreted as the topic in the informational structure: Taroo in (a) and hon 'book' in (b). In this way what I am arguing for directly reflects the intuition of traditional grammarians such as Mikami ( 1 960) that Jap anese is a topic prominent language, not subject prominent like English. Finally, there is one point I need to note about the "anti-focus" phrase in an inform ational focus structure. Recall from our discussion of identificational focus with indeterminate pronouns that, in a structure that contain s the indeterminate pronoun in, for example, the object position it i s p oss ibl e optionally to raise the subject. The object indeterminate pronoun is in a long-distance agreement with FOe without moving. This kind of long distance agreement never happen s in informational focus structure. What is marked for "anti-focus" must necessarily raise to Spec of TP. This is logical. If something other than this phrase raises to the Spec of TP, the structure would end up with two '·topics," something not allowed at least in Japanese (cf. Kuno 1973). "
,
-
5.
Kinande (Baker 2003)
this section I will look at Kinande, a Bantu l anguage 1 w ill depend on the Baker (2003). I am going to assume that Kinande is a focus prom inent language. However, Kinande also has agreem ent, in fact quite an elaborate agreement system that allows agreement between not only the subject and the verbal affix but also with the obj ect and with a locati ve (and p ossib ly other arguments). What I will propose is that while Kinande is a focus prominent language, it also manifests agreement on C, similar to West Flemish and Turkish that we saw at the stan of this article. Kinande falls together with Japanese in b ein g a focus promirient lan guage despite the appearance of agreement because Kinande shows a property that is not shared by the agreement prominent languages of Indo-European. As noted by. Baker, the agreed-with category must be interpreted as definite/speci fic . This is illustrated for the object below. In
.
insights of
(44)
Eritunda n a-ri-gul-a fruit.5 1 SG.S-T-OMS-buy-FV 'The fruit, I bought it ,
-
.
.'
In this "reversal construction," the object is in a position to trigger agreement on the verb and it must be interpreted as s pecific/defini te, i.e., the topic. Baker notes that " [t]rue polysynthetic languages . . . always have agreement and always ,
222
Shigeru Miyagawa have dislocation" (2003 : 1 1 2). What Baker means by "dislocation" is that he views the agreed-with category as ultimately being somewhere above the "basic" Spec of TP - he assumes that it is in the higher Spec of TP in a position comparable to the dislocation position in languages such as Italian. Dislocation in these languages is only possible if the phrase is not indefinite/nonspecific (Rizzi 1 986). The agreement, therefore, occurs in a specifier higher than the normal S pecifier of TP. Baker (2003: 1 09) expresses this as a biconditional for languages such as Kinande (and Mohawk, etc.). -
(45)
A verb X agrees with an NP Y ifand only if Y is in a dislocated, adjunct position.
How is the agreed-with phrase " dislocated"? Baker forces the agreed-with phrase to occur in a higher position - thus the agreement to hold between this higher specifier and some head - by requiring pro to occur in the "normal" Spec ofTP for Case reasons (see his article for details) (Ibid., p. 124). (46)
[yp NPj
[TP proj T+Verb . . . [vp ti . . . ]]]
The occurrence of pro in the lower Spec of TP forces the agreed-with phrase, NPj, to be in the higher specifier where the agreement takes place according to the biconditional in (45). What I wish to key in on is Baker's intuition that the DP triggering agreement in Kinande must occur in a higher specifier position than the normal IS Specifier of TP, which he calls the "dislocated" position. In our system, we have a way to capture this without having to postulate the pro in the lower Spec ofTP. A natural way under our approach is to postulate the following structure.
(47)
Kinande, etc. CP
�.
�/FOCU1
T
�
TEPP
AGREEMENT
�
IS
Baker's system is designed, in part, to account for the difference between IE languages such as Greek (Alexiadou and Anagnostopoulou 1 998). which appears not to have the EPP on T, and languages such as Kinande. I will not go over this portion of Baker's analysis.
223
On the EPP
As a focus prominent langua'ge, the focus feature on FOC probes an "anti-focus" phrase, and the EPP feature on T raises it to the Spec of TP. This leaves agreement on C. Here. let us make the following " design" stipulation. (48) Agreement or focus must occur in the context of EPP. Agreement, for example, may occur at C, and the EPP on T. But if the EPP on T is satisfied by focus, the agreement on C must come with its own EPP feature, which requires that a category be raised to C. 16 The closest one is the category already raised to the Spec ofTP by EPP/focus, and it is raised to the Spec of ep, where agreement takes place. By assuming that Kinande is a focus prominent language, but with also agreement in the higher node (C), we capture Baker's intuition embodied in his biconditional in (45) which states that the agreement holds of a specifier higher than the basic Spec of TP. Below, 1 will present some of the major data from Baker (2003). Baker observes that Kinande allows non-subjects as well as subjects in the Spec of TP, and the verb agrees with whatever occurs there. In the following pair, the subject occurs in the Spec of TP in the first example, and the object in the second example. (49)
a.
Omukali mo-a-seny-ire okukwi (lw' -omo-mbasa). woman.l AFF-LSIT-chop-EXT wood. I I LKl l-LOC. 1 8-axe.9 'The woman chopped wood (with an axe).'
b.
Olukwi
si-lu-li-seny-a bakali (omo-mbasa). wood. l l NEG-I I .S-PRES-chop-FV women.2 LOC. l S-axe.9 'WOMEN do not chop wood (with an axe).'
In (a) the subject occurs in the Spec of TP and the agreement is with this thematic subject. In (b) the object occurs in the Spec of TP, and the verb agrees with this object. The translation for (b) indicates that this construction, the so called "reversal construction," "expresses contrastive focus on the thematic subject" (Ibid. p. I t 3). This is telling of the purpose of the reverse construction. Although Baker does not pursue this notion of contrastive focus, it seems to be an indication that this effect comes at least in part from removing the object from the position of sentential nuclear stress, which is the region naturally interpreted as focus (new information). By moving the object out of this position, the subject gets the focus reading (new information). This parallels what we observed in Japanese, and it is a sign that Kinande is focus prominent. Auother kind of inversion i s with the locative expression. This construction is found with unaccusatives and passives of transitives.
16 Alternatively. the EPP feature could be merged on FOe, so that the category in the Spec of TP
needs to raise just to the Spec of FOCP. This will
be identical in modus operandi to the EPP on T. I
will not attempt to argue for one or the other, but will assume that the category moves to the Spec of CP in the exposition.
224
Shigeru Miyagawa
(50)
?Omo-mulongo mw-a-hik-a omukali. LOC. I S-village.3 1 8.S-T-arrive-FV woman , At the village arrived a woman.'
Unlike in English. in Kinande locative inversion leads to the verb agreeing with the inverted locative expression. It is important to note that although English has a similar construction of locative inversion (e.g., Bresnan and Kinerva 1 989) , the locative PP in English is not in the same position. It is not in the "basic" Spec ofTP. We can tell this easily by the fact that the locative PP does not allow Aux in version in questions, one of many pieces of evidence for this fact. 17
(5 1 ) *Whoj was on the wall hung a picture ofti? (52)
Who was there a picture oflj on the wall?
(5 1 ) indicates that the auxiliary was cannot invert across the locative PP on the wall, suggesting that this locative PP is positioned somewhere other than in the Spec of TP. possibly adjoining to the Spec ofTP. In (52), we see that there is no problem in inverting the aux was across the expletive there, which arguably is in the Spec of TP. That the Kinande locative in (50) enters into agreement with T indicates that the locative inversion in Kinande is fundamentally different from English. In Kinande. locative inversion moves the locative phrase into the Spec of TP. This difference is in part due to the fact that in Kinande the locative phrase is a DP. but in English it is a PP; P in English does not enter into agreement. In all of these cases, Baker argues that the phrase that is "agreed-with" must occur on the left edge, where it is interpreted as definite/specific. One piece of evidence for this has to do with "augment vowels." Nouns often begin with an "augment" vowel that matches the vowel of the class that the noun prefix belongs to. This augment vowel may drop under the scope of negation and in some other contexts; a noun without this augment vowel has a nonspecific, indefinite interpretation. The following exemplifies an object with and without its augment vowel. (53)
a.
Yohani si-a-nzire o-mu-kali. NEG- l .Srr-l ike AUG-CLI -woman John 'John does not like the woman.'
b.
mu-kali. Yohani si-a-nzire NEG-I.Srr-like eLl -woman John 'John does not like a(ny) woman.'
A point relevant to our discussion is that the agreed-with phrase can never drop its augment vowel. Because the agreed-with phrase only occurs on the left edge, 17 1 thank Noam Chomsky (personal communication) for pointing oul this fact.
225
On the EPP
the agreed-with phrase is always specific/definite. By a transitivity of reasoning, a nonspecific, indefinite phrase can only occur in a lower position, probably in their originally merged positions within vP/VP. In the first example below the subject has the augment vowel, while in the second example it does not.
(54)
ki-ndu. Omukali mo-a-teta-gul-a AUG-CL. I -woman AFF- l.S-NEG/PAST-buy-FV CL7-thing 'The woman didn't buy anything. •
a.
b.
* Mu-kali mo-a-teta-gul-a eritunda. CL l -woman AFF-1.Srr-NEGIPAST-buy-FV fruit.S 'No woman bought a fruit. •
As shown in (b), it is not possible for a phrase without the augment vowel, which is nonspecific and indefinite, to occur in the Spec ofTP. What we can observe by using the augment-vowel phenomenon is that in Kinande, something that moves to the left edge is marked for being the specific/definite topic of the sentence and also for agreement. Being just in the Unormal" Spec of TP does not force a category to be interpreted as specific/definite, as we can see from English and Japanese. This added interpretation arises from the category moving to a higher node, which I assume is the Spec of CP (or the Spec of FOCP), which is a position that the interface system interprets as specific/definite. This movement to the higber specifier is forced by the occurrence of agreement in a focus prominent language. What Kinande demonstrates is that a language that is focus prominent can also have agreement. It is just that this agreement raises a category to a higher specifier than the TP because agreement occurs on C (universally), and the EPP occurs with it.
5.1 Turkish
-
further evidence
Turkish provides a particularly dramatic demonstration of what we just saw for Kinande. The following from Ozturk (2003) shows the effect of adding agreement (example (b».
(55)
a.
Butun �ocuk-Iar dergi-yi sev-me-di. all child maga.zine-Acc love-not-past • All children didn't like the magazine. • not > all
b.
Biltun �ocuk-Iar dergi-yi sev-me-di-Ier. all magazine-Ace love-not-past-pl child 'All children didn't like the magazine.' *not > all, all > not
As Ozturk points out, without the plural agreement suffix -fer in (a), the universal quantifier in the subject position is interpreted as being in the scope of
226
Shigeru Miyagawa
negation, not > all. But if the plural suffix is added as in (b), the universal quantifier which the verb now agrees with can only be interpreted as being outside the scope of negation. Ozturk analyzes the universal quantifier here as being higher than negation. Setting aside the details of her analysis, what this shows is that the presence of agreement can raise a category higher than when the agreement is absent. In Kinande, the agreement on C raises the category in the Spec of TP to a higher specifier where it receives the definite/specific interpretation. This is the "dislocation" effect Baker notes. In Japanese, another focus prominent language, agreement in the traditional sense does not occur on C (but see below for something else that in effect instantiates "agreement" in Japanese), hence the category raised to the Spec of TP stays there, thus it need not be interpreted as definite/specific. Other languages that have this property of specific/definite include Hindi (cf. Mahajan 1 990) and Tagalog (cf. Rackowski 2002 and references therein) among others. 6. Wh-movement vs. Wh-in-situ
In this section I will show that the theory of the EPP I have outlined can predict with a measure of precision the presence/absence of overt wh-movement in a given language. The presentation will have to be brief due to limitation of space. In Miyagawa (200 I) I suggested that one can predict when a language requires overt wh-movement, and when it doesn't. In that work, I suggested that the feature [+wh] in overt wh-movement languages occurs on C, while it occurs on T in non-wh-movement languages. I argued that when it occurs on C, there is no option but to move a wh-phrase to the Spec of CPo However, when occurring on T. the [+whJ feature may be satisfied without moving the wh-phrase to the Spec of TP. because the T also has other features, e.g., features that go with a non-wh phrase category. If one moves such a non-wh category, the EPP on T is met, and there is no reason to move the wh-phrase. This allows the wh-phrase to stay in situ. Given the theory of the EPP outlined in this article, we can now refine the notions involved. We can make a simple statement, as follows.
227
On the EPP
(56)
Overt wh-movement: when it must occur, and when it does not Overt whwmovement to the Spec of CP only occurs in agreement prominent languages. The movement occurs when, along with agreement, FOe is projected due to the presence of a wh-phrase.
The structure below illustrates this for a sentence with an object wh-phrase. (I am going to somewhat arbitrarily assume that FOe in an argument prominent language occurs above C.) (57)
Overt wh-movement CP
�"
�
T
�
SUB whOBJ . . . TEPP
<
mcw
AGREEMENT
�
Being that this is an agreement prominent language such as those of Indo European, the agreement feature on C works in tandem with the EPP on T. This raises the agreed-with phrase, in this case the thematic subject, to the Spec of TP. This takes care of the agreement and the EPP on T. But there is the focus feature, which has been merged into the structure because of the presence of the object wh-phrase, which has a focus feature. The focus feature requires (another) EPP, because of the requirement that focus works in tandem with the EPP. Suppose that this EPP is merged at C. The focus feature agrees with the focus feature on the wh-phrase, and in tandem with the EPP raises the wh-phrase to the Spec of CP. Hence, what raises the wh-phrase is the focus feature that is not checked as part of satisfying the EPP on T. It is not checked because this is an agreement prominent language. Now consider a focus prominent language. Take Kinande, for example. (58)
Kinande, etc. CP
----......
�AG
�
SUB whoBJ- . . TEPP
�<
.. EMENT
FOCUS
-----. -.J
_________
Given that this is a focus prominent language, the focus feature works in tandem with the EPP on T. Because this is a wh-construction, the focus feature "agrees"
228
Shigeru Miyagawa
with the wh-phrase focus. Recall from our discussion of the identificational focus structure in Japanese that it is possible to have long-distance agreement, just as in the existential construction in English. In such a case, some other phrase meets the EPP on T. The following is such an example from Kinande (Baker 2003: 1 1 2).
(59)
Kambale
a-gul-a
I .srr-buY-FV 'What did Kambale buy?'
Kambale
ebihi? what. 8
Here the subject "Kambale" raises to the Spec ofTP, then to the Spec of CP (or FOCP) for agreement. The wh-phrase stays in-situ, since there is no need for it to raise under this "long-distance" agreement. As Baker points out. a wh-phrase cannot occur in the agreement position because that position requires the phrase
in it to be definite/specific.
In languages such as Japanese, there is no overt agreement (but see next section), hence once focus is satisfied in tandem with the EPP on T by 10ng distance agreement with the wh-phrase, nothing more happens, and there is no
instance in which overt wh-movement is required. It is. of course. possible for
the wh-phrase to move to the Spec of TP to satisfy the EPP. This is similar to
overt wh-movement in agreement prominent languages, but, as I noted in Miyagawa (200 I ), unlike in an overt wh-movement language, it is not necessary for the wh-phrase to be picked to move in this way to the Spec of TP in a language such as Japanese. What must be satisfied is the EPP on T, and there are
a number of options including the wh-phrase to accomplish this. Finally, Turkish, which we also saw is a focus prominent language by virtue of the fact that it need not have agreement, also does not have overt wh movement as expected even when there is agreement Our approach predicts that when a language has no agreement, iJ does not have wh-movement because it is a focus prominent language. Kuroda ( 1 988)
(and also Fukui 1 986), too, makes the same prediction. What differentiates ours from his system is that he predicts that when there is agreement, there should be wh-movement. In our system, a language with agreement only requires wh movement if the language is agreement prominent, as in the case of IE languages.
7.
18
Where is the Agreement in Japanese?
At the beginning of this article I said that I will assume a strong version of the Uniformity Principle (Chomsky 200 1) for the inflectional system. This means that features such as focus and agreement should be present uniformly in every language.
• 8 Norvin Richards has pointed out to me that the Nonh Gennanic languages might be a problem for
the claim that nuU-agreement languages always have wh·in-situ-Norwegian, for example, basically
doesn't have subject-verb agreement, and does have wb-movement. On the other hand, dtese are V2
languages. so possibly V2 is another way for EPP to
be expressed, not directly related to agreemenl.
229
On the EPP
Uniformity Principle (Chomsky 200 1 ) I n the absence of compelling evidence to the contrary, assume languages to be unifonn, with variety restricted to easily detectable properties of utterances.
(60)
We saw that Kinande and Turkish, both focus prominent languages, also have agreement. We also saw that English (in fact all of IE), is agreement prominent, but the focus feature is also instantiated (at least) in the wh-construction. That leaves Japanese. What instantiates agreement in Japanese'? Japanese has no overt agreement. I suspect that a possible candidate for "agreement" is the topic construction, which in Japanese is typically overtly marked with the topic marker -wa. hon-o Taroo-wa kana. Taro-Top book-Acc bought 'As for Taro, he bought a book.'
(6 1 )
Parallels with agreement include the fact that there can only be one true topic per sentence (Kuno 1 973). Also, the topic phrase, which typically occurs on the left edge, occurs high in the structure. Whitman ( 1 991 ) suggests that it is in the region of the Spec of CPo If this is correct, it is precisely in the specifier of the head, C, that I have argued is the locus of agreement. Finally, topics do not occur in most subordinate clauses such as a relative clause (Kuno 1 973). (62)
'"
hon-o yonda. Taroo-ga [Hanako-wa katta] Taro-Nom [Hanako-Top bought] book-Acc read Taro read the book that Hanako bought. '
While other explanations may be possible, foHowing the idea that topic is "agreement," it is possible to correlate the lack of topic in subordinate clauses to the lack of agreement in subjunctive clauses. Uchibori (2000) gives compelling evidence that a variety of subordinate clauses in Japanese behave like subordinate clauses in languages where the subjunctive is well established, e.g., Romance. The indicative/subjunctive distinction in Japanese, i f real , is not represented by verbal inflection. It corresponds approximately to the conclusive/attributive distinction in Classical Japanese of some one thousand years ago, which was represented by distinct inflections (cf. Miyagawa and Ekida 2003 and references therein). The conclusive fonn typically appeared in root clauses, while the attributive fonn appeared most commonly in subordinate clauses. There was one instance in which the conclusive fonn appeared in a subordinate clause; this is when the subordinator is the complementizer -to. Because the conclusive fonn in Classical Japanese correlates with the indicative fonn in present-day Japanese, if Uchibori 's analysis is right, we predict that the
230
Shigeru Miyagawa
subordinate clause with the following is sueh an example. (63)
-to
complementizer should allow a topic. The
Hanako-ga [Taroo-wa kuruma-o katta to] omotteiru. Hanako-Nom [Taro-Top car-Acc bought C] think 'Hanako thinks that as for Taro, he bought a car.'
There may be better examples to demonstrate this point, but even with this example it contrasts sharply with "subjunctive" clauses such as below. (64) * Hanako-ga [Taroo-wa kana] kuruma-o Hanaka-Nom [Taro-Top bought] car-Ace 'Hanako saw the car that Taro bought.'
mita. saw
While speculative, and with a number of problems remaining (e.g., what to do with sentences that do not have a topic phrase with -wa?), it does make a prediction. Namely, we should only find a robust topic structure in languages that do not evidence any overt agreement. At . least this is true for Chinese, Japanese, and Korean. Further study is necessary to see whether this hypothesis of "topic-agreement" holds up. 8.
Concluding remarks
If what I have outlined in this article is anywhere near on target, we have a view of languages in which all the features of universal grammar are uniformly present in every language, at least for the domain of inflection which we explored. This is what Sigurdsson (2003) suggests. This idea is based on the notion that the universal stock of features is uniform (Chomsky 2001 ). I took the strong version of this to be true and hypothesized that these features appear in every language in the domain of inflection. The differences among languages, which appear vast on the surface, as in the case of overt wh-movement languages versus languages that do not move the wh-phrase, or between "scrambling" and "nonscrambling" languages, are the result simply of varying the way some of the features interact with each other. It does not reflect some deep difference among languages.
References Ackema, Peter and Ad Neeleman (200 1 ). Context-sensitive spell-out and adjacency. Ms., Utrecht University and University College London. Alexiadou, A. and Elena Anagnostopoulou ( 1 998). Parametrizing word order, V movement, and EPP-checking. Natural Language and Linguistic Theory 1 6, 491 - 539. Alexiadou. A. and Elena Anagnostopoulou (2001 ). The subject-in-situ generalization and the role of case in driving computations. Linguistic Inquiry 32, 1 93-232. Bailyn. John (20 0 1 ). On scrambling: a reply to Boskovic and Takahashi. Linguistic Inquiry 32. Bailyn. John (2003). Does Russian scrambling exist? In Simin Karimi. eds., 1 56- 1 76.
23 1
On the EPP
Carnie, Heidi Harley. and M ary Willie (eds.), Formal approaches to function in
grammar: in honor of Eloise Jeli nek, Amsterdam/Philadelphia: John Benjamins, pp. 1 07- 1 34.
Baker, Mark (2003). Agreement, dislocation, and partial configurationality. In Andrew Carnie, Heidi Harley, and MaryAnn Willie, eds., Formal approaches to function in grammar, pp. 1 07 - t 32. John Benjamins Publishing Company. Boeckx, Cedric (2003). is lands and chains: resumption as stranding. John Benjamins Publishing Company. Bresnan, Joan, and 1. Kanerva ( 1 989). Locative inversion in Chichewa: a case study of factorization in grammar. Linguistic Inquiry 20, 1 -50. Brody, Michael ( 1 990). Some remarks on the focus field in Hungarian. VCL working papers 2:201 -225. Carstens, Vicki (2003). Rethinking complementizer agreement: agree with a case checked goal. Linguistic Inquiry 34, 393-4 1 2. Chomsky, Noam ( 1 97 1 ). Deep structure, surface structure and semantic interpretation. In D. Steinberg and L. Jakobovits (eds.), Semantics. an interdisciplinary reader in
philosophy, linguistics and psychology. Cambridge University Press. Chomsky, Noam ( 1 976). Conditions on rules of grammar. Linguistic analysis 2:303-349. Chomsky, Noam ( 1 9 8 1 ). Lectures on government and binding. Dordrecht: Foris Publications. Chomsky, Noam ( 1 993). Minimalist program for linguistic theory. In Kenneth Hale and Samuel Jay Keyser (eds.), The view from B U ilding 20, Cambridge, Mass.: MIT Press, I -52. Chomsky, Noam (2000). Minimalist inquiries: the framework. In Essays on minimalist syntax in honor of Howard Lasnik. Roger Martin, David Michaels, and Juan Uriagereka (eds.), 89- 1 56. Cambridge. Mass.: MIT Press. Chomsky. Noam (200 t ). Derivation by phase. In Michael Kenstowicz (ed.), Ken Hale: a life in language. Cambridge, MA: MIT Press, pp. 1 -52. Chomsky, Noam (to appear). Three factors in language design. Linguistic Inquiry. Chomsky, Noam and Morris Halle ( 1 968). The sound pattern of English. New York: Harper & Row, Publishers. Cinque, Guglielmo (1 990). Types ofA '-dependencies. Cambridge, Mass.: MIT Press. Cinque, Guglielmo ( 1 993). A null theory of phrase and compound stress. Linguistic . Inquiry 24: 239-298. Cinque, Guglielmo ( 1 99 9). Adverbs and functional heads. Oxford: Oxford University Press. Culicover, Peter and Michael Rochemont ( 1 983). Stress and focus in English. Language
59, 1 23- 1 65. Dayal, Veneeta (2003). Bare nominals: non-specific and contrastive readings under scrambling. In Karimi Simin, ed., 67-90. E. Kiss, Katalin ( 1 998). Identificational focus versus informational focus. Language 74: 245-273. E. Kiss, Katalin (2003). Argument scrambling, operator movement, and topic movement in Hungarian. In Simin Karimi, ed., 22-43. Fox, Danny (2000). Economy and semantic interpretation. Cambridge. Mass.: MIT Press. Fukui. Naoki ( 1 986) . Theory of category projections and its theoretical implications. Doctoral dissertation•. MIT, Cambridge. Mass. Fukui, Naoki ( 1 99S). The principles-and-parameters approach: A comparative syntax of English and Japanese. In Masayoshi Shibatani and Theodora Byron (eds.), Approaches to language typology. Oxford: Clarendon Press. Gussenhoven, C. ( 1 984). On the grammar and semantics ofsentence accents in Dutch. Dordrecht: Foris. Haegeman. Liliane ( 1 992). Some speCUlations on argument shift, clitics and crossing in West Flemish. Ms .• University of Geneva.
232
Shigeru Miyagawa Halle, Morris and Jean-Roger Vergnaud (J 987). An essay on stress. Cambridge, Mass.: MIT Press. Hiraiwa, Ken (2002). Indetenninate-agreement and varieties of rais ing. Ms., MIT. Holmberg, Anders ( 1 999). Remarks on Holmberg ' s generali zati on. Studia iinguistica 53: 1 -39. latridou, Sabine ( 1 993). On nom i native case assignment and a few related things. Papers on case and ugreemenl Il, Colin Phi llips (ed.), MIT working papers in linguistics 19: 1 75- 1 96. Ishihara, Shinichiro (2000). Stress. focus, and scrambling in J apanese. MIT working papers in Iinguislics 39, 142-1 75. Jackendoff, Ray ( 1 972). Semantic inlerpretation in generative grammar. Cambridge, Mass.: MIT Press. Jelinek. Eloise ( 1 984). Empty categories, case, and configurationality. Nalural Language and Linguislic Theory 2, 39-76. Jung, Yeun-Jin (2002). Scrambling, edge effects, and AlA '-distinction , The Linguistics Association 0/Korea Journal 10.4. Karimi, Simin, ed., (2003). Word order and scrambling. Blackwell Publishing . Kato, Yasuhiko ( 1 988). Negation and the discourse-dependent property of relative scope in Japanese. Sophia Linguistica, 23-4: 3 1 -37. Kishimoto, Hideki (200 I ). B inding of indetenninate pronouns and clause structure in Japanese. Linguislic Inquiry 32, 597-634. Kitahara, Hisatsugu (2002). Scrambling, case, and interpretability. In Samuel David Epstein and T. Daniel Seely (eds.), Derivation and explanation in the minimalist program. Oxford: Blackwell Publishing, 1 67- 1 83. Klima, Edward ( 1 964). Negation in English. In Jerry A. Fodor and Jerrod Katz (eds. ) . The slructure of language. Englewood Cliffs, New Jersey: Prentice-Hall . pp. 246- 323 . Koizumi, Masatoshi ( 1 995). Phrase structure in minimalist syntax. Doctoral dissertation, MIT, Cambridge. Mass . Korntlit, Jaklin (2000). Some syntactic and morphological properties of relative clauses in Turkish, in A. Alexiadou, C. Widler, and P. Law (eds), The syntax olrelative clauses. AmsterdamlPhildelphia: John Benjamins, 1 2 1 - 1 59. Komfilt, Jaklin (2004). Unmasking covert complementizer agreement. Presented at the Linguistic Society of America conference, Boston, Mass., January 1 0, 2004. Kuno, Susumu ( 1 973). The structure o/ the Japanese language. MIT Press, Cambridge, Mass. Kuroda, S.-Y. ( 1 965). Generative grammatical studies in the Japanese language. Doctoral dissertati on, MIT, Cambridge, Mass. Kuroda. S.-Y. ( 1 972). The categorical and the thetic judgments. In Foundations 0/ language 9, 1 53- 1 85. Kuroda, S.-Y. ( 1 988). Whether we agree or not: a comparative syntax of English and Japanese. Linguisticae Investigationes 1 2, 1 -47. Kuroda, S.-Y. ( 1 992). Judgment forms and sentence fonns. In Japanese syntax and semantics: collectedpapers, 1 3-77. Dordrecht: Kluwer Academic Publishers. Laka, I. 1 990. Negation in syntax: on the nature offunctional categories and projections.
Doctoral dissertati on, MIT, Cambridge, Mass.
Lasnik, Howard ( 1 999). Chains of arguments. In Samuel Epstein and Norbert Hornstein (eds.), Working minima/ism. Cambridge, Mass.: MIT Press. McCloskey, James (200 1 ). On the distribution of subject pro perties in Irish. In Objects and other subjects, ed. William D. Davies and Stanley Dubinsky. Dordrecht: Kluwer Academic Publishers, 157-1 92. Mahajan, Anoop {I 990). The AlA-bar distinction and movemenl theory. Doctoral dissertation, MIT, Cambridge, Mass. Mikami. Akira ( 1 960). Zoo-wa hana-ga nagai. Tokyo: Kurosio Shuppan.
233
On the EPP Miyagawa, Shigeru ( 1 989). Structure and case marking in Japanese. New York: Academic Press.
.
Miyagawa, Shigeru (200 1 ). EPP, sc rambling. and wh-in-situ. In Michael Kenstowicz (ed.), Ken Hale: A Life in Language. MIT Press, pp. 293-338.
Miyagawa, Sh igeru (2003). A-movement scrambling and options without option ality . In Sim in Karimi (ed.>, Word order and scrambling. Blackwell Publishers. Miyagawa, Shigeru, and Fusae Ekida (2003). Historical development of the accusative case marking in Japanese as seen in classical literary lexts. A special issue of
Journal 0/Japanese Linguistics 1 9.
Neeleman, Ad and Tanya Reinhart (1 998). Scrambling and the PF interface. In Miriam Butt and Wilhelm Geuder (eds.), The projection of arguments; lexical and compositional/actors, Stanford, California: CSLI Publications, pp. 309- 353. N i shigauch i , Taisuke ( 1 990). Quantification in the theory of grammar. Kluwer, Dordrecht. Ordoiiez, Francisco. ( 1 997). Word order and clause structure in Spanish and other Romance languages. Doctoral dissertation, CUNY. Ozturk, Balklz (2003). Configurationality: a typology based on case and referentiality. Ms., Harvard University. P ap er presented at CASTL, Trsmsa, Norway. Pesetsky, David, and Esther Torrego (200 1 ). T-to-C : causes and consequences. In Michael Kenstowicz (ed.), Ken Hale: a life in language. MIT Press. Pollock, J.-Y. 1 989. Verb movement, universal grammar, and the structure of IP.
Linguistic Inquiry 20:365-424.
Rackowski, Andrea (2002). The Structure of Tagalog: Specificity, Voice, and the Distribution of Arguments. Doctoral dissertation, MIT, Cambridge, Mass. Reinhart, Tanya (1995). Interface strategies. OTS working papers, Utrecht University. Rizzi, Luigi ( 1 986). Null objects in Italian and the theory of pro, Linguistic Inquiry
1 7:50 1 -557.
Rizzi, Luigi ( 1 997). The fine structure of the left periphery. In Liliana Haegeman ( ed. ), Elements of grammar: handbook in generative syntax. Dordrecht: Kl uwer Academic Publishers, pp. 281-337. Rochemont, Michael (1 986). Focus in generative grammar. Amsterdam: Benjamins. Saito, Mamoru ( 1 992). Long distance scrambling in Japanese. Journal of East Asian
Linguistics 1 :69- 1 1 8.
Selkirk, Elizabeth ( 1 984). Phonology and syntax: the relaiion between sound and
structure. Cambridge, Mass.: MIT Press. Selkirk. Elizabeth (l 995). Sentence prosody: Intonation, stress, and phrasing. In John A. Goldsmith (ed.), Handbook of phonological theory. London: Basil Blackwell, pp. 550-569. Shibatani, Masayoshi ( 1 990). On parametric syntax. In Festschrijt for Professor Hisao
Kakesh; s Sixtieth Birthday. Tokyo: Kurosio, pp. 397-420. Sigurdsson, Halld6r Armann (2003). Meaningful silence, meaningless sounds. Ms., Lund University, Sweden. Simpson, Andrew and Zoe Wu (200 1 ). Agreement, shells. and focus. Language 78: 287313.
Szabolcsi, Anna. ( 1 98 1 ). The semantics o f topic-focus articulation. In Jan Groenendijk, Theo Janssen, and Martin Stokhof (eds.), Formal methods in the study 0/ language, Amsterdam: Matematisch Centrum, pp. 5 1 3-4 1 . Tada, Hiroaki ( 1 993). A/A ' partition i n derivation. Doctoral dissertation, MIT, Cambridge, Mass. Uchibori, Asako (2000).The Syntax ofsubjunctive complements: evidencejrom Japanese. Doctoral dissertation, University of Connecticut, Storrs, Conn. Watanabe, Akira, ( 1 993), Agr-based case theory and its interaction with the A-bar system. Doctoral dissenation, MIT, Cambridge, Mass.
234
Shigeru Miyagawa
Webelhuth, Gert ( 1 989). Syntactic saturation phenomena and the modern Germanic languages. Doctoral dissertation, University of Massachusetts. Amherst. Whitman, John ( 1 99 1 ). String vacuous V to Compo Paper presented at GLOW. Yamashita. Hideaki (200 I ). EPr and the ordering effects on interpretation: a preliminary study. Nanzan studies in Japanese language education 8. 300-338. Nagoya: Nanzan University. Yang. Dong-Whee (2004). Scrambling, interpretive complex, and cyclic spell-out. A paper presented at the Linguistic Society of America, Boston, Mass.
Shigeru Miyagawa Department of Linguistics and Philosophy 32-D886 MIT Cambridge, MA 02 1 39 USA [email protected]
235
Transparent parsing: phases in sentence processing* Iris
Mulders University
Utrecht
This paper discusses the role of phases in reanalysis during sentence pro cessing. First, it is shown that whereas reanalysis in syntactic processing is very difficult to perfonn in some cases, it is automatic in other cases. This calls for a constraint on reanalysis. It is argued that on conceptual grounds, we expect reanalysis to be impossible across phase boundaries. It is shown that indeed reanalysis is possible within phases, and that reanalysis across phase boundaries can only take place from the phonological border of a CP or VP-phase. Finally, it is investigated whether nominal projections constitute boundaries for reanalysis.
I.
The Basic Problem: Constraining Reanalysis in Sentence Processing
The goal ofthis paper is to convince theoretical linguists, especially syntacticians, that sentence processing is a field ofstudy which is directly relevant for their own work. My intention is to demonstrate that syntactic processing is not at all remote to syntax, but is actually governed by familiar principles. Therefore, sentence processing should not be dismissed by syntacticians as 'interesting but a different baUgame'; to a large extent, it is the same baUgarne as syntax. The empirical domain that this paper is concerned with consists of so called garden path sentences. Garden path sentences are sentences that are difficult to process due to a temporary ambiguity. A well-known example is ( 1 ): i. After Susan drank the water evaporated.
(I)
The element the water in (1) is temporarily ambiguous, because at the point in time where the parser encounters it, it could in principle be analyzed in two positions: in the subject position ofthe main clause (which would be the correct analysis for (1», or in the direct object position of the embedded clause, as in the minimally different (2): (2)
After Susan drank the water she disappeared.
It is generally accepted that the processing difficulty associated with ( 1 ) arises because the parser analyzes the water as the direct object of the preceding verb drank, rather than as the subject of a predicted main clause. One might attempt to account for the contrast in ease of processing between ( 1 ) and (2) along the following lines. In a string like After Susan drank the water.. .. , the parser initially analyzes the water as the direct object of drank. This analysis happens to be correct in (2), but it is not correct in (1); the initial analysis has to be revised in (J-) when the parser encounters evaporated: •
Many thanks to Kriszta
paper.
Szendroi and Eddy Ruys for their comments on an earlier version of this
49, 23 7-264. 2005 Jris Mulders
MIT Working Papers in Linguistics Perspectives on Phases.
©
Phases in sentence processing
i , � " !
(3)
TP
�
PP
.L oA. I
Susan
�
T
Jib.
TP
�
�
VP
o
Ap I
V evaporated
I drank
One might hypothesize that this reanalysis is costly for the parser, and that this is why (2) is so difficult to process. However. things are more complicated than this: not every instance of reanalysis leads to processing problems. To see this, consider the sentences in (4): (4)
a. b.
John believed Mary. John believed Mary came.
Note that (4b) does not induce any noticeable processing difficulty, compared to ( I ). It is important to note that the difficulty in processing ( I ) is very strong: native speakers of English tend to get rather confused when they are confronted with sentences like ( l ). This psychological state of confusion is indicated here with i. No such confusion arises when processing (4b). Nevertheless, (4b) does display a temporary ambiguity s imilar to (1): at the point it is being processed, Mary could be either analyzed as the subject of ·the verb following it, or as the object of the verb preceding it. The parser makes the same decision as in ( 1 ), and attaches Mary as a direct object of believed. This means that Mary will have to be reanalyzed as the subject 'of came when came is encountered. But this in tum means that our account for the processing difficulty in ( 1 ) cannot be as simple as stated above: it can'tjust be that reanalysis in itself leads to conscious processing problems: reanalysis leads to problems i n ( 1 ), but not in (4b). So, we can conclude that reanaly�is is possible in certain environments, but not in others. In other words, it is constrained. The fonnulation of the conditions under which reanalysis can occur and under which it cannot occur, is thus a topic for psycho linguistic study. The importance of intuitive processing judgments is emphasized in the work of Pritchett ( 1 992). Pritchett gives numerous minimal pairs of temporary ambiguity which do lead to a conscious processing problem in some structural environments, but not in others . Some of these examples are given below: (5)
a.
b.
238
Object-Subject Ambiguity John bel ieved the ugly little man hated him. i.. After Susan drank the water evaporated Double Object Ambiguity Rex gave her presents to Ron i.. Todd gave the boy the dog bit a bandage.
Iris Mulders
c.
Lexical Ambiguity i.. The old train the children. The church pardons many sinners.
(Pritchett 1 992: ex.22a,d,e)
In order to account for these types of facts, Pritchett argues for a serial processing model which consists of I . a theta-based strUcture·building component, which attaches incoming material in maximal compliance with the theta criterion, and 2. a reanalysis constraint, which rules reanalysis in or out based on structural differences between the 'easy' and the 'difficult' cases. Although Pritchett's model accounts for an impressive amount of facts, it has been widely rejected in the field of psycho linguistics. This rejection is party explained by the fact that in the field of psycho linguistics, the recognition of the importance of intuitive judgments by native speakers is not a matter of course. Furthennore, there has been some criticism on Pritchett's attachment procedure; this criticism does not hold up to closer inspection though (for details, see Mulders (2002), chapter 3). Finally, Pritchett'S reanalysis constraint has been shown to run into problems in accounting for some Japanese processing facts (Mazuka and Itoh
J 995) .
My work aims at solving these empirical problems, by taking a new take on the reanalysis constraint. For reasons of space, I cannot go. into the specifics of Pritchett's framework here, or the specifics of the objections raised against it, and how my approach remedies these objections while maintaining some of Pritchett's basic insights. The interested reader is referred to Mulders (2002), which discusses the development of the current proposal in detail. 2.
A
Transparent Parser
Having established the empirical domain of this work, we now move to the de mands we pose on our processing theory. We have already seen some evidence suggesting that in order to account for the presence vs. absence of strong gar den path effects. we need a parser that consists of two components: an analysis component, and a constrained reanalysis component. But what fonn could these components take? Ideally, we are looking for a transparent parser. Le. a parser which is as close as possible to grammatical theory as we know it, without any parser-specific operations. The rationale behind this wish is, firstly, a matter of a simple desire for elegance and simplicity in the theory of language (subsuming grammatical theory and processing theory). Syntactic theory already contains mechanisms for building structure (Merge), and it also contains a way to restrict grammatical operations to certain domains (the Phase Impenetrability Condition, Chomsky 1 999). If these building blocks rum out to suffice for explaining garden paths in sentence processing, no processing-specific operations are called for. Another requirement that our parser should meet is that it should be universal: it should apply to all languages, just like grammatical theory does. This seems like an even more acute concern than the similarity between grammatical and processing theory, for a simple reason: a child needs a parser to be able to learn his or her language in the fIrSt place, so assuming different parsers for different languages induces a bootstrapping paradox: the child will need a parser to learn its language, but it will have to kn ow its language before it can decide which parser to use. However, it has been argued in the processing literature that for instance Japanese is parsed by entirely different heuristics than English. I will come back to
239
Phases in sentence processing this in more detaiL For now, let us take the universality of the parser as a guideline that simply must be met. Now, let us see what these conceptual guidelines do for restricting our search space for the analysis component ofthe parser, and its reanalysis component. The form of the parser's analysis component is not very important for the work presented here; for present purposes, I will adopt without further discussion the theta-based structure-building mechanism of Pritchett ( 1 992):
(6)
Theta Attachment: The theta criterion attempts to be satisfied at every point during processing given the maximal theta grid.
Note that Theta Attachment implies that attachment during sentence processing is head-driven. i.e. theta-requiring element are stored in working memory until their licenser is encountered. Theta Attachment does seem rather close to Merge, the major difference being that it proceeds left-to-right rather than bottom-up (for more information, see Mulders (2002». As for the reanalysis component. which is the topic of this paper, I would like to pursue the intuition that our requirement for the parser to be transparent implies that the constraint on reanalysis should be on a par with constraints on movement in syntax. Of course, movement in syntax is contrained by the Phase Impenetrability Condition: (7)
Phase Impenetrability Condition (PIC): The domain of H is not acces sible to operations outside HP, but only H and its edge. (Chomsky (1999), p. l O)
The PIC seems like a particularly nice starting point for looking at the constraint on reanalysis in sentence processing. The reason for th is is that we can be sure that during sentence processing, constituents will need to be 'closed' in order to avoid overflow of WOrking memory. It seems conceptually plausible that these units for closure would be identical to phases. In order to be testable, we need to fonnulate more precisely the intuition that phases should constrain reanalysis in syntactic processing. My proposal is to (onnulate it as follows: ( 8)
Phasal Constraint on Reanalysis (PCR) (to be revised); In case of reanalysis from source position S to target position T: ifS is phase-embedded with respect to T, S must be in the edge ofa phase.
Where phase-embedded can
(9)
be defined as follows:
X is phase-embedded with respect to Y iff there is a phase boundary P that dominates X but not Y (where P '# X and P '# Y).
The intuition behind the reanalysis constraint in (8) is rather straightforward: reanalysis should be barre d whenever the parser has to remove an element from a closed phase in order to perform the reanalysis. The only 'escape hatch ' from this constraint should be the edge of the phase, just as in syntax. To get a feel for how the peR works to constrain reanalysis, consider again ( l ), repeated here as ( 1 0): ( 1 0)
l After Susan drank the water evaporated
As we have seen, the water will be attached initially as the direct object of drank, due to (6).
240
Iris Mulders (I I)
PP
�
P
TP
I
� afterOp 81 T' � I Susan
T
VP
�
V
OP 82 ..<::>... the water
I
drank
Now, the reanalysis that is necessary to arrive at the correct interpretation of ( 1 0), can be seen as 'movement' from the object position of the embedded clause, to the subject position of the main clause. This can be depicted as follows: TP
( 1 2)
� I DPij)' � � TP
PP
I
after
OP
Susan I
T'
VP
/'--.... � T VP the water I evaporated
� I
drank
In order to determine whether the reanalysis in (I I ) is permitted by the peR, we look at the reanalysis of the argument the water. We indicate the source position of the water with a singly boxed node. In this case, the source position is the complement position of drank. so the node that occupies that position in the end structure, is marked with a single box. The target position is marked with a double box. As we can see in ( 1 2). the source position of the water is phase-embedded with respect to its target position; the VP phase dominates the source position, but not the target position. The embedding phase is indicated here by the box around the VP-phase Since the source position is not in the edge of the VP, automatic reanalysis is barred by the peR. This means that automatic reanalysis is not possible for the parser in ( 1 0); we are basically stuck with a structure in which evaporatedhas no subject, and the water is 'trapped' in direct object position. It seems to be true intuitively that this is indeed the automatic interpretation of(lO): the interpretation seems to be missing a subject role for evaporated. Of course, a native speaker confronted with ( 1 0) will attempt to repair this, "and in doing so helshe may arrive at the grammatical analysis; the important point is that conscious processing is needed to arrive at the correct interpretation. Reanalysis is not done automatically and effortlessly by the 1 Throughout this paper. I use simplified VP-structures rather than an elaborate vP-structure, unless the difference between a vP and a VP structure really matters.
24 1
Phases in sentence processing
automatic human parser. Now, let us contrast this with the easi ly processible (4b), repeated here as ( 1 3 ). (13)
John believed Mary came
The initial parse ofthe string John believed Mary is depicted below: ( 1 4)
TP
�
DP I
John
T'
/'-..... T VP
�
V I
beJieved
DP I
Mary
This initial analysis has to be revised when came is encountered. The result in g structure is given below:
( I S)
TP
�
T'
DP I
John
/"-...
T
VP
� �
beli ved
�P
C
� Ug!J rf't, � I
Mary
T
VP _ I V
came As we can see in ( 1 5), the source position is not phase-embedded with respect to the target p osition: there is no phase boundary that dominates the source position but not the target position . This means that the peR correctly predicts no conscious processing problems in (14). The intuition with respect to examples Jike (l4) is that reanal ysis is possible as long as a constituent is 'still be ing worked on' � We have now seen two simple examples of how the PCR discerns auto matic reanalysis from non-automatic reanalysis. The approach may seem oddly representational in nature, especially for a theory about processing: it looks at the grammatical analysis for a garden path sentence and evaluates the relation be tween the source and the target position in order to determine whether reanalysis could take place automatically. However. as we will see, this way of evaluating reanalysis, which was introduced by Pritchett, turns out to be rather productive. In the remainder of this paper, we will see how the peR accounts for a cross-linguistic variety of garden path effects as well as for instances ofautomatic reana lysis. In section 3, I will discuss reanalysis from the VP-phase. Section 4 will discuss reanalysis from CPs. In section 5, we will see that the PCR has
242
Iris Mulders
to be constrained somewhat more. Finally, in section 6, I would like to make a contribution to the debate as to whether next to CPs and VPs, there is also a nominal projection that counts as a phase for reanalyis; I will review evidence from processing that bears on that issue. 3.
Reanalysis from the
VP
In this section, I will discuss reanalysis of arguments crossing a VP boundary. With respect to the VP-phase, the PCR makes a particularly interesting prediction: the peR allows for reanalysis of direct objects in OV-Ianguages, like Japanese, if we assume that direct objects in OV-Ianguages are located in the edge of the vP-phase. It may be good to note here already that the refonnulation of the PCR that will be undertaken in section 4 predicts that direct objects can be reanalyzed in OV-languages regardless of any assumptions about their exact position inside the VP. The prediction that it should be possible underthe PCR to reanalyze direct objects in OV-Ianguages sharply contrasts with the prediction the PCR makes for reanalysis of direct objects in VO-Ianguages like English: direct objects in these languages do not occupy an edge position, and therefore automatic reanalysis of direct objects is predicted to be impossible in YO-languages. We have alread seen that the latter prediction is borne out; (1) is a garden path sentence. As far as I know, it is cross-linguistically true that reanalysis from a direct object position to the subject position of a higher clause is impossible in YO-languages. Consider for example the following Hebrew case: ( J 6)
i,axarej §e+dana §ateta ma'yim zarmu me+ha+berez after that Dana drank water flowed from the tap After Dana drank water flowed from the tap. (Sadeh Leicht (2003), ex. (2»
The garden path effect in ( 1 6) is of course also accounted for by the PRC. Now, let us tum to the prediction of the PRC that reanalysis of direct objects in OV-languages is allowed. An example by Mazuka and Itoh ( l 995) shows that this prediction is in fact borne out: ( 1 7)
OK
0 Yumiko-o
yobidasita kissaten-ni nagai koto mata-seta Yumiko-ACC summoned tea room-LOC long time wait-made
• 0 made Yumiko wait for a longtime at the tearoom to which he summoned her' (Mazuka and Itoh 1 995: ex. l 8b) This'sentence is parsed as follows. When the verb yobidasita 'summoned' is encountered, the following structure can be built: ( 1 8)
TP
�
DP I
pro
T' �
VP
T
�V
DP I Yum iko-o
yobidasita
summoned
243
,
l
Phases in sentence processing
I ,
i '
This is a simple main clause analysis. Note that in Japanese, it is generally possible to drop arguments; in ( 1 8), a small pro is inserted in subject position. Note furthennore, that for ease of representation I use a flat structure to represent the VP. Bear in mind that the direct object, which is located in the left of the VP, is actually assumed to be in the edge of vP for our current purposes. The next input is the noun kissaten-ni 'tea room' . Relative clauses in Japanese are right-headed; the structure in ( 1 8) is attached as a relative clause modifying kissaten-ni 'tea room '. The resulting structure looks like this: ( 1 9)
NP I N'
�N '
CP
o�, I
h
� kissaten-ni
TP
C tea room
� DP T' I �
pro
VP
T
�
DP
NP
Yum iko-o
ti
I
I
V
yobidasita summoned
'The tea room where I1he summoned Yumiko. '
The next input is the matrix verb mata-seta 'made-wait'. According to Mazuka and Itoh, this verb triggers reanalysis of Yumiko-o from the embedded object position
to the matrix object position. It is not entirely clear why reanalysis is preferred over .simply inserting pro in the matrix object position. This probably has something to do with the distribution of pro in Japanese. The reanalysis is depicted in (20):
1 r
Ii
I 244
'
Iris Mu lders
(20)
TP
� r DP I �
pro
VP
T
mata-seta
wait-made
CP
�
OPj
C' /'-... TP C
�T' DP
N' I N kissaten-ni
tea room
I
V
yobidasita summoned
In this structure, there are two phase boundaries that embed the source position with respect to the target position: the relative clause Vp, and its CPo However, the peR predicts the reanalysis to be unproblematic, because the argument that is being reanalyzed, originates in the edge of the VP. Note that for the PCR to allow this reanalysist the source position has to be in the edge of a phase, so the presence of the extra embedding CP-boundary does not rule out the reanalysis. As we have already seen, this prediction is correct: ( 1 7) is processed effortlessly by native speakers of Japanese. I would like to stress again how similar the reanalysis in ( 1 7) is to the reanalysis in (10). In both sentences, a direct object has to be reanalyzed from an argument position in an embedded clause to an argument position in a main clause. But while this turns out to be possible in Japanese, it is impossible in English. The PCR correctly predicts this cross-linguistic state of affairs. The universal 'escape hatch' for reanalysis of direct objects crossing a VP-phase boundary, simply does not apply in VO-Ianguages. 4.
Reanalysis from CP-phases
Having looked at reanalysis from the vP-phase, we now move to the CP-phase. We have already established that direct objects can be reanalyzed in Japanese, and we may wonder about the reanalyzability of subjects in Japanese. It turns out that reanalysis of subjects in Japanese is in fact allowed. Consider (2 ] ):
245
Phases in sentence processing
(2 1 )
N akamura-ga tyuuko-no pasokon -o kana toki Nakamura-NOM sec ond-hand PC-ACC bo u ght when syuuri-site-kureta repaired (for me)
' When lI( he/she) bought a second-hand com puter, Nakamu ra repaired it for me. ' (Mazuka and Itoh 1 995: ex.9=37) At the point of discovery of katta 'bought', the string is straightforwardly analyzed as a main clause. When toki ' whe n ' is encountered, it is added to the i n iti al main clause, and we now have an adverbial clause:
(22)
CP
�C TP
� DP I Nakamura-ga
T'
toki when
/'....
VP
T
�
N
� pasok tyuuko-no
second-hand PC
on o
kana
bought
Now, when the matrix verb syuuri-site-kureta 'repaired for me' is encountered, it becomes clear that this is not the right analysis.2 Mazuka and Itoh report that it is poss ible to interpret the string in (2 1) with a pro in matrix subject position referring to a third person, but according to them this reading is difficult to obtain without a proper context. This means that at the point of discovery of the matrix verb, the subject in side the embedded clause has to be reanalyzed as the subject of the matrix clause, and the s ubject position inside the embedded clause has to be filled with a pro. lAs Mazuka and ltoh explain. the verb syuuri·site.kurela is a special verb form that is used when the speaker is the beneficiary of the action expressed by the verb - sile kureta means something like 'someone did - for my benefit'. Because of this meaning, lite verb syuuri.site-kureto 'repaired for me' cannot take apro as its subject which is interpreted as the speaker. Mazuka and Itoh contrast (21) with the following example:
(i)
Nakamura-ga tyuuka.no pasokon.a katta lOki 0 syuuri-site·yatta Nakamura-NOM second-hand PC-ACC bought when repaired 'When Nakamural bought a second·hand personal computer, (Ij) repaired (it) for him, ' (Ma zuka and ltoh 1995: ex. lO)
J n this case. because a different verb form is used which does not mean that the speaker is the beneficiary, a pro referring to the speaker can be inserted in its subject position, and no reanalysis is necess8J)' when the matrix verb is encountered; Nakamllra therefore Slays in the subject position of the embedded clause. According to Mazuka and ltoh, (2 1) and (i) 'are alike in their ease of comprehension' (p.303).
246
Iris Mulders
TP
(23)
�
Nakamura-ga
CP
�
TP
�
� I
pro
T
C toki
when
/'...T
VP
�V
NP
�
tyuuko-no pasokon-o
T'
/'-... T VP
�V
DP
I
pro
syuuri-site-kureta repaired (for me)
katta bough
second-hand PC
Two points can be made with respect to this example. The frrst one is a rem inder of the enterprise we are undertaking here, namely finding a constraint on reanalysis . Note that the vel)' fIrst analysis made is the analysis of a simple, complete, Japanese matrix clause. There is no (serial) processing theol)' that would predict otherwise. In order to arrive at the structure in (23), reanalysis has to take place. However. this leads to no conscious processing difficulty at all. So, we must conclude again, that reanalysis is not generally impossible for the parser; problems due to reanalysis only occur in certain structural environments. The second point with respect to this example is that we see that a subject can be reanalyzed from an embedded clause to a main clause. As it turns out, reanalysis of a subject from an embedded clause to the matrix clause is generally possible in Japanese. We will see some more examples in section 5. But under the PCR formulated in tenns ofthe standard conception of edge, we would not expect reanalysis of subjects to be possible, since subjects do not occur in the edge of a phase: they are not in the specifier of CPo One way to solve this, is to say that the notion of edge may be defined differently in processing. Note that whereas the subject in (23) is not in the edge ofCP syntactically speaking, it is in the edge in some phonological sense. In these examples, the subjects do constitute the edge ofthe CP-phase, in the sense that they are the first edge-positions within the CP-phase that are filled with overt material. A similar m ix-up ofa phonological notion ofedge versus a purely structural notion of edge plays a role in the discussion of object shift in Scandinavian in Chomsky ( 1 999); in orderto solve it, Chomsky introduces the notion ofphonological border, which refers to the leftmost phonologically oven constituent in a phase. It looks like something similar is needed to account for reanalysis from subject position. This means that the PCR will have to be refonnulated along the following lines:
(24)
Phasal Constraint on Reanalysis (PCR) (revised, and to be revised):
In case of reanalysis from source position S to target position T: if S is phase-embedded with respect to T, S must be in the phonological border of a phase.
247
Phases in sentence processing
To my mind. the implicit assumption here that the parser can only be constrained by overt material does not seem unreasonable� after all, overt material is the only material the parser can be certain of; elements like empty operators traces and pro argum e nts are postulated by the parser itself, do not appear as such in the input. and may tum out to be wrong. Note furthermore that i f we use the formu lation in (24). reanalysis of direct obje cts in OV-Ianguages w ill be allowed regardless of the precise syntactic analy sis of direct obj ects in OV-languages, since the direct obj ect will always be in the phon o lo gical border, even ifwe assume that it is base-generated in a position to the left of the verb, and stays there in the syntax. ,
s.
Fu rther Constraining the peR
So far, we have seen that the PCR can account for reanalysis from subject and object pos ition in OV-Ianguages, and that it successfully bars reanalysis from object p osition in VO-Ianguages. In this section, I will discuss some facts from Japanese that seem to indicate that the PCR in its current formulation needs to be c onstrained a little bit further. The Japanese facts that I have in m ind concern reanalysis of more than one element. M azuka and Itoh (1 995) g ive the following minimal pair: (25)
a.
ox.Yoko-ga
kodomo-o koosaten-de mikaketa onnanoko-ni Yoko-NOM child-ACe intersection-LOC saw girl-DAT koe-o kaketa called ' Yoko called the girl who saw the child at the inte rsection (Mazuka and Itoh 1 995:ex.8b) b. lYoko-ga kodomo-o koosaten-de mikaketa takusii-ni Yoko-NOM child-ACC intersect ion -LOe saw taxi-OAT noseta put on 'Yoko put the child on the taxi she saw at the inters ection (Mazuka an d Itoh 1 995:ex. 14a) '
-
.'
Let us first consider the parse of (25a). When the verb mikaketa 'saw' is encoun tered, the first three nominal constituents can be theta-attached in a single clause, as foll ows: TP
(26)
�
DP
T' � Yoko-ga VP T
I
�
NP I kodomo-o child
NP I koosaten-de intersection
V mikaketa saw
This main clause analysis has to be revised when onnanoko-ni' girl' is .encountered. Yoko has to be taken out of the structure, and then a relative clause can be built,
248
Iris Mulders
where the trace of the relative operator associated with onnanoko-ni ' girl ' can occupy the su bje ct position, as in (27):
(27)
I�I I
NP I N'
Yoko-ga
CP
�
OPj
C' � C TP
onnanoko-ni girl
�,
� T
VP
�
NP I
NP I
kodomo-o
koosaten-de
child
intersection
V
mikaketa saw
Note that the reanalysis of Yoko in (27) does not violate the PCR. The source position of Yoko is phase-embedded with respect to its target position. The relative clause CP dominates the source position (the subject position of mikaketa 'saw'), and it does not dominate the target position, because the target position is not even in the same structure: Yoko has simply been sent back to store. However, since the source position occupies the phonological border of this phase, the reanalysis is allowed by the PCR. The two nominal constituents in (27) can be theta-attached when the matrix "verb" koe-o kaketa 'called' is encountered:
249
in sentence processing
Phases
(28)
TP
�T'
NP
I
�
Yoko-ga
T
VP
�V
�alled NP I N'
CP
N' I N onnanoko-ni
�
OPi
C'
�T'
ti
girl
�
TP
koe-o kaketa
C
�T
VP
N� I
I
kodomo-o
koosaten-de
child
intersection
mikaketa saw
As We have seen, the parse of (25 a) does not run into problems at any point. However, (25b) is less easily processed. Its initial analysis is identical to the initial analysis of (25a), i.e. the simple main claus.e structure in (26). But when the head noun takusii-ni 'taxi' is encountered, a more drastic reanalysis is invoked.
I
I
250
Iris Mulders
I�I I
(29)
Yoko-ga
I�I I
NP I N'
kodomo-o child
CP
�
C'
OPj
takusii-ni taxi
�C TP
�, � pro
VP
T
~ t.I
I koosaten-de intersection
mikaketa saw
The trace ofthe relative operator cannot be attached in subject position, because the relative clause verb mikaketa has a thematic requirement for an animate subject, and takusii-ni is inanimate. This means that the trace of the relative operator has to be attached in direct object position, and the original object kodomo-o ·child', has to be sent back to working memory. This in turn means that Yoko-ga also has to be removed from the initial analysis, because of the linear order of Yoko-ga and takusii-ni. So in (29), both the subject and the object of the initial analysis in (26) have to be reanalyzed. This reanalysis turns out to be very difficult to perform for native speakers of Japanese. Mazuka and Itoh ( 1 995) claim that the reason is that it is very difficult in general to reanalyze more than one argument in Japanese. They provide numerous examples to prove this point. One of these examples is the ,
following: (30)
l,H uruhasi-ga
Yumiko-o 0 0 yobidasita kissaten-ni Huruhashi-NOM Yumiko-ACC pro pro summoned tea room-LOC nagai koto mata-seta long time wait-made Huruhasi made Yumiko wait for a long tim e at the tea room to which he (Mazuka and Itoh 1995 :ex. 1 7c=37) summoned her. ' •
(30) is almost identical to ( 1 7), the only difference being that (30) has an overt subject. The initial analysis looks like this at the point of discovery of the head noun of the relative clause:
25 1
Phases i n sentence processing
(3 1 )
NP I N'
�N'
CP
I
�
N � kissaten-ni TP C tea room
OPj
C'
�
T' DP I � Huruhasi-ga VP T
�
DP I Yum iko-o
V yobidasita
NP I tj
summoned
'The tea room where Huruhasi summoned Yumiko. ' When the main verb is encountered, the subject and the object ofthe relative clause have to be reanalyzed as arguments of the main verb: TP
(32)
�
II D
VP
I Huruhasi-ga
T
mata-seta wait-made
CP � C' OPj
/'-....
TP
�, I
C
N' I N kissaten-ni tea room
�
pro VP
T
� I
TO
I yobidasita ti summoned
But where reanalysis of just the object in (1 7) is unproblematic, as we saw, and reanalysis of just a subject is also possible (remember (2 1 », it is not possible to reanalyze both the subject and the object at the same time: (30) is a garden path. The peR however, predicts that it should be fine, since the source position of
252
Iris Mu l ders
Huruhasi-ga occu rs in the phonological border of the relative clause CP, and the source position of Yumiko-o appears in the phonological border of its VP. The pattern is general; Mazuka and ltoh give a number of minim al pairs with the same structure.
(33)
a. i.Yakuza-no kanbu-ga wakai kobun-o sagasi-dasita gang-GEN leader-NOM young member-ACC found kenzyuu-de utikorosite simatta gun-with shot to death 'The leader of the gang [shot; 1M} the young member to death with the gun he found.' b. OJ(.Yak uza-no kanbu-ga wakai kobun-o sagasi -das ita gang-GEN leader-NOM young member-ACC found otoko-ni rei-o itta man-OAT th anked 'The leader of the gang thanked the man who found the young (Mazuka and Itoh ] 995:ex. 15) member of the gang'
(34)
a.
i.Hati-gatu-ni natte kara, Yamasita-ga yuuzin-o August-to became after Yamasita-NOM fri end-ACe hoomonsita kaisya-de mikaketa company-at saw visited 'After it turned into August, Yam asita saw his friend at the company he visited. ' b. oI:Hati-gatu-n i natte kara, Yamasita-ga yu uzin -o August-to became after Yamas ita-NOM friend -ACe hoomonsita siriai-ni tegam i-o kaita visited acquaintance-to letter�ACe wrote 'After it turned into August, Yamasita wrote a letter to an acquain (Mazuka and Itoh tance who visited his friend. ' 1 995:ex. 16)
The unprocessability of cases like (25b), (30). (33b), and (34b) implies that the peR is too wide; it seems that reanalysis from a phonological border position has to be restricted to one element at a time. The exact formulation is no trivial matter. We have seen that simultaneous reanalys is of both a subj ect and an object from their base positions is impossible in Japanese. However, Mazuka and !toh show that it is possible to reanalyze both a subject and an object if the object is scrambled over the subject: (35)
OKY
umiko-o Hurahasi-ga yobidasita k issaten-ni nagai koto Yumiko-ACC Hurahasi-NOM summoned tea room-LOe long time mata-seta wait-made
This is quite a spectacular finding. Let us consider step by step how (35) is processed. The initial analysis at the first verb will look something like this:
253
Phases in sentence processing
(3 6)
TP
�TP
�
DPj
�
Yum ko-o
D I
'
/'--....
Huruhasi-ga
VP
T
�
V
tj
yobidasita
summoned Note that the subject and the object are both located in a specifier of TP in this structure. This structure can be incorporated in a relative clause when the head noun kissaten-ni 'tea room' is encountered:
NP I
(3 7)
N'
�N'
CP
I N
� CO
OPj
/'---... TP C
kissaten-ni
tea room
�TP t Yumlko-o � DPj
DP I
T' �
Huruhasi-ga VP
T
�
tj
tj
V
yobidasita
summoned 'The tea room
where Huruhasi summoned Yumiko.'
This analysis has to be revised when the matrix encountered:
254
verb mata-seta
' m ade-wait' is
Iris Mulders
(38)
�P IHUru� l oU TP
i
Yum ko.o J
I
VP
tj
NP
I
N'
T
V mata-seta wait-made
CP
�
OPj
C'
� TP C
kissaten-ni tea room
� �, P
J
�
pro VP
T
�V
pro
�p
I yobidasita
tj
summoned
Note that in (38), the source positions both occupy the phonological border of the embedded CPo We can now attempt to refonnulate the PCR in such a way that it will account for the facts discussed in this section. We have seen that reanalysis of both a subject and a preverbal object is disallowed, unless the object is scrambled across the subject, which results in a structure where they are in specifiers of the same (TP) projection. This means that we will have to restrict the PCR so that it only allows for reanalysis from one phase at a time:
(39)
Phasal Constraint on Reanalysis (peR) (twice revised):
In case of reanalysis involving source positions 81 Sn and target posi tions T1 T n (where for each i t 1 5 i 5 n. 8j is the source ofTi) : if some 8i is phase-embedded with respect to Tit then there must be a phase P such that for all i, 8i is in the phonological border of P. ...
••.
Furthennore, we can specify the notion of phonological border, which has been used rather loosely so far, to mean the following:
(40)
The phonological border of a phase consists of the specifier(s) of its highest projection which contain(s) overt material
255
Phases in sentence processing
This allows for reanalysis of both the subject and the scrambled object in (38), because the source positions ofthe subject and the object are in the specifiers ofthe highest projection (TP) inside a phase boundary (the embedded CP). (39) excludes reanalysis in (32), because the source positions there are not in the phonological border ofone phase: the source position ofthe subject in (32) is in the phonological border of the CP, and the source position ofthe direct object is in the phonological border of the VP. (32) shows that the parser can only look for material that must be reanalyzed, in (the phonological border of) one phase at a time. (39) also accounts for the following example, which is presented as a puzzle by Mazuka and Itoh: (4 1 )
oKHiroshi-ga aidoru kashu-o kakusita kamera-de totta Hiroshi-NOM popular singer-ACC hid camera-with photographed ' Hiroshi photographed the popular singer with the camera he was hiding' (Mazuka and Itoh 1 995: ex.20)
Mazuka and Itoh's processing theory looks like this. They also assume that attachment is head-driven, just like is assumed here. They furthermore state the following: (42)
The Tentative Attachment Strategy:
In Japanese, a parsing decision is tentative until the sentence is finished. By tentative, we mean that reanalysis ofeach decision will have a psycho logically measurable cost (i.e., it is not cost-free), but any single reanalysis will not be costly enough to cause conscious processing difficulty. When reanalysis is combined with other complexities (e.g., lexical ambiguities, multiple reanalyses, pragmatic naturalness, etc.), it becomes increasingly costly and may become conscious. (Mazuka and Itoh 1 995: ex,45) It is importantto note that this approach entails a departure from a universal parser: attachment is tentative in Japanese, but not in English. To see this, consider again the English garden path in (43 ): (43)
�r
i After Susan drank the water evaporated
In order to solve the garden path here, all the parser has to do is to take out the water, and attach it in the subject position of evaporated. So if the Tentative Attachment Strategy would also apply in English, the prediction would be that (43) is not a garden path. This is of course not correct; the Tentative Attaclunent Strategy is specifically meant to account for processing of Japanese alone. Now, in Mazuka and Itoh's view, the conscious processing difficulty'that is found in sentences like (25b), (33b) and (34b), arises because iri those cases, two arguments have to be reanalyzed. The ease with which (42) is processed, is a problem for this theory, since this case also involves reanalysis of two arguments, but is nevertheless processed without any conscious effort. In order to account for the difference in processing difficulty between (42) on the one hand (their (20» , and (25b), (33b) and (34b) on the other hand (their (1 4), (1 5), ( 1 6), and ( 1 7» , they say: _
'The difference between the garden-path sentences and (20) can prob ably be accounted for by factors including the meaning ofan individual verb, the strength of the relation between the NP and the verb, and other pragmatic knowledge. For example, hiding a popular singer in (20) is not as likely an event as seeing a child in ( 1 4), finding a gang
256
I
t
I f
i
r
I
I I �
ir
Iris Mulders
member in ( 1 5 ), visiting a friend in ( 1 6), or summoning a person in ( 1 7).' (Mazuka and ftoh 1 995: p.308) To my mind, it makes no sense at all to suppose that hiding a popular s inger is less likely an event than finding a gang member, so I don't believe this to be the right explanation. But when we look at the parse carefully under the PCR, it turns out that under the assumptions we have been making here, the difference between (43) on the one hand, and (25b), (33a), and (34a) on the other hand, is not surprising at a ll The initial analysis is a main clause: .
(�)
TP
�T' �T VP
Hirosi-ga
�V
NP
�
aidoru kashu-o
kakusita hid
popular singer
When the head noun kamera-de is discovered, the trace of the relative operator associated with it can be attached in subject position, as kamera can in principle be the subject of kakusita 'hide'; kakusita does not select for an animate subject. This means that only one DP needs to be sent back to working memory, and this is allowed by the peR. The resulting structure is the following: (45)
I �I
NP I N'
I Hirosi-ga CP
�
OPj
C' � TP C
�I
N' I N kamera-de camera-with
�
J
t·
VP
T
�V
NP
kakusita hid
'Hiroshi' 'with the camera that hid the popular singer' Next, aidoru kashu-o 'popular si n ge r has to be taken out of the relative clause '
257
Phases in sentence processing
when the matrix verb is discovered:
(46)
TP
�T' DP Hiroshi-ga
�
VP
T
NP I N'
CP
�
OPj
C'
r0
�
DP pro
totta photographed N' I N kamera-de camera-with
T'
/"-.... T VP
�v I tj
kakusi hid
The revisions are slightly more complex than in cases like (25a) and (23); the trace of the relative operator is put in the object position in the relative clause, and Jhe subject position in the relative clause, which was occupied by the trace of the relative operator in the first relative clause analysis, is filled with a pro argument. Still, the only reanalysis of an overt argument in this step is reanalysis of aidoru kashu-o 'popular singer'; this argument can be taken out under the PCR, because it is located in the phonological border of the CP phase. The main point of difference between this case and problematic cases of reanalysis of two arguments such as (25b), (33a) and (34a) under the current approach, is that in this case, the reanalysis of the two arguments takes place at two different points in the parse and is pennitted at each of those points because the reanalysis takes only one argument at a time, which is allowed by the PCR. In (25b), (33 a) and (34a), the reanalysis of the two arguments has to take place at a single point in the parse. This is disallowed by the PCR.
6.
Which Nominal PrOjections Count as Phases for Sentence Processing?
Up until now, we have just been looking at reanalysis crossing VP and CP phase boundaries. In this section, I will look at reanalysis that takes place in nominal projections, or that crosses nominal projections. Whether these projections count as phases, is under debate (see for instance Matushansky, this volume).
258
Iris
6. 1 .
Mulders
Reanalysis that does not cross a phase boundary is perm itted
In this section, I will discuss cases of automatic reanalysis. Under the peR, reanalysis that does not cross a phase boundary is predicted to not le ad to garde n paths effects. We have already seen one case: (4b). It is generally possible for the parser to perfonn reanalysis of this type. Consider for instance (47):
(47 )
1 like her students' papers' quality
(47) involves three instances of reanalysis of the type discussed above. See the consecutive steps in (48): (48)
VP V DP
a.
like
VP
b.
�
��p � II D D
I her
�
l e
NP
her
� j() VP
c.
d.
ike
I DP I her students'
� D
NP
I
papers
I
VP
students
�
like
D l op i
her students' papers
•
�
o
'
NP
I
quality
In each of the instances of reanalysis· here, the new input word tak es over the complement position of like, and the material that was attached in that position before the new input word was encountered, is 'pushed' to the specifier of the complement position. Since the material that is reanalyzed does not cross a maximal projection, there is no phase that could embe d the source position with respect to the target position. Therefore this type of reanalysis is allowed by the peR. Note that it is crucial here that phase-embedding, as defmed in (9), does not count target or source positions as potential embedders. 6.2.
Reanalysis crossing a PP proj ection
Evidence from processing suggests that processing. To see this, consider (49): (49)
PPs count as a boundary for sentence
l Below the stairs collapsed.
Th is sentence is processed as follows.
When the stairs is encountered, it is attached as the complement ofthe theta-assiging preposition below:
259
Phases in sentence processing (50)
PP
�
P below
DPel
�
the stairs
When the verb collapsed is encountered, this analysis has to be revised, since collapsed requires a (nominal) subject. TP
(5 1 ) pp
TP
UD�
�
the stairs
T
VP I V collapsed
The reanalysis in (5 1) violates the peR, if PPs count as phases - there is no other phase that embeds the source position but not the target position of the stairs. Since (49) is a garden path sentence> we can conclude that PPs do constitute boundaries for reanalysis in sentence processing. 6.3.
Reanalysis with a phase-embedded target positio n
Up until now, we have only looked at cases of unpermitted reanalysis where the source position is phase-embedded with respect to the target position. That is, we have only looked at cases where reanalysis 'removes> an element from a phase. There are, however, also cases where reanalysis 'adds' material to a closed phase. In this section, I will discuss cases like these. As we will see, they provide evidence that �ere is some nominal projection that counts as a boundary for reanalysis. Consider the following example, which is a garden path: (52)
i.
I put the candy in tbejar into my mouth
Initially, the parser attaches in the jar as an argument of put. The reanalysis in (52) then looks like this:
260
Iris Mulders
(53)
TP
�
NP 91
,
r
�VP
T
: - - DP 92-
-
-----:
NP - - - - ," : : the� I I I' I
t
I
N'
into my mouth
II
1 �<['[ffi I 1 ' N
�llndy
_ _
� II
Ln.!h� ia[:
In the structure in (53), there are two maximal projections that embed the target position, but not the source position: the NP and the OP. Since (52) is a garden path, we can conclude from this that either the NP or the OP is a phasaJ boundary for reanalysis. On the basis of (53), we have no way of deciding which ofNP or OP is the relevant boundary, so we will have to postpone judgment of that. We can, however, add the following statement to the peR: (54)
1fT is phase-embedded with respect to S, reanalysis is disallowed.
(54) will exclude the reanalysis in (53) under the assumption that NP or DP is a phase boundary. All that needs to be determined is whether the NP or the OP is the relevant phase boundary. To do that, compare (52) with (55): (55)
They gave her books to Ron
(55»s processed without any problems whatsoever. However, it does involve reanalysis. The initial analysis, maximally satisfying the theta criterion, is given in (56):
261
Phases in sentence
(56)
processing
TP
�
NP I they
T'
�
vP
T
�
V-v gave
VP
�
V'
DP I her
�
tv
NP I
books This analysis must be revised when an extra argument, Actually, two instances of reanalysis are necessary :
10
Ron, is encountered.
TP
(57)
�T'
DP
� �
th y
T
vP
�
VP
V-v gave
0'
0Uk11 I
N books In (57), her is reanalyzed from the specifier of VP to the spe c ifier position of the DP. This reanalysis is unproblematic, since the source position and the target position occur in the same phases. The second instance of reanalysis in (57) is reanalysis of booles from the comp lem ent position of V to the complement position of D. Now, there is a maximal projection that dominates the target position in this case (the maximal DP proj ection dominating booles) that does not dominate the source position. Since (55) is not a garden path sentence it seems that we must conclude that DP projections do not count as phases for sentence processing. We just concluded on the basis of (52) that either the NP or the DP count as a phase boundary. Given that in (57) a DP proj ection embeds the target position. and (57) is easily processed, we can conclude that the relevant projection that constirutes the relevant phase boundary in (52). must be the NP. ,
262
i
I I I
Iris Mulders
7.
Conclusions
In this paper, I have shown that reanalysis in sentence processing is constrained by structural factors. I have fonnulated a constraint on reanalysis, the peR, which makes use of the syntactic notion ofphase. We have seen that the PCR can account for cross-linguistic garden path data. The PCR as developed in this paper gives us an interesting new perspec tive on what is widely perceived as a mystery in the processing of Japanese as compared to English. In the words ofMazuka and Itoh (1 995), the problem is this: Most models of sentence processing developed for English predict that almost all sentences in a language such as Japanese should induce a garden path. But the majority of Japanese sentences predicted to induce a garden-path effect by these models do not produce difficulties in processing (e.g. Mazuka and Lust, 1988, 1 989; Mazuka et aI., 1 989). (Mazuka and Itoh 1 995: p.296) As we have seen, the step that Mazuka and Itoh take to solve this problem boils down to saying that processing Japanese is done in an entirely different way than processing English. Evidently, the concept ofhaving different parsers for head-final and head initial languages, raises the question of how these language-specific parsers would be acquired. Mazuka (1 998) argues that there is no problem in departing from tile ideal of a universal parser for aU languages; she argues that children at a very early age can decide on the basis of prosodic properties of the language they are learning whether their language is head-final or not; and that they can choose a parser according to that property. While this approach may be feasible when applied to the Japanese-English contrast, it raises significant questions for. for instance, Dutch. Japanese is head fmal throughout its grammar, and English is head-initial. But Dutch has a mixed system: it is head-initial in all projections except the verbal projections. What would this mean for the Dutch child hying to decide which parser to use? Would it use a different parser for different projections in the same sentence? The reanalysis constraint developed here tackles the problem of English versus Japanese from a different angle. It says that reanalysis from the phonological border of a phase is possible for the Human Sentence Processor. This type of reanalysis will occur quite frequently in Japanese, and it will only rarely occur in English, due to the different syntactic properties of these languages. Dutch,-with its mixed system, provides an excellent opportunity to test whether the predictions ofthe constraint are correct; in Mulders (2002), I argue that this does indeed appear to be the case. Thus the puzzle of the virtual absence of garden path effects in Japanese compared to the robustness of garden path effects in certain contexts in English can be solved with a universal parser which is fonnulated in tenns that can directly be linked to competence theory. References Chomsky, N. ( 1999). Derivation by Phase, MIT occasional papers in linguistics No. I S,
Cambridge, Mass. Matushansky. O. (2005). Going through a phase, MITWPL 49: 1 57-1 8 1 . M azuka. R. (I 99S). The development oJlanguage proceSSing strategies: a cross-linguistic study between Japanese and English, Lawrence Erlbaum Associates.
263
Phases in sentence processing
Mazuka, R. and l toh, K. ( 1 995). Can J apane se speakers be led down the garden pat h?, in
R. Mazuka and N. Nagai (eds), Japanese Sentence Processing. Lawrence Erlbaum Associates. pp. 295-329.
M ulde rs, I. (2002). Transparent Parsing: Head-Driven Processing o/Verb-Final Structures, PhD thesis, Utrecht University. *http://www. library.uu.nlldigiarchief/dip/dissl2002-1029-094527/inhoud.htm Pritchett, B. ( 1992). Grammatical competence and parsing performance, University of Chicago Press, Chicago. Sadeh Leicht. O. (2003). Sporadic occurrence of the garden path effect, in W. Heeren, D. Papangeli and E. Vlachou ( c ds), UiL OTS Yearbook 2003, UiL OTS, Utrecht, pp. 59-68.
Universiteit Utrecht Faculteit der Letteren. UiL OTS Trans 1 0 35 12 J K Utrecht The Netherlands
iris. mulders@let. uu.nl
264
On Phases and Cyclicity Juvenal Ndayiragije University of Toronto
This article argues thaI mosl of the assumptions surrounding Chomsky's (200 I) Phase Theory are dispensable. It will be shown that MLC alone is
strong enough 10 minimize "search" for both Agree and Internal Merge (Move), without appealing to the Phase Impenetrability Condition (PIC) and ancillary assumptions. Empirical evidence supporting the proposal comes from Kirundi (i) raising asymmetries in high applicative constructions. (ii) muhiple probes valuation by a single goal in compound tenses, and (iii) long distance inverse in obligatory control constructions.
1. Introduction I will first show that on theoretical grounds alone, most of the assumptions surrounding the current Phase Theory, when carefully scrutinized and put together, bad to undesirable internal inconsistencies. The latter are highlighted in section 2 below. Then. I show in sections 3-5 that those assumptions arising from a PIC based approach to cyclicity turn to be untenable on empirical grounds as well. a fact that raises doubt on their virtual necessity. We will see that those imperfections disappear entirely under MLC. an independent locality condition on Agree and Internal Merge that is . powerful enough to minimize "search" alone. Section 6 concludes the discussion.
2. Derivation by Phase: four assumptions 2.1. Assu mption 1 : Strong Phases
According to Chomsky (2001). strong phases are those that have an OCC (=EPP) position as an escape hatch for movement; that is. yP and CPt On phonological grounds. these are targets for movement On semantic grounds. they are propositional objects: verbal phrases with full argument structure and CP with force indicators. but not TP alone or "weak" verbal configurations lacking external arguments (passive, unaccusatives).
MIT Working Papers in Linguistics 49. 265-282 Perspectives
on Phases © 2005 Juvenai Ndayiragije
On
Phases and Cyclicity
A remark is in order here, regard ing TP and Cpl. In Chomsky's system, TP is not a strong phase, unless selected by C (as i n control constructions but not in raising ones). In other words, the strong phase status of TP. if any, is contingent upon selection by C, whereas that of CP is an intrinsic feature. Yet, there is compelling ev idence in Kirundi2 suggesting exactly the opposite view. As a matter of fact, we will see in section 4 that finite TP alone fully qualifies as a strong phase, independently of C-selection. This should n ot be surprising. though. Indeed . phonological and semantic considerations aside), if the rule is that a strong phase must have an OCC (=EPP) feature, then finite TP should be on the short list. Furthermore, there is evidence in Kirundi showing that CP is not always a strong phase. Indeed. we will see in section 5 that only finite CPs arc strong phases; non-finite CPs are not. Accordingly. CP's strong phasehood is determined by T, not th e reverse as in Chomsky' s system. If so, interesting consequences will follow.
2.2. Assumption 2: The Phase Impenetrability Condition (PIC) This is an accessibility condition a imed at minimizing "search" in probe-goal matching. To see how it works, consider the abstract configuration ( 1). with two strong phases ZP and HP. YP )]]
(1)
According to PIC. the complement domain of H. YP in ( 1 ), is not accessible to operations at ZP, but only the head H and its edge a, the edge being the residue outside of H-bar. either SPEC or adjunction to HP.
2.3. Assumption 3: A Phonological Condition (PC) Consider the confi guration (2), with a probe P, two strong phases ZP and HP. and t'Yo positions XP and SPEC at the edge of HP. (2)
[zp
. .
. . P· . · .IHP
XP [SPEC [H
Y P]]]
According to the Phonological Condition (PC). XP prevents Match of probe P and a goal G in SPEC, under MLC. only if XP has phonological content. This
I
1
I will come back to the assumed strong phase status of �P later in this section. A Bantu language spoken in Burundi.
3
As a matter of fact, semantically speaking, if strong phases are propositional objects. TP alone fulfill all the requirements, at least in an Aristotelian view of what a proposition is, that is, a string containing a subject. a predicate and a tense. Furthennore, on phonological grounds, if extractability is the test as suggested by Rizzi (1982), there are languages in which TP are extractable. One of them is Fongbe (see Ndayiragije 2000). Furthennore, for languages in which CP is claimed to be a phase, say English, there are cases where CP extraction is not possible (see notably Bowers 2001). Therefore, those two criteria are not slrOng enough 10 tell for sure, in a tandem means, what is (not) a strong phase.
266
1 ;
luvenal Ndayiragije
eliminates the Equidistance Condition, a welcome result. But its merits end here; for three reasons: First, on minimalist grounds, PC is not an optimal solution. It is relevant for Internal Merge only; that is. when the matching relation between P and SPEC in (2) involves Move, but not A gree. Indeed, if PC were applying to Agree as well, that would be an interesting discovery . Yet, on empirical grounds, PC fails to prevent Agree of Probe C-T and subjects in-situ in such cases as Icelandic-like transitive expletive constructions in which Object always intervenes between T and the associate. A tool that is useful only for one operation of CHL -Internal Merge- . but not the other -Agree- raises scepticism on its explanatory adequacy. Second, PC is a complexity. Indeed, in a computational system that independently appeals to MLC, PC becomes superfluous. To see this, consider the definition of MLC in (3), which is free of the Equidistance Condition, thus restricting MLC to the c-command requirement (i) of Chomsky's ( 1 995) original formulation of Closeness4•
(3)
Closeness (MLC) If f3 c-commands a and "t' is the target of Match, then � is closer to
"t' than
a.
(Ndayiragije 1 999: 406)
Under (3), PC becomes entirely useless. MLC alone prevents Match of probe P and SPEC in (2) if the intervener HP qualifies as a potential goal. Third, PC is an imperfection in its own design specificationss. Consider (2) again repeated below:
(4)
[zp . . P . . " [HP ..
XP [SPEC [ H
YP]]]
According to PC, SPEC cannot raise to Spec of Probe P if XP has phonological content. Assume so, then a question arises: how did XP reach its position in (4) in the first place? Suppose XP is part of a non-trivial chain, that is, a derived position rather than a first merge one. Under a restrictive PC. SPEC must be phonologically empty first, or overt raising to XP will fail. It is not clear to me why a constituent in the complement domain of the head H in (4) is free to raise to XP over a SPEC with phonological content, whereas a constituent in SPEC is barred to raise to Spec of probe P over an XP with phonological content? Under a good design v iew, this is a clear defect in the manufacture of Pc.
" The latter is provided below. for clarity 's sake. Closeness
If tl c-(:ommands a and 't' is the target of raising, then � is closer to K than minimal domain of (i) 't' or (ii) a. (Chomsky 1 995:356)
Cl
unless � is in the same
S
PC is parasitic to another theory-internal assumption -the multiple specifiers hypothesis-. a questionable assumption on both conceptual and empirical grounds, but I will not pursue the issue here.
267
On Phases and Cyclicity Chomsky (2001) didn't address this issue. He rather focused on PC effects of a phonologicaJly overt XP in (4), by pointing out the following empirical consequence: if XP is an object (OB) raised to the outer Spec-v * , it can 't remain there, or the Subject (SU) in SPEC will not be accessible to the h ig her complete Probe P (C-T or v *) and the derivation will crash. In other words, if Object Shift is to Spec-v*, it must move on to a hi g her position. This v iew raises two problems pointed out by Chomsky himself. First, the issue of cou ntercy cl ic i ty : under strict cyclicity, SU in SPEC raises to Spec.T before the operation is authorized by vacating the outer Spec,v P, at the subsequent CP phase. Second, the issue of backtracking to rescue a failed derivation along a different path: for instance, cancell i ng O bjec t Shift to HP if the position is not later evacuated. Chomsky offers a partial solution (Assumption 4 below) which only solves the first puzz l e but not the second one.
2.4. Assumption 4: Procrastinating Spell-Out In the configuration (4), repeated in (5), Spell-Out applies only at the next higher strong Ph ase ZP, MLC too is evaluated at ZP. Specifically, if HP and ZP i n (4) are yP and CP respec tiv el y then it is only at CP (=ZP) that the phonological content of XP, hence the phonological-edge status of SPEC, is determined. ,
lzp . . . . P. . ··LHP
(5)
XP [SPEC lH
YP I1J
The countercyclicity puzzle is thus solved: SU in SPEC is free to raise to Spec,TP over XP ·the shifted OB- with p honologi cal content, because Spell-Out is at the next strong phase ZP, and MLC too is evaluated at this stage of the derivation. Yet, the problem of backtrac king to recover a failed derivation along a different path remains, as noted by Chomsky himself. Moreover, the Assumption 4 above leads to an inescapable conclusion (6). The strong phase status of yP is useless, hence dispensabl e .
(6)
In the following section, we present empiri cal evidence for (6).
3. High applicatives and the strong phase status of yP It is standardly assumed that subject and object of a transitive v erb are first merged in the same yP domain, as schematized in (7). (7)
vP
-----______ y ------
SU
ROOT
268
08
Juvenal Ndayiragije
I will argue for a slightly different configuration (8) in which subjects are generated outside the yP domain. If (8) is correct. then the strong phase status of yP becomes irrelevant. XP
(8)
-----SU _____
' � � , '_
X
.y.P
-----SPEC ______ y -----ROOT
OB
Sentences (9b-c) illustrate applicative constructions in Kirundi. formed from (9a). (9 )
a.
ibitabo Yohani a-a-ra-rungik-ye John 3s-pst-AF-send:perf books 'John sent books'
b.
Yohani a-a-ra-rungik-k-ye umwana ibitabo John 3s-pst-AF-send-APPL-perf. child books 'John was sending books to the child'
c.
Yohani a-a-ra-rungik-it-ye ibitabo John 3s-pst-AF-send-APPL-perf books (same reading)
(BEN-TH)
umwana (TH-BEN) child
Note first that the applied (benefactive) argument is dependent upon the presence of the underlined applicative morpheme -k- suffixed to the verbal stem. As a matter of fact, dropping the latter from the verb leads to ungrammaticali ty. as illustrated by ( 1 0). (10)
* Yohani a-a-ra-rungik-ye umwana John 3s-pst-AF-send-perf. child 'John was sending books to the child'
ibitabo books
(BEN-TH)
Furthermore, as shown in (9b-c), the word-order between the applied argument (BEN) and the direct object (TH) seems variable. Yet, there is clear evidence that BEN c-commands TH. First, in passive constructions, only BEN may raise to Spec,T to value T's Case and phi-features, as shown by ( l Ia) vs.(* 1 2a). Note that cliticization of the non-raised argument is possible for TH ( l I b), but not for BEN ( l 2b), an indication that ( 1 1 a) is not an inverse construction (see section 5 for details on the latter).
269
On Phases and Cy cl icity
( I I)
a.
b.
( 1 2)
u mwana
a-a ra-rungik- ir-u-ye 3s-pst-AF-send-APPL-PASS-perf 'The child was sent books' -
child
ibitabo
books
umwana a-a-ra-m.-rungik-ir-u-ye. (TH-c1 iticization) 3s-pst-AF-OM-send-APPL-PASS-perf. child 'The child was sent them'
a.
umwana * ibi labo bi-a-ra-rungik-ir-u-ye books 3p-pst-AF-send-APPL-PASS-perf. child 'books were sent to the child'
b.
* ibilabo bi-a-ra-mu-rungik-ir-u-ye
(BEN-cliticization)
books
3p-pst-AF-OM-send-APPL-PASS-perf. 'books were sent to him' Second. a quantifier BEN Q-binds a possessive pronoun within TH on ly if BEN precedes TH, as illustrated by ( l 3a). If not, Q-binding is barred ( l 3b). Furthermore. a quantified TH is incapable of Q-binding a possessive pronoun within a preceding BEN, as shown by ( 1 3c): ( 1 3)
a.
Yohani a-a-ra-rungik-ir-ye
John
I ibitabo vyiwed books his
'lohn was sending every child i his books i
b.
umwumwe]i each
[umwana 3s-pst-AF-send-APPL-perf child '
Yohani a-a-ra-rungik-ir-ye[ibitabo vyiwed [umwana umwumweL John V books his child each 'Johnj was sending hisi books to every child j c. Y ohani i a-a-ra-rungik-ir-ye
[umwana wiweJi
child John his V I umuvyeyi u mwumwe)j every parent 'John i was sending every paren� hi s il&j child I
For those two reasons alone, it is clear enough that the applied (BEN) argument c-commands the direct object (TH). as depicted in the partial structure below:
270
Juvenal Ndayiragije
( 1 4)
VoiceP SU
� � voice
ApplP
� BEN � Appl
yP
� Spec � y. � ROOT
TH
If ( 1 4) holds. and indeed it does, an immediate conclusion follows: the strong phase status of y.P is irrelevant, useless. hence dispensable. As a matter of fact, within an MLC-based approach to cyclic derivations, if SU in ( 1 4) is demoted by passivization. the applied argument (BEN), but not the direct object (TH), is the next closest goal to be probed by T. This nicely accounts for the contrast of grammaticality between (1 l a) and ( 1 2a). On the other hand, under a PIC-based approach, some unmotivated assumptions must be introduced to rule out ( 1 2a). One of them would be to say that ApplP is not a strong phase. Accordingly, the direct object (TH) cannot reach ApplP's phonological edge, a PIC-precondition for TH to be probed by T and raised to Spec,T. Such an assumption would be hard to buy. though. As a matter o f fact, if strong phases are those that have an acc (=EPP) feature, it would be unfair to grant a strong phase status to y'P, while denying it to AppiP. 4. TP as a strong phase: a view from Compound Tenses
Sentences ( I 5b-c) constructions. ( 1 5)
a.
b.
c.
illustrate
Kirundi
compound
tense
(henceforth
en
ibitabo Yohani a-§1S-ra-soma John 3s-pres-AF-read books 'John reads books' Yohani a-a-riko hp t' a-§1S-ra-soma John 3s-pst-be at 3s-pres-AF-read 'John was reading books' Yohani a-oo-ba John 3s-cond':'be hp ttl a-a-riko[TP t' a-§1S-ra-soma 3s-pst-be. at 3s-pres-AF-read 'lohn might have been reading books'
ibitabo) books
ibitabo]] books
Importantly, the bracketed sentential complement displays overt tense-marking and subject-verb agreement. Indeed, ( 1 5b) is ( 1 5a) plus an auxiliary verb, and ( l Sc) is ( I Sb) plus one more auxiliary. This leads to the following conclusion:
27 1
On Phases and Cyclicity
the bracketed complement in ( I Sb-c) is at least a TP, an inescapable conclusion which has an i mportant consequence on phasehood, that I return to shortly. Furthermore, the bracketed complement is not a CP, for the following two reasons, among others. First, passivization of the direct object is possible i n ( l S b-c), as ill ustrated b y ( l 6b), formed from (1 6a). The raised object to the highest Spec,T agrees with both the aux iliary and the main verb, like the subject of active sentences in ( l Sb-c). ( 1 6)
a.
b.
Y ohani a-a-riko hp t' a-f/S-ra-soma John 3s-pst-be at 3s-pres-AF-read 'lohn was reading books'
ibitabo] books
(na Yohani) ibitabo bi-a-riko lTP t'bi-f/S-ra-som-u-a books3p-pst-be at 3p-pres-AF-read-PASS-a by John 'books were being read (by John)'
( 1 7a) contains a true embedded CP complement ( 1 7b) is an ungrammatical passive sentence in which the direct object of the embedded CP is raised to matrix Spec,T.
( 1 7) a.
b.
Yohani a-a-vug-a rep ko fTP ba-f/S-soma John 3s-pst-say-asp C 3p-pres-read 'John was saying that they read books' * ibitabo bi-a-vug-u-a books 3p-pst-say-PASS-asp [ep ko [TP bi-f/S-som-u-a C 3p-pres-read-PASS-a
(na by
ibitabo] books
Yohani»)) John
The contrast of grammaticality between ( 1 6b) and (* 17b) clearly shows that the sentential complement in ( 1 6a) is not a CPo Second, in idioms, the CT construction does not destroy the idiomatic interpretation of the sentence, as illustrated by ( 1 Sb), from ( 1 Sa). A standard property of raising constructions. ( 1 S)
a.
b.
ibisaka vy' imbwa bi-f/S-ra-hi-ye hidings of-dogs 3p-pres-AF-burn-perf. [Lit: 'Dogs'hiding-places are burning' ] ' We are close to surrender' ibisaka vy'imbwa bi- a-riko [TP t' bi-f/l-ra-hi-a] hidings of-dogs 3p-pst-be at 3p-pres-AF-burn-imperf. ( Lit: 'Dogs'hiding-places were burning' ] ' We were close to surrender'
From the facts in ( 16- 1S), it follows that the sentential complement is not of the CP category. With this conclusion in mind. let us now return to feature valuation. As illustrated again in ( l 9a-b). the CT construction is a clear case of multiple Probes (T) v alued by a single Goal, the subject.
272
Juvenal Nday iragije
( 1 9)
a.
b.
Petero a-a-riko I TP t' a-i1S-ra-andika igitabo] Peter 3s-pst-be at 3s-prcs-A F-write book 'Peter was reading a book' a-oo-ba 3s-cond-be a-a-riko I TP t' a-i1S-ra-andika [TP til 3s-pst-bc at 3s-pres-AF-write 'Peter might be writing a book' Petero Peter
igitaboJ] book
Multiple agreement directly fol lows from cyclic Spec-to-Spec raising of the subject, as schematized below: (20)
TP
�
SU i
� T
vP
AGREE� v
AUX
TP
� � yP T AGREE ______ I'; ______ y �
t" ;
ROOT
OB
An important question arises from (20) regarding phi-completeness of lower T. In Chomsky 's system, strong phases must have an OCC-feature but that is not enough; they must be phi-complete too. Thus, 'y' is phi-complete, the same is true of C-T, that is, T selected by C, either finite T or non-finite T of control but not raising constructions. In (20). lower TP is not selected by C, as shown above, but it is clearly finite . The question then is whether or not lower TP is phi-complete. If so, it is a strong phase. In order to demonstrate that lower TP is indeed phi-complete. we have to find evidence showing that Agree between SU and T in lower TP values the un interpretable Case-feature of SU and the phi-features of T. Fortunately, that evidence does exist. As a matter of fact, Kirundi displays a clear agreement distinction between Nominative and Accusative Case valuation. as illustrated below. (2 1 a) is an intransitive sentence. The subject agrees with the verb. The agreement marker precedes the tense marker. (2 1 b-c) show that the subject may be dropped (21 b) but not the agreement marker (2I e).
273
On Phases and Cyclicity
(2 1 )
a.
Petero a-�-ra-tamb-ye Peter 3s-pst-AF-dance-asp 'Peter danced'
b.
pro
c.
a-a-ra-tamb-ye 3s-pst-AF-dance-asp 'He danced'
*
Petero a-ra-tamb-ye Peter pst-AF-dance-asp
(22a-c) are transitive sentences. In (22a), the subject is the second person singular pronoun whose features are spelled-out by the subject marker. In (22b), a weak pronoun replaces the object in (22a) and appears next to the verbal stem, in a position different from that of subject markers in (2 1 ). Finally, in (22c) the direct object is a second person singular pronoun, and its morphological form is clearly distinct from the one displayed in subject position (22a-b). (22) a.
b.
pro u-a-ra-bon-ye
Petero 2s-pst-AF-see-asp Peter 'Y ou saw Peter'
pro u-a-ra-mu-bon-ye.
2s-pst-AF-3s-see-asp 'You saw him'
c.
pro a-a-ra-ku-bon-ye.
3s-pst-AF-2s-see-asp 'He saw you'
From the facts above, there is no doubt that Nominative and Accusative Case valuation have morphological instantiation. With that conclusion in mind, consider now CTs i n (23a-b) which are nothing other than (2 1 a) and (22a) with an additional auxiliary. (23)
a.
b.
274
a-a-ri [TP t1 Petero 3s-pst-be Peter 'Peter had danced'
a-a-ra-tamb-ye] 3s-pst-AF-dance-asp
n-a-ri [TP ( u-a-ra-bon-ye Petero] 2s-pst-AF-see-asp Peter 2s-pst-be 'Y ou had seen Peter'
pro
'
Juvenal Ndayiragije
Interestingly enough. the subject-verb agreeement marker on lower TP in (23a b) is exactly the same as the one in (2 I a) and (22a), respectively. Therefore, Nominative Case feature of the subject (SU) in (20) is valued upon agreement with lower Probe T first. Accordingly . lower TP is phi-complete6• This leads to two inescapable conclusions: (24) a. b.
Phi-completeness of T is nru. incumbent on a selecting phi complete C Finite TP is a strong phase alone, contra current Phase Theory
Under a PIC-based approach to cyclicity, some undesirable consequences follow from (24b). If Phase-level evaluation (5-0) must apply at the next strong phase; that is, high TP in (20), then the latter must have an extra SPEC (aCC) to be used as an escape hatch for wh-phrases to reach SPEC-C without violating MLC at this strong phase. On empirical grounds, PIC predicts that in Kirundi wh-movement, the high T in (20) might agree with a wh-phrase en route to SPEC-C, and the low T by a subject raised and stranded in SPEC of lower TP. The prediction is not borne ouL The next section presents two other undesirable consequences of PIC. 5. The strong phase status of CP and Inverse in Obligatory Control
5.1. On Inverse
(2Sa) illustrates an SVO transitive sentence. The verbal morphology includes a subject marker and the underlined particle -ra- that follows Tense. This marker is dubbed "antifocus" in Ndayiragije ( 1 999). (25b) is (2Sa) minus the antifocus marker. In this case, the direct object receives a contrastive focus reading. (25)
a.
Yohani a- a-m-som-ye ibitabo 3s-pst-AF-read-perf books John 'John read books'
b.
Yohani 8- a-som-ye ibitabo John 3s-pst-read-perf books 'John read books (e.g. not newspapers)'
SVO
(26a) illustrates an inverse construction. Here, the verb agrees with the preverbal object Moreover, the antifocus marker is precluded and the logical subject receives a contrastive focus reading.
I> 1 wish to emphasize that the bracketed complement of Cfs in (23a-b) is a tensed TP in Kirundi.
not some other category such as Asp. as suggested by Carstens (2001 ) for KilegalKiswahili. and
others for languages like Italian or Icelandic. In (23a-b). both T and Asp are overtly realized on V in the bracketed complement and occur in
V position for T.
two distinct positions: the final position for Asp and the pre
275
On Phases and Cyclicity
(26b) illustrates a transitive expletive construction (TEe). The latter shares two important properties with inverse (OVS) in (26a). First, TEe assigns a contrastive focus interpretation to the logical subjecL Second. it precludes the presence of the antifocus marker -ra-. Yet, TEC (26b) and OVS (26a) differ with respect to verb-agreement. Whereas the verb agrees with the preverbal object in OV S, in TEe, it agrees with an empty pro subject of the there-type. This is expressed by the locative marker -ha-prefixed to the tense marker, the same position as for the agreement marker of the raised object in (26a). (26c) shows that (26b) becomes ill-formed if the verb agrees with either the logical subject or the direct object. (26)
a.
ibitabo bi-a-som-ye Yohani. books3p-pst-read-perf John [Lit: Books read John] 'JOHN (e.g. not Paul) read books'
b.
pro
c.
* pro
ha-a-som-ye ibitabo Yohani. Loe-pst-read books John lLit: There read books John] 'JOHN (e.g. not Paul) read books' albi-a-som-ye AGRSIO-pst-read
ibitaboYohani. books John
Inverse (OVS)
TEC (Exp-VOS)
TEC (Exp-VOS)
5.2. Analysis In Ndayiragije ( 1 999), I argued for a unified analysis of OVS and TEC according to which the logical subject in both cases undergoes A-bar movement to SPEC7 of a Focus projection (FocP) located i n between TP and yP, as depicted in (27):
7
To account for the word order in (26a-b), I stipulated that this Spec is on the rightside of its head Foe, a consequence of still unclear PF properties of theme-rheme conslrUctions. An alternative solution to the word-order problem would be remnant yP-movement above FocP as suggested to me by Chris Collins. Yet, the trigger is still lacking. Anyway, nothing in the present 'discussion hinges on the word-order issue, hence I will leave it aside.
276
Juvenal Ndayiragije
(27)
TP
------SPEC _______ T
FoeP
------SU i _______ Foe
y.P
------Ij _______ Y. � ROOT
' OB
After A '-movement of SU to the TP-internal Focus position, either OB raises to Spec,T and this gives rise to an inverse construction (26a), or a there-type empty expletive is merged in Spec,T. and we obtain a TEC (26b). For a detailed discussion of empirical arguments supporting (27), I refer the reader to Ndayiragije ( 1 999). I now turn to an interesting case of inverse that is directly related to PIC and the assumed strong phase status for CPo 5.3. Long Distance Inverse in Obligatory Control Let us first recall the two main minimalist approaches to OC. The first approach argues that there is no PRO in Obligatory Control constructions. Thus, Wurmbrand ( 1998) suggests that in OC the matrix verb selects a VP complement Manzini & Roussou ( 1 998) eliminate PRO in OC by suggesting to view a-roles as aspectual features associated with a DP in the inflectional domain. Hornstein ( 1 999, 2000) claims that OC involves DP Movement (raising) and that a-roles are features valued via DP-Movement. The second approach, followed notably by Landau ( 1 999, 2001), adopts a more classical view, by claiming that PRO is present throughout. I show below that Landau is right in assuming an empty category PRO in OC. Consider first (28a-b). (28a) is an DC construction. The matrix verb agrees with the matrix subject. The bracketed sentential complement contains a silent subject, an infinitival verb and its object. On semantic grounds. the silent external argument of the embedded verb must be coreferential with the matrix subject. (28b) is an inverse (OVS) construction. The entire sentential complement in (28a) is raised to the preverbal position and the logical subject appears postverbally. (28)
kugura Inf-buy
a.
Y ohani a-jil-ra-ifuza [PRO John 3s-pres-AF-wish 'John wishes to b�y a book'
b.
[PRO kugura igitaboJ bi-jil-ifuza AGRCP-pres-wish Inf-buy book 'JOHN (not Paul) wishes to buy a book'
igitabo] book Yohani. John
277
On Phases and Cyclicity
The subject-verb agreement marker bi- in (28b) appears when a (non-)finite CP or an it-type pro expletive fills the subject position. as ill ustrated by (29a-c), respectively: (29)
a.
b.
c.
inzu] I cp ko Yohani a-f1)-gura C Yohani 3s-pres-buy house bi-fiS- ra-shobok-a. AGRCP-pres-AF-be possible-asp That JOHN buy a house is possible' [ PROarb kugura inzu] bi-f1)-ra-shobok-a. Inf-buy house AGRCP-pres-AF-be possible-asp 'Buying a house is possible' pro bi-fiS-ra-shobok-a [cp ko AGRCP-pres-AF-be possible-asp C 'It is possible that John buy a house'
Yohani a-fI}-gura Yohani 3s-pres-buy
The agreement similarity between (28) and (29) clearly indicates that the sentential complement in (28a) is of the CP category, not a V P (contra Wurmbrand ( 1 998)H. Furthermore, the fact that that sentential complement is movable suggests that it is not an IP (contra Boskovic 1997). Furthermore, the following facts suggest that the inifinitival complement in (28a) contains PRO. First. we have already seen that the silent external argument of the infinitival complement in (28a) must be coreferential with the matrix subject. (30) shows that that restriction does not apply in inverse (28); that silent argument is free to receive an arbitrary reading. (3�)
[ PROarb kugura inzu] bi-f/S-sab-ye . Yohani. Inf-buy house AGRCP-pres-ask-asp John 'JOHN (not Paul) asked for some people to buy a house'
Second, as shown in (3 1 ). the raised infinitival may include a reflexive anaphor.
8
Note that Kirundi VPS cannot be fronted (which suggests that they are presumably nol VP but non tensed TP). as shown by the ungrammaticality of (ib-c), fanned from (ia) by topicalization and left dislocation/elitie-doubl ing. respecti vely. (i)
278
a.
Yohari a-a-n [kugura inzu] John 3s-pst-be Inf-buy house 'John would/should/could buy a house'
b.*kugura
Inf-buy
inzu. house
Yohari John
a-a-ri. 3s-pst-be
c.*kugura Inr-buy
inzu. house
Yohani John
a-a-bi-ri. 3s-pst- �LlAQRCP-be
inzu] house
Juvenal Ndayiragije
[PROarb ku-i-shira m u ngorane] Inf REFL p ut in trouble bi-0-sab-ye Yohani. AGRCP-pres-ask-asp John 'JOHN (not Paul) asked for some people to put themselves in trouble'
(3 1 )
-
-
The well-formedness of (3 1 ) and the availability of an arbitrary reading in inverse (30) but not in (28a) are mysterious if OC constructions do not contain PRO. Put together, the facts above lead to the conclusion that the infinitival complement in OC is a CP, and that the latter contains an empty category PRO acting as the goal G of the non-finite C-T (Probe), without phonological content, . as predicted by feature-matching under Agree. With that conclusion in mind, we can now look at long distance inverse in Oc. This is illustrated in ( 32b) from (32a). ,
(32)
a.
b.
Yohani a-a-emer-ye [cp PRO kwandika John 3s-pst-agree-asp Inf-write 'lohn agreed to write a book' igitaboj ki-a-emer-ye lcp PRO kwandika book 3s pst-agree-asp Inf-write 'JOHN (not Paul) agreed to write a book' -
igitabo] book
ti J Yohani John
In (32b), what has been promoted to the matrix Spec,T is the direct object, raising from the embedded CP complement. The raised object agrees with matrix T. (32b) is what I refer to as "long distance inverse in OC". Interestingly enough, a PROarb reading is absolutely impossible in (32b). Likewise, control of PRO by the raised object is barred. PRO in (32b) is rather obligatorily controlled by the postverbal logical subject, exactly as it has to be in non-inverse OC (32a), but not in CP-raising to Spec,Tin (3 1). Furthermore, "Iong distance inverse in OC" can be replaced by external Merge of an expletive into Spec of matrix T, giving rise to a transitive expletive construction (TEC). An illustration is given in (33), formed from (32a): (33)
pro ha-a-emer-ye
Icp PRO kwandika igitabo] Yohani. Loc-pst-agree-asp Inf-write book John [Lit: "there agreed to write a book John"] 'JOHN (not Paul) agreed to write a book'
Moreover, like inverse (32b), TEe (33) does not allow a PROarb reading. Only obligatory control of PRO by the logical subject in the right periphery is permitted. Finally, the logical subject of the matrix cannot appear in the position filled by PRO i n (33), as (34) shows.
279
On Phases and Cyclicity * pro ha-a-emer-ye lcp Yohani kwandika igitaboJ. Inf-write book Loc-pst-agree-asp John I Lit: "there agreed to write a book John" ] 'JOHN (not Paul) agreed to write a book'
(34)
The ungrammaticality of (34) challenges Hornstein's ( 1 999, 2000) DP movement analysis of OC. Any attempt to save the latter by appeal ing to technicalities such as obligatory raising out of control infinitives or mismatch between phi-features of the lexical DP John in (34) and the non-finite T (probe) leads to other complications that I will not discuss here. On the other hand. a PRO-based approach to OC predicts (*34). Furthermore. the control asymmetries between (3 1 ) on one side and ( 32b-33) on the other side are exactly what our analysis of inverse (OVS) and TEC in (27) predicts. Let us see how. Consider (3 1 ) first. According to the analysis in (27), the raised CP complement A-raises to Spec,T. Incidentally. it is standardly assumed that A-movement does not reconstruct Therefore. PRO in (3 1 ) is not c commanded anymore by the logical subject in postverbal position, hence is freed of obligatory control. Consider now (32b). Here. what has raised to Spec of matrix T is the direct object. not the entire embedded CP that contains it. According to (27), the logical subject in inverse A-bar moves to the (rightward) Spec of Foc. In (32b). what has moved to Spec,FocP is the external argument John of the matrix verb; the same is true of (33). �econstruction being a trademark of A-bar movement, the analysis in (27) correctly predicts obligatory control of PRO by the trace in matrix Spec.vP of the A-bar moved logical subject in (32b) and (33). Those control facts provide additional support (not d iscussed in Ndayiragije 1 999) for the A-bar movement analysis in (27). Any alternative should not ignore them. Coming back to the relevance of inverse to the Phase Theory, we saw that long distance raising out of a non-finite CP is possible in Kirundi. as illustrated by (32b). On the other hand, as (35b) shows from (35a), such a movement is impossible out of a tensed CPo Note that SpectCP and Spec,T have no phonological content, hence PC is not at work here. (35)
a.
b.
Yohani a-a-emer-ye [cp ko proa-zo-kwandika igitabol John 3s-pst-agree-asp C 3s-zoo-write book 'John agreed that he will write a book'
* igitaboj ki-a-emer-ye [cp ko pro a-zo-kwandika til Yohani book 3s-pst-agree-asp C 3s-zoo-write John 'JOHN (not Paul) agreed that he will write the book'
An inescapable conclusion follows: (36)
CP is a strong phase only if selecting a finite T
This clearly goes against Chomsky' S proposal that CP is an intrinsic phase, and that TP's phasehood. if any, is dependent upon selection by C. We just saw
280
Juvenal Ndayiragije
ev idence showing exactly the opposite view; that is, the strong phase status of CP is l ikely inherited from T's finitess.
5.4. Some undesirable consequences First, if finite CP is indeed a strong phase and the phonological condition (PC) holds, then the PIC-based approach to cyclicity fails to account for the absence of long distance inverse out of finite CPs (*35b). Second, if non-finite CP is a strong phase and PIC holds, then long distance inverse in OC (32b) involves improper A-movement of OB 'book' to SPEC-T through embedded (A-bar) SPEC-C. Third and finally, neither PIC nor PC is equipped enough to account for the availability of long-distance inverse in (32a-b), repeated in (38a-b), but not in (39a-b), where an applied argument has been added to the matrix 'verb of (38a-b): (38)
a.
b.
(39)
a.
b.
Yohani a-a-emer-ye [ep PRO kwandika 3s-pst-agree-asp Inf-write John 'lohn agreed to write a book'
igitaboJ book
igitabo; ki-a-emer-ye [ep PRO kwandikat; Inf-write 3s-pst-agree-asp book 'JOHN (not Paul) agreed to write a book'
) Yohani John
Yohani a-a-emer-ir-ye abana John 3s-pst-agree- APPL-asp kids fep PRO kwandika igitabo] Inf-write book 'John agreed tolfor kids to write a book' * igitabo; ki-a-emer-ir-ye abina 3s-pst-agree-APPL-asp kids book lep PRO kwandika t;] Yohani Inf-write John 'JOHN (not Paul) agreed tolfor kids to write a book'
The ungrammaticality of (39b) is straightforwardly accounted for under MLC (3) alone. The applied argument prevents Match of acc feature of matrix T and Goal OB out of the embedded clause as much as it does in high applicative constructions discussed in section 3. Nothing else needs be added.
6. Conclusion From the facts discussed in this article, it appears that all of the imperfections and complexities arising from a PIC-based approach to cyclicity disappear entirely under a recursive "Single-Cycle Search" (SCS) derivation, minimized (i.e.delimited) by finiteness, and constrained by MLC, an independent locality condition on both Agree and Internal Merge. If this proposal is on the right track, then the current Phase Theory is dispensable.
281
On Phases and Cyclicity
References Boskovic, Zeljko . 1 997. The Syntax of Nonfinite Complementation: An Economy Approach .
Bowers. John. 200 1 .
MIT Press. "On Reducing Obligatory Control to Movement". Ms. Cornell
University. Ithaca.
Carstens, Vicki. 200 1 . "MUltiple agreement and case deletion: Against phi (in)completeness". Syntax. 4.3. Chomsky, Noam. 200 1 . "Derivation by Phase" In Ken Hale: A life in language, ed. M. .
Kenstowicz. MIT Press.
Chomsky, Noam. 1 995. The Minimalist Programme. MIT Press. Hornstein. Norbert. 1999. "Movement and Control", LI 30: 69-96 Hornstein, Norbert. 200 1 . Move! A Minimalist Theory of Construal. B lackw ell Landau, Idan. 1 999. Elements of Control. PhD dissertation. MIT Landau. Idan. 200 1 . "Control and Extrapositi on, The case of Super-Equi". NLLT 19: 1 091 52 Manzini, Rita and Anna Roussou. 1999. "A Minimalist Theory of A-movement and Control". Lingua I t o
Ndayiragije. Juvenal.2000. "Strengthening PFn. Linguistic Inquiry 3 J :3, 485-5 12 Ndayiragije, Juvenal. 1 999. "Checking Economy ". Linguistic Inquiry 30:3. 399-444 Rizzi, Luigi. 1 982. Issues in Italian Syntax. Dordrecht: Foris Wurrnbrand, Suzanne. 1 99B. lnfinitives. PhD. Dissertation. MIT Department of French
University of Toronto Scarborough Campus 1 265 Military Trail, Toronto ON. Canada MIC lA4
[email protected]
282 "
Derivations without the Activity Condition· Andrew Nevins Massachusetts Institute a/ Technology The EPP cannot be reduced to the Inverse Case Filter (contra Boskovic 2002) a nd remains for the moment axiomatic in phrase-marker construction. Feature-splitting, Agree. and th e EPP al low for a range of derivational options that have been constrained by the Activity Condition (Chomsky 2001 et p req). The phenomenon of non-nominative subjects (including A-movement of structuraJIy-Case marked elements) runs afoul of the activity condition. This paper explores the hypothesis that all derivation al options ruled out by the activity condition can be ruled out by independent princ iples of locality and a constraint against mu lti ple Case valuation.
1.
Movement in a Post-Agree Landscape
The major conclusion of this paper will be the proposal that the Activity Condition is not necessary nor desirable constraint on derivations. Prior to the discussion of the Activity Condition however, a few fundamental questions about the nature of displacement in syntax will be addressed and discussed at length. Why do OPs move from their theta position to a functional head FO? Two relatively uncontroversial answers used to be to check agreement features on FO, or to be assigned Case (which FO can do)l . However, with the introduction of Agree, a feature-valuation operation "at a distance" (Le. under closest c command), there . is no need for actual movement (i.e. re-merger: creation of a new sisterhood relation) to accomplish feature valuation. Schematically, valuation of agreement features on Tense occur purely under Agree with the postverbal OP in ( 1 ) :
(1 )
There seem to be two fish in that tank (seem > 2, ? 2 > seem)
The associate, twa fISh, can't be "moving at Logical Fonn" since its scope with respect to the intensional verb would have to be obligatorily wide in that case, * The conclusions reached in this paper can be chrono l ogi cal l y traced as the result of conversations in chronological order: Milan Rezac, Karlos Arregi. Pranav Anand. Shigeru Miyagawa, Joh n Bailyn. Howard Lasnik . Usama Soltan, Noam Chomsky. Brent DeChene, Julie Legate, Klaus Abels, Norbert Hornstein. and Cilene Rodrigues. In ad dition. audiences at the MIT Phase Workshop (Jan 2003) and at Kanda University (Oct 2003) and Tsukuba Un iversity (Oct 2003) were vel)' encouraging and thought-provoking. I d oubt thaI any of the aforementioned parties are in full agreement with the conclusions I have drawn. I As the locution "It is the case that. . ... or "It is not the case that" is highly prevalent in this paper I . will capitalize all instances of Case when refening to the technical tenn.
MIT Working Papers in Linguistics 49. 283-306 Perspectives on Phases CO 2005 Andrew Nevins
f"
Derivations without the Activity Condition
counter to fact. Similar arguments (due to H. Lasnik, among others) against remerger of the DP at any conferred-upon level of representation can be mustered from the inability of t�e postverbal DP to bind an anaphor license a negative polarity item (3),
(2) or
(2)
* There seem to each other to be some biolinguists making progress.
(3)
* There seem to any of the biolinguists to be no philologists making progress.
The evidence that features can be valued at a distance (whether ultimately modeled with Agree or another operation) demonstrates that feature-valuation
cannot be what drives movement of a OP to FP as Agree should always be an
option. There are two less-conferred upon answers as to why OPs move (to A positions). The Inverse Case Filter answers that FO needs to assign Case to something (in a spec-head relation, no less). The EPP answers that there is a structural requirement that the immediate projection of FO is in a sisterhood relation with a DP. As the Inverse Case Filter is stated in tenns of the
requirement for an overt specifier-head configuration, the only divergent predictions from the EPP pivot upon whether or not the moved OP is actually assigned Case in the specifier of FO. A range of evidence to be discussed below demonstrates that the re-merged DP is not assigned Case by FO, The EPP in its pure form remains as the best answer to our opening question. Whether or not this is a sad conclusion for the hypothesis that movement is only driven by uninterpretable feature-checking is an evaluation left for the reader. We tum to the arguments that Movement can occur without Case, Agreement, or Inverse Case valuation. 2.
The Inverse Case Filter does not Subsume EPP Effects
The Inverse Case Filter (ICF henceforth) is "the requirement that traditional Case assigners assign their Case features" (Boskovic 2002: p. 1 70; see also Martin 1 999 for an earlier fonnulation) in an overt spec-head relationship. Thus.
(4) is modeled as illicit because fmite matrix T has not assigned nominative Case to anything. (4)
* Is likely that Agboh helped Mbaya.
The failure of matrix T to assign nominative Case in (4) poses no legibility problems for the LF interface. as the Visibility requirement (Chomsky 1 98 1 ) - that all arguments of all predicates bear Case -- is met (even by the CP, though not morphologically visible). It could be that the unmet lexical requirement on T to assign Case renders it ineligible for phonological spellout at PF. but I prefer to view the ICF as a syntax-internal constraint within models that posit it.
(5)
284
* was told Njoroge that Assefa left.
Andrew Nevins
(6)
* Qasba believes to have been told Sasha that Yves left.
After all, (5) shows that nominative Case assignment to the OP object isn ' t enough: Nominative must b e assigned i n a configuration resulting from overt movement2. (6) shows the same for accusative: it cannot be assigned at a distance in order to satisfy the ICF, but requires displacement. Why should PF be unable to spellout a predicate that has discharged its Case feature, but not in a spec-head configuration? The answer remains obscure; hence, let's consider the ICF purely syntax-internal . Note that it still does not subsume all EPP effects. Though (4)-(6) are ruled out by the EPP, there is a class of derivations that the EPP can rule out that the ICF can't: those without successive-cyclic movement through intermediate specifiers of TP. Evidence for this intermediate movement is easy to find: in (7), neither the surface nor the theta-position are suffjcient to bind the anaphor, but the intermediate specifier ofnonfmite T is3:
(7)
Thelonious seems to Mrs. Monk to appear to himself to be a genius
Thus, above and beyond the ICF, a representational principle ensuring successive-cyclic movement is required in models that attempt to model derivations without the EPP. Suppose we make this move, temporarily. What about (8), an Icelandic sentence in which there is no nominative Case being assigned?
2 The lCF may also be employed to rule out instances of there-existentials wirhout an associate, e.g. *There seems that Kate Icnows everyth ing . However, sentences like rhese independently crash due to lack of a DP to val ue the phi-features of matrix T. An leF-based explanati on in terms of O vert Case assignment adds nothing to this independent sy,ntactic requirement. In addition, blocking effects (the availability of an it-expletive variant) and perhaps semantic requirements on there-existentials are
additional sources of explanation.
3 Grohmann, Drury, and Castillo (2000) attempt to provide an alternative explanation for (7), in which there is nO intennediate A-movement.. but rather, that himself is a logophor in the complement of appear. citing the following sentence as evidence (with the judgement that it is grammatical):
(i)
John kissed Mary because it appeared to himself that rhe earth was flat
I find the status of thi s sentence egregious, as do all of the people that I have asked. In any event,
more nuanced arguments for successive-cyclic movement can be found with quantifiers. Sauerland
(2003) uses data such as the following to argue for successive-cyclic movement through
(must > neg > every)
(i)
Every student mustn't get an A.
(ii)
Every child; doesn't seem to hisj father to be smart. ( neg > every)
[. .
.
vP:
At most a third of them can get one.]
(i) rules out the possibility that when ne gation scopes over a universaJly-quantified subject, it is due to neg -rai sin g , rather than reconstruction. as negation cannot scope above the modal. (ii) shows that the universally-quantified subject can reconstruct below negation. to a position still high enough to
bind the experiencer pronoun. (II also cannot be the case that the lowest copy of every child is
Quantifier-Rai sin g to bind the pronoun, as that would constitute a weak-crossover violation). The LF
posi ti on, then. must be an i ntcnnediate projection, which Sauerland argues to be specifier of the v
above seem, with movement triggered by an EPP feature.
285
Derivations without the Activity Condition
(8)
a.
Okkur var hjelpao We-dat was helped 'We were helped'
b.
1>ao logaoi a kertinu There flamed-3.sg at the.candle 'There flamed at the candle'
To maintain the ICF as operative in (8a), we must analyze it as follows: " Quirky subjects have a structural nominative, not morphologically realized, on top of the inherent Case" (Boskovic 2002, p. 17 1). (8b) represents a class of unaccusatives which have no argument capable of bearing Case (Sigurdsson 1 989/1 992: chapter VI), not even a quirky one. Perhaps a similar claim could be made, that there bears nominative Case in these constructions. Suppose we make these moves, temporarily, to keep the I CF afloat, and assume, contra morphological evidence, that datives and expletives actually bear an invisible nominative Case. Consider now structures in which the only DP in the clause is postverbal: (9)
Into the room rolled the basketball
Let's not enter into the debate about the ultimate surface position (A or A') of the locative; it's irrelevant for the present discussion. But recall that nominative Case must be assigned overtly (i.e. in a Spec-Head configuration) on the ICF account. For (9) to converge, the matrix T must assign nominative Case overtly . Hence it would have to be to into the room for the ICF to remain empirically correct. The conclusion is thus forced that the preverbal PP bears nominative Case, the lone instance of a preposition bearing nominative Case in the literature. It cannot be the case that the basketball ha� checked nominative Case overtly in Spec, T (say, with subsequent verb-raising even higher). Coopmans (1 989) demonstrates that it is not structurally high enough to control into an adjunct ( 1 0), while canonical preverbal subjects are ( 1 1 ): ( 1 0) (1 1 )
* Near the oasis lay two sheiks [without PRO talking] Two sheiks lays near the oasis [without PRO talking] .
The postverbal OP cannot be receiving nominative Case on the ICF account, as it never targets a high enough position for a spec-head configuration with T. Hence, the next question is, what Case is assigned to the basketball in (9)1 Notice that the ICF, as stated, requires overt movement for accusative case as well (cr. (6) above). But it can't be the case that the postverbal subject in (9) is moving overtly to, say, AgrO or v, as that would license pseudogapping (if we follow Lasnik's account). (1 2)
286
* Into lohn's office rolled a basketball, and into Mary's did a volleyball
Andrew Nevins
Pseudogapping ( 1 2) is extremely degraded. The explanation without overt movement is simple: the postverbal subject cannot move to AgrO, hence can't escape the ellipsis site. With the ICF, something extra must be stated. Let' s conclude, then, that the basketball isn't being assigned accusative Case. The standard account, of course (see Collins 1 997, So11an 2003) is that the postverbal subject has nominative case, assigned at a distance: incompatible with the ICF. Suppose then that locative inversion structures don't contain a nominative-assigning Tense, and that the postverbal subject is being assigned 'partitive' case (though definite NPs in this position sound quite acceptable, so the semantic determination of partitive Case would have to be abandoned) under sisterhood with the verb. Lasnik ( 1 992), based on Belletti ( 1 988) suggests that some instances of postverbal subjects (e.g., in expletive constructions) are assigned partitive Case. If this were to occur in (9), it would not constitute a violation of the ICF, as partitive case isn't structural, and all structural case assigners (trivially satisfied here, as there would be none, by stipulation) have assigned their Cases. But then why is ( 1 3) completely ungrammatical: ( 1 3)
* rolled the basketball into the room
The ICF cannot rule out ( 1 3) on the partitive account for the postverbal DP, cannot rule in (9) on the overtly-assigned nominative account, and cannot rule out ( 1 2) on the overtly-assigned accusative account. Suppose we amend the ICF account of ( 13): partitive Case can be assigned to the postverbal DP, and nominative Case must then be assigned to the PP in an overt spec-head relation. Baroque or not, this captures the facts considered so far. The reader is no doubt weary of the number of theoretical options we have considered and discarded in order to preserve the ICF account of word order. In the next section, however, we tum to evidence from quantifier scope that even this last emendation cannot be maintained. 2.
Moving without Structural Case & Structural Case without Moving
Kuno ( 1 97 1) noticed a scope contrast between locative inversion structures and postverbal locative structures, demonstrated here: ( 14)
Some actress stood on every stage (3 > \/, \/ > 3)
( I S)
On some stage stood every actress (3 > \/, *\/ > 3)
The contrast can be explained by a version of the generalization from Nevins & Anand (2003) in ( 1 6) ( 1 6)
Non-nominative subjects cannot take narrower scope than their surface position
Obviously, the assumption needed by the ICF for theory-internal reasons that the PP has nominative case in (IS) leads to the inapplicability of (16) in explaining the quantification contrasts with (14), In short, the ICF assumption that
287
Derivations without the Activity Condition
nominative case is always overtly assigned (even to PPs and morphologically dative subjects) not only complicates PF, it obscures contrasts at LF. The alternative is the EPP. XPs move without doing so in order to value structural case. The locative in ( 1 5 ) moves and does not receive structural Case. The postverbal subject in ( I S) does not move (at least, not enough to license pseudogapping or adjunct control) and yet does receive structural Case. We have what biolinguists call "a double dissociation". The EPP as a purely fonnal principle of structure-building enables us to model this. The contrasts in ( 1 4)-{ l 5) extend to many kinds of non-nominative subjects (ergatives in Hindi, datives in Greek). We return to Russian accusative subjects below. For the moment, though, consider another domain in which movement seems to occur completely independently of structural Case assignment. Though it is obligatory that at least one left-peripheral head (usually T(ense» bears the EPP property in English, it is often optional for lower functional heads to bear an EPP property. Concretely, we will assume that there is a head H distinct from (but quite cartographically close to ECM v)4 which may optionally bear EPP, inducing a movement of the ECM subject completely independent of Case assigment. Recall that the ICF requires overt movement for structural Case assignment. However, the order of the ECM subject and a particle of the matrix verb may vary, as Lasnik has shown with the ECM verb make oU15: ( 1 7)
1 made Cuijian out to be a fool
( 1 8)
I made out Cuijian to be a fool
The optionality of word order in ( 1 7) & ( 1 8) looks like good evidence for a head H with an optional EPP, whose specifier is linearly after the verb and before the particle. In fact, as ( 1 6) can more strongly be formulated as ( 1 9), there is a pre9iction for quantifier scope, attested in (20) and (2 1 ), from Lasnik (200 1 ) ( 1 9) (20)
4
Movement of a QP only for the EPP does not reconstruct for scope I made out every Mersenne number not to be prime (neg > � \7' >
neg)
ln the current model. perhaps rather empirically naively, it is assumed that
V-to-v raising
is overt
in English.
5
Make out, for the unfamiliar, has two relevant meanings in the ECM conslrUction: one is roughly convince. e.g. (i)
I made out my missing homework to have been eaten by the dog
The other is roughly discover. e.g. (ii)
I finally made out close tabs to have been kept on me all along by the CIA
A third meaning, used later in this paper. is roughly Sleai, as in (iii)
I made out like a bandit
The fourth, meaning roughly reciprocally kiss and embrace. is probably more common than any of these, but unhelpful in this paper (though the imaginative reader can substitute it in (22».
288
Andrew Nevins
(2 1 )
(*neg > V.
I made every Mersenne number out not to be prime
> neg)
tf
To be clarify the readings: (20) can mean either that I discov ered/let on that all Mersenne numbers aren't prime (i.e., they're all composite): this is the tf
> neg reading. It can also mean that I discovered/let on that it's not the case that h they ' re all prime: the 23 7 t one isn't. This is the neg > li reading, which (2 1 ) lacks. Recall, now, that the ICF does not allow structural Case assignment
without movement. Hence Boskovic maintains that object shift is always overt in all of (1 7),( 1 8),(20),(2 1 ). and hypothesizes that the particle has two different positions, one before and one after the position of object-shift for the ICF; l et 's call them both GutP for clarity ' s sake. How then, to capture the scope contrasts
(20)-(2 1 )? Boskovic suggests that the ambiguity is not due to reconstruction of the DP, but to neg-raising, " a sort of QR . . .bounded by GutP (p. 206)6. In other words, in (20), negation can neg-raise over the universal quantifier, leading to the non-surface reading. However in (21 ), negation cannot neg-raise past GutP, hence it cannot move to a position c-commanding (and outscoping) the universal quantifier. This captures all of the facts in (20)-(2 1); after all, by themselves they underdetennine a unique analysis. The claim that OutP bounds QR, however, is false. (22) can clearly describe a scenario in which a different guest absconds with each dessert. in
"
(22)
Some guest (or other) made out with every dessert (3
) V, V > 3)
Suppose then, that the account is revised, and that outP only blocks QR negation, but not universal quantifiers. Then it is predicted that while (22) may allow the inverse scope reading, (21 ) cannot. Again the facts underdetennine an analysis, and this aspect of the revised ICF model is consistent with (2 1 )-(22). It's wrong for the contrast in (23)-(24). ,
(23)
I made out some guest to have eaten every dessert (3
> V, V > 3)
(24)
I made some guest out to have eaten every dessert (3
> V, • V > 3)
Putative movement of outP should be irrelevant for QR of the universal quantifier if it is only n eg- raising that is affected. However, the contrast in (23) (24) receives an explanation if the different structures are due to movement of
the DP. The contrast is straightforwardly pred icted by ( 1 9) and the explicit axiom that structural Case can be checked without movement. It has been demonstrated that Case valuation can occur without
movement ( leading to vP internal nominative subjects and IP-internal ECM subjects) and that movement can occur without Case valuation (leading to frozen scope). The ICF predicts neither, and hence misses empirical generalizations about the interaction between Case, word order, and logical 6 Further doubt can be cast
on neg-raising due to ilS unavailability to provide negation with scope leave has only the must > not reading. No proponen t of n eg -rais i ng (when invoked as a counterexplanation to DP-reconstruction) has provided
over modals and quantificational adverbs: You must not a principled set of constrainlS on its application.
289
Derivations without the Activity Condition
form. The best answer we have at present for why a DP moves to the specifier of a functional head FO is the EPP. Non-nominative subjects in A-positions (as diagnosed by lack of weak crossover, ability to bind subject-oriented anaphors� and ability to control into adjuncts; cf. Ura 2000) are precisely that: non nominative.
3. Non-nom inative Subjects and the Activity Condition We have seen that while inherently-case marked (i.e. dative, ergative) and prepositional-phrase subjects can be argued to have a (morphologically invisible) nominative case. such a theory-internal stipulation complicates Case theory. On the empirical side, there is a more difficult problem: the lack of a distinction between agreeing and non-agreeing subjects eliminates the representational vocabulary to describe scope contrasts between nominative and non-nominative subjects; only the former display narrow scope with respect to a lower DP (Nevins & Anand 2003). Even if a characterization of subject properties were to include the stipulation that inherently-case marked elements have obligatorily wide scope, the adversity impersonal construction in Russian (Lavine & Freidin 2002, Bailyn 2003) presents a more serious challenge to proposals that inherent-case bears a hidden nominative, as the accusative subject is structurally case-marked: (25)
ranilo puljami Soldat Soldiers-ace wounded-past.nonagr bullets-jnstr
The accusative subject is in an A-position: it can bind the Spec-T oriented anaphor svoj and it does not induce weak-crossover violations (Bailyn 2003). Moreover, it is an instance of structural case: it undergoes the genitive of negation, which inherent accusative does not (Lavine & Freidin 2001). The verb is in default fonn and shows no agreement with , the subject. Finally, the accusative subject shows obligatorily wide scope with respect to the instrumental (Nevins & Anand 2003). This latter fact, that the accusative-initial order has semantic effects (namely, freezing scope at LF), effectively rules out a "Stylistic Fronting" analysis (e.g., Holmberg 2000) for these constructions, in which non-nominative subjects are displaced only at PF. All three of the above papers thus agree that the most fruitful analysis of the construcition has the . following properties: (26)
Subject Position is Spec-Ti The accusative subject is in Spec-T, an A-position, allowing anaphor binding and obviating weo.
(27)
EPP-Only Movement Exists: Movement of the accusative subject to Spec-T is triggered by the EPP, a requirement that a functional head have a DP specifier. Movement of the accusative does not involve instances of Agree or uninterpretable feature valuation.
The analysis of adversity impersonals in Russian, whereby movement of a non-nominative subject can occur ··only for the EPP", finds parallels in
290
Andrew Nevins
analyses of ergative and dative subjects crosslinguistically; however, it is the clearest instance of an already structurally case marked DP satisfying the EPP. Movement of this type, however, involves an operation between T(ense) and a phrase that has already valued all of its uninterpretable features. In the framework of Chomsky (200 1 et. preq), however, an element that has already elim inated its uninterpretable features (28) cannot enter into further operations
(29) (28)
(29)
Inactivity of an XP: An XP that eliminates its uninterpretable features (case, wh-) is rendered inactive. The Activity Condition: Inactive elements are not accessible for further operations.
The models in (26)-(27) and in (28)-(29) are directly incompatible. Hence I will argue that the latter (specifically (29), as (28) may remain terminologically useful) must be abandoned, and that its role in constraining derivations must find other sources. I argue that the Single Case Constraint, the locality of A-movement (in its formulations under Relativized Minimality or the Minimal Link Condition) and the Phase Impenetrability Condition (itself a locality condition) are sufficient to rule out derivations that (29) does, rendering (29) unnecessary.
4.
The Role of the Activity Condition in Phased Derivations
Once (27) becomes an active part of the model, the question arises as to what rules out movement of case-marked elements to satisfy the EPP. Specifically, consider a configuration in which a nominative DP in a fmite clause is c commanded by an athematic-subject verb, which in tum will be c-commanded by a T(ense) with
an EPP requirement and uninterpretable phi-features.
Consider (30)-(3 1), where unbracketed · left-to-right linear order denotes c command, lower occurrences of moved phrases are shown as copies, and some functional structure is omitted: ( 30)
TEPP. � is likely RhodaNom is intelligent
(3 1 )
RhodaNom.Nom T is likely RhodaNomNom is intelligent
Agreement of matrix T with the embedded subject is an illict derivational move. This derivational step could be ruled out by (29), as Agreement and Movement occurs for an inactive XP. However, consider another detail. As nominative case assignment will accompany cp-valuation, the DP in (3 1 ) is assigned nominative case twice. A universal generalization about derivations is that DPs cannot be valued with more than one case feature: (32)
The Single Case Constraint: A DP that is valued with more than one case feature is illegible to PF.
29 1
Derivations without the Activity Condition
The generalization in (32) has been upheld by Schutze (2001 ), who demonstrates that an apparent counterexample, the appearance of "case stacking" in Korean is not an instance of multiple case markers, but rather that the second morpheme is a focus marker and simply cannot be analyzed as case related. I propose that the implementation of (32) occurs in the morphological component7• Incidentally, some readers might wonder, if (32) is all that is active in constraining movements, why is *A book read John ungrammaticaL Satisfaction of the EPP by the object in fact presents no problem, but in order to move above the subject, a book must undergo topicalization. The ban on improper movement (from an A' to an A position), discussed below, will thus rule out this sort of example. As the subject position in (30) is athematic, the only derivational option is to merge an expletive, which will yield a convergent derivation. A variant of (30), however, that will not run afoul of (32), is a configuration with a non-fmite T(ense), (spelled out as lo), with no uninterpretable cp-features. Consider (33) (34), where unbracketed left-to-right linear order denotes c-command, lower occurrences of moved phrases are shown as copies, and some functional structure is omitted (italics intentional this time): (33)
tOEPP be likely RhodaNom is intelligent
(34)
RhodaNom to be likely RhodaNomNom is intelligent
The derivational step in (34) is illicit (as can be verified by the fact that a continuation of the derivation with an expletive, e.g. *It is certain Rhoda to be likely is intelligent is ungrammatical). If movement of the DP is only to satisfy EPP on nonfinite T, however, what rules out this step? Suppose we put aside de_rivations in which an expletive is available in (33). If an expletive is not available at the point of (33), then the fmite TP that eventually dominates (33) will not have access to an expletive either. Hence movement of the DP in (34) will always result in movement to specifier of a finite TP, yielding an eventual violation of (32). Note that this explanation, however, though correct due to the fact that finite TPs are at every root, provides no account for the illicit local step in (34). In ruling out the local derivational step in (34), we note that (29). clearly bans movement of inactive elements, yielding an apparent argument for the activity condition. Closer examination of the structure of (33), however, reveals that there is an independent factor that prevents this movement. An active hypothesis in models of the relationship between functional structure and case assignment is that nominative case valuation by T(ense) requires the "participation" of a c-commanding C(omplementizer}. Specifically, -
7 In a lexicalist morphology. (32) would be implemented as a ban over precompile d OPS with more than one fonnal case feature. In a realizational morphology. (32) can be implemented through the ordering of rewrite rules (Anderson 1992. Noyer )992) or through Vocabulary Insertion (Halle & Marantz 1993) that requires a one-to--onc pairing between the Vocabulary Item inserted and the feature realized under a single tenninal node.
292
Andrew Nevins
Iatridou ( 1 993) demonstrated that nominative case assignment in Greek always implicates the presence of CP structure. Rizzi ( 1982, Chapter 4) implicates a C Nominative relationship, Aygen (2002) continues this line of research for Turkish nominative case, and Chomsky's Beyond Explanatory Adequacy formalizes within the Agree framework that the C-T relationship enables nominative case. (35)
c!-Nominative Valuation: Nom. case is valued by T immediately
within a CPo
The empirical data in Greek & Turkish, in addition to English, if we consider the optional nonpronunciation of that in the fmite complement of is likely and seems, demand (35). Further confirmation comes from Icelandic (facts due to Halld6r Sigurdsson), in which a matrix verb may agree with an embedded subject (36), just as long as the embedded clause is not tensed, regardless of the presence of a complementizer (37) (36)
(37)
(38)
Mer virOast margar baekur vera Me-OAT seems-pi many books-nom-PL be-INF skemmtilegar interesting 'To me seem many books to be interesting' ... Mer viroast margar baekur eru/seu Me-DAT seems-pi many books-nom-PL are-pl-ind/subj skemmtilegar interesting ' *To me seem many books are interesting' Mer virOist aO margar baekur seu Me-DAT seems-sg that many-books-nom-PL are skemmtilegar interesting 'To me, seems that many books are interesting'
A closely related Germanic language thus reveals that fmite T in (37) implicates the presence of a strong phase boundary, preventing agreement into it. With (35) in mind, we return to the structure of (34) and the illicit derivational step. The light verb v above be likely, athematic in this case as it introduces no external argument, is included as well. (39)
tOEPP v be likely [cp C RhodaNom is intelligent ]
(40)
RhodaNom to v be likely [cp C RhodaNom is intelligent ]
The derivational step in (40) is illicit because it is A-movement out of a CP. Research from van Riemsdijk ( 1 978) to Chomsky (200 1 ) has established that no syntactic operations can relate a phrase outside a CP to a phrase c commanded by the head of that CP:
293
Derivations without the
(4 1)
Activity Condition
Phase Impenetrability of CP: N o phrases i n th e complement of CO may Move to or Agree with phrases higher than CP.
Since (40) involves Movement from the complement of an embedded CO to the sp ecifier of a T which dominates the embedded C, (4 1 ) rules out (40). In the implementation of Phase I mpenetrabil ity in Chomsky (200 1 ), in which the
complement of CP becomes inaccessible only upon the introduction a do minating v, nonfmite T will not be able to induce EPP movement of the embedded subj ect. No appeal to the Activity Condition is necessary. Recall that all of the non-nominative subjects whose existence m otivates the impossibility of (29) originate within the vp, A likely possibility is
that vP in these configurations is simply not a strong phase8. To summarize, (4 1 ), an independently motivated locality condition on
syntactic operations, rules out EPP satisfaction by DPs that are embedded too far from the attracting T. The case valuation of the attractee is irrelevant. Before concluding, we should however consider what Bob Franks has called the implicit Phase Penetrability Condition within (4 1 ): that movement to
the specifier of CP allows further movement out of Cp, This derivational option should be considered for ( 1 2), traditionally ruled out under the rubric of improper movement.
(42)
RhodaNom to v be likely intelligent ]
[cp
RhodaNom CEPP,Top RhodaNom is
The first question that arises with respect to (42) is what features the embedded CO bears. Clearly an EPP feature is present, to enable movement to
the specifier position. All intennediate specifier,CP landing sites of successive cyclic A ' movement bear an EPP feature. And the DP in question is not a wh pbrase, so the embedded CP cannot be interrogative. Suppose that it is
topicalization-driven. Then movement to specifier of the higber nonfinite T will still be banned, as that would constitute improper movement. The ban on improper movement should be considered in tandem with another effect of
A ' movement: it renders the moved argument invisible
as an
8 When considering the role of the Phase Im�netrability Condition, howeVcr, it is im�rtant
to
recal l that vPs (inc lud ing unaccusatives and passives, according to the successivc-cyclic movement and nuclear stress d iagnostics advanced in Legate 2003b) may constitute a strong phas e, at least in
English. If (41 ) is to be invoked in blocking derivational steps in which T attracts or agrees with a DP ins ide a strong phase, it looks like the same constraint wiIJ block T from Agreeing with the vP
internal DP in an unaccusative expletive construction:
(i)
There T seem to have v arrived three ships.
Although the \I in (i) does not introduce an external argument, it still constitutes a strong phase. Legate (2003a) proposes that the apparent agreement of T and the postverbal OP is actually
mediated by \I. The unaccusative v Agrees with its internal object, valuing its case as nominative.
Recall that the complement of v is inaccessible. However, T(ense), requiring t-valuation. can enter into an Agree relationship with
v, resulting in
plural agreement. The difference betwe en C- and
v
headed phases. then. with respect to apparent accessibility of the complement, is due to the fact that a higher Probe can �Agree with \I, but not with C.
294
Andrew Nevins
intervener for cp-Agree. Consider negative quantifier movement in Icelandic, which displaces an object DP above the base position of the subject, licenses parasitic gaps, and induces weak crossover v iolations (Svenonius 2000): (43)
Strakarnir, hofdu [engu grj6ti] t, [hent t2 ibilana] The.boys had no rock thrown at.the.cars
If the Activity Condition were correct, (43) should induce a defective intervention constraint, similar to superraising, as the moved object intervenes between T and the in-situ position of the subject. Why should a case-valued DP that intervenes between a Probe and Goal while occupying an A position incur a Minimality violation, while a case-valued DP intervening between Probe and Goal while occupying an A' position does not? It appears as though A'-moved elements become invisible for cp-Agree once they have targeted their final landing site (see e.g., Rezac 2003): (44)
A'-agreement renders a DP 's categorial and cp-features invisible
The solution pursued here does not assume that A' -operations render A-operations invisible, but simply that they occur necessarily later. This rests on changing a few assumptions about the way in which phrase markers are built. Bottom-up application of Merge requires that embedded clauses are fully constructed before matrix clauses. Suppose that trees are not built in this way, but rather, that embedded clauses and matrix clauses are built in parallel, in separate workspaces. The embedded vP and matrix vP are constructed at the same derivational step, and the embedded IP and matrix IP are constructed in the same derivational step. Finally, the embedded CP and matrix CP are joined at the same cycle of phrase-marker construction. In essence, this is a derivational rendering of the model in Williams (2003), and is schematized in (45) (45)
Workspace 1 John TpBSt V say cP
Workspace 2 Mary T ["pwho v saw <who>]
At this point, C is merged in each of the subtrees and the wh- phrase moves to Spec, C: (46)
Workspace J Workspace 2 [c did John T v say CP ] [cp who C Mary T [vpwho v saw <who>]]
At this point, the syntactic object is merged with the CP, and the wh- phrase may move on to matrix CP (47)
Workspace 1 [cp who did John T v say [cp <who> C Mary T ["pwho v saw <who>]]]
I will address technical questions that have no doubt arisen. What is the status of ? In this model, it serves essentially as a 'placemarker' for the embedded CP. One theoretical option is that it literally is a place marker� and
295
Derivations
without the Activity Condition
that fusion of two subtrees always involves variable-substitution: the sister of
say is literally replaced with the CP, reminiscent of a generalized transfonnation. No de-merge of cj> would be necessary; it would be replaced by the CP. Though the number of operations in the model has increased, it is not
immediately obvious that any empirical problems of overgeneration necessarily arise from substitution of this sort9, given constraints on where variables can be inserted and what types of phrases may substitute for them. However, I will pursue another theoretical option, needed for independent reasons. The representation in (47) does not commit to intennediate movement through the matrix vP. However, there is a some evidence that A' movement is successive-cyclic through every vP, which is theoretically enforced by making vP a strong phase. Strong phases, however, necessarily contain no uninterpretable features. The placemarker cj> is clearly uninterpretable at the interfaces. Hence the matrix vP in (47) is not a strong phase until substitution of the CP occurs. When substitution occurs, the vP is eligible for spellout, and requires movement of the wh- phrase to its edge.
(48)
Workspace 1 [cp T [vp who v say [cp <who> C Mary T [vp <who> v saw • . .
<who>]]]
Thus, three new aspects of phrase-marker construction have been introduced. First, trees are not built strictly bottom-up, but rather respect Parallel Cyclicity. Second, interarboreal fusion occurs as a substitution operation. Third, a phrase becomes a Strong Phase only once it is interpretable at the interfaces. That is, transfer of a strong phase to the interfaces can take place after, say, the vP has been merged with higher material. The reader may fmd these three new aspects of syntactic computation wholly unmotivated. Let us see how, in the spirit of Williams (2003) (though his model is, as its name implies, thoroughly representation�l). improper movement is ruled out by these very principles of tree construction. Improper movement is A-to-A'-to-A movement. However, as subtrees are built in parallel, the highest A position will necessarily be filled by the time the embedded A' movement occurs. Consider the point in the derivation in which a T has been merged in both subtrees:
(49)
Workspace 1 Workspace 2 [ T v seems � ] [ T [vp who [Mary [v saw <who>]]]]
EPP satisfaction of both T(ense)s must occur immediately. In Workspace 2, this is accomplished by movement of the agent, Mary. In Workspace 1 , the EPP property of T must be satisfied somehow. Workspaces 1 and 2 have not been fused yet, and will not be until a C is merged. Hence no 9 Of course, it may be possible to view ell as more than a placemarker: suppose that the visibility criterion for arguments is correct in every detail, but that the formal implementation is that every
argument is dominated by a Kase head that that must be valued. CP arguments will need Case. ell could be a Kase head that say values the Case of. in tum merging with the embedded CP.
296
Andrew Nevins
existing X P in the derivation can satisfy the EPP of matrix T, and an expletive must be merged. (50)
Workspace
I
Workspace 2
[IP It T v seems cp ] [IP Mary T [vp who [<Mary> [v saw <who>]]]]
The CP layer is built, and the embedded wh- moves to specifier of the embedded C. It's clearly too late for improper movement to occur, as the A position in the matrix clause is already filled. (5 1)
(52)
Workspace 1 Workspace 2 [cp does [ it T v s eem lj> ] [cp who C Mary T
..
.] .
The Generalized Ban on Improper Movement (Williams 2003): A
movement operation cannot move an element from Xn+ 1 in the embedded clause to Xn in the matrix clause (where X l..n denotes the selection sequence of functional projections, Le. from embedded C to matrix T, from embedded T to matrix v, etc.)
(52) is implemented here in a derivational model by a Relativized Extension Condition that forces extension within each derivational workspace, without the
possibility of delaying EPP satisfaction until after CP-Ievel merger. Identical remarks apply in banning the derivational step in (42) that motivated the present discussion. (42) and the class of movements from an embedded A-position to a matrix A-position are thus prevented by (4 1 ), which rules out movement from within (i.e. the non-edge of) a strong phase, while movement through the edge will occur after the matrix A-position has necessarily been filled, by (52). No appeal to the Activity Condition is needed to rule out movement from one Case position to another.
5. A note on Hyperraising Though the focus of the current paper is to dispense with the Activity Condition, as it wrongly rules out A-movement of a structurally case marked DP to satisfy EPP, it is worth briefly considering other configurations that pose a problem for the Activity Condition about which I have only a little to say at present. In the last section, I have shown that A_movement out of a tensed clause in English is ruled out by the Phase-Impenetrability Condition, with no appeal to the Activity Condition necessary. Hyperraising is a phenomenon in which there is apparent movement of a DP from a tensed embedded clause (often with a complementizer) to matrix Case position, often of an athematic subject verb. Ura (1 994) presents a comprehensive study of such constructions. l will report facts here from Brazilian Portuguese (BP; Ferreira 2000, Rodrigues 2003): (53)
0 Joao parece que esta doente The John seems-sing. that is-sing. sick-sing.
297
Derivations without the Activity Condition
Ninguem parece que esta doente Nobody seems-sing that is-sing. sick-sing.
(54)
A few remarks on (53) are in order. Skeptics that this is hyperraising would have to reckon with the fact that if 0 Joao were base-generated in the higher clause, it could not receive a theta-role. Hence it seems reasonable to assume that 0 Joiio entered the derivation in the embedded clause. Another objection to waylay is that 0 Joao might be topicalized (A'-moved) from the embedded clause, with a null expletive in the matrix Spec, TP. But Ferreira (2000) shows that negative quantifiers, which make lousy topics, can undergo hyperraising (54). What could be the difference between languages that allow hyperraising, such as BP, and languages that do not, such as English? If the account provided here for English is correct, there must be two properties of BP that differentiate it from English. The frrst is that the surface subject is not receiving Case in either the matrix or embedded position. Ura in fact suggests that the embedded position does not assign Case hyperraising configurations. He points out a striking one-way correlation: if a language has hyperraising, it will have pro. One reason that Brazilian Portuguese is particularly interesting is because it seems to have lost pro due to morphological change (Rodrigues 2003). Hence, I will not pursue the pro-related analysis, and instead adopt another possibility here, based on the phenomenon of copy-raising (also discussed by Ura). Consider the English example below: (55)
John seems like he is often in trouble
(56)
John seems like he just told everyone that he will resign
(57)
... John seems like it was told him that he must resign
In (55), John must receive a theta-role. The only place to get it is in the embedded clause. I will adopt the proposal of Fujii (2003), in which (55) involves movement of John from the embedded clause, with evidence from the impossibility of superraising (57). How does the DP come to bear two Cases, in apparent contravention of the single case constraint? It doesn't; I suggest that what is going on here is peeling (a term due to M. Starke): moving a DP out �f a larger KaseP, and re-merging the DP in the matrix clause undet anothei KaseP (see also Boeckx 2003 for a very similar approach to A'-resumption). The remnant KaseP in the embedded clause is spelled out as he, as it only bears nominative Case features, and agreement features (valued through concord with the DP): (58)
[K KasCNom [op <Joh n>]] seems like [K KasCNom lop <John>]] is often . . .
(59)
PF: John seems like he is often . . .
(60)
298
*
. . John to seem that (he) left .
Andrew Nevins
Contrast the current proposal, which rules out (60) due to the Strong-Phasehood of the CP, with the Activity condition, which rules it out due to freezing of DPs in a Case position. Both proposals must assume something like a Kase-peeling analysis for (56)-(60). Why is Kase-Peeling not applicable in (60)? In fact, I will suggest that it is, but that the movement still runs afoul of the Phase Impenetrability Condition. The Activity Condition. for its proponents, must be bolstered with a statement that Kase-peeling is not available with that, but it is available with like, with no connection between the environment where it is available and the
mechanism itself. In other words, the Activity Condition could be relaxed specifically for as-ifconstructions. In the current proposal, where there is no Activity Condition, Kase peeling is always available 10, but the movement is bounded by phases, and as-if does not head a strong Phase l l . Potsdam & Runner (200 1 ), who pursue a non
movement analysis in which an A-chain is "generated" between the matrix subject and the embedded pronoun, pursue a similar intuition: like and as are prepositions, not complementizers (see also Heycock 19 84)1 2 . Returning to hyperraising: the suggestion would be that hyperraising
always constitutes an instance of Kase-peeling 13. Languages that allow Kase peeling spell out the remnant as an empty category, recoverable through morphological agreement or other means. English allows Kase-peeling, but must 4 spell-out the remnant as a pronoun 1 .
1 0 The reader m ay wonder if Kase-peeling is aJlowed, say, from object position to subject position.
Of course, the perspective of Hornstein (200 1 ), this could be precisely what is happening: there is movement to a second theta position, and die spellout of a remnant accusative KaseP is himself.
This would be fine, but dlen, of cou rse, die DP would get two theta-roles, which is illicit. from
1 1 Pr epositional as and like heading an IP seem to select for a 'defective' IP, as the EPP of the
tensed verb does not even seem to need to be satisfied (Postal 2002):
(i)
Lasers can, as is[-EPP) obvious, cut through walls
(ii)
Lasers can, as was[-EPP) proved by Mike, cut through walls
My hope is that the non-phasehood of as-headed IPs and die lack of an EPP on the T head can be related i n further research.
12 Further evidence comes from the fact that like can co-occur with that (Shimada 2004):
(i)
Their idea was something like dlat die government was airport security to be tough
13 Ura ( 1 994) has one argument (p. 133) against hyperraising as an instance of copy-raising:
replacing the empty position in a hyperraising structure with a full pronoun blocks the idiomatic interpretation with idiom chunks. Note that dlis does not occur in English: The cal seems like it's out
of Ihe bag. My guess is that die emphatic nature of full pronouns in pro-drop languages is incompatible with an idiomatic interpretation. Stressing die pronoun in English copy-raising with idioms yields a fairly bizarre result: The cal seems like idioms:John seems like HE ;s the one to do the job.
14
IT is out of the bag, while not with non
It is tempting to relate die necessity of spelling out a remnant in English to the necessity of overt
possessors. Consider get-passives, such as John got his wallet stolen. My analysis of these
conslrUctions is that John receives a dleta-role as possessor, and moves to the higher position for Case, where it receives no theta-role. The obligatory "affectedness" and "possessor" requirements of get-passives (Oehrle and Nishio 1 98 1 ) are automatically derived. If John is in fact A-moving from
Spec, DP - a left branch extraction, it must leave a resumptive his. The corresponding g et-passives
299
Derivations without the Activity Condition
remnant
Brazilian Portuguese is not out of the woods, yet, however. Recall that, the difference between licit and illicit finite
spell-out notwithstanding,
clause-to-fmite-clause raising has been reduced to the Phase Impenetrab ility Condition. How do configurations like (53) skirt this derivational locality requirement? An uninteresting solution is to suggest that the C heading the embedded clauses in these cases does not head a Strong Phase. While this seems like an unmotivated stipu lation, hyperraising often out of subjunctive and irrealis/future embedded clauses, known to be more penneable that indicative tensed clauses (and hence, perhaps not Strong Phases). As the repertoire of functional heads includes a Vde/ (passives) and a Tdef (infmitives), the existence of a Cdef (subjunCtive/irrealis) might be expected. However, I will pursue another possibility here, based on an interpretation of the Phase Impenetrability Condition offered in Lightfoot & Rodrigues (2003), in their analysis of clitic
climbing in Romance. The formulation of the Phase Impenetrability Condition in (4 1 ) was sufficient for our discussion at that point, and indeed, could even be understood as a representational constraint on well-fonned A-chains, if one wanted such
things in one's theory. However, let us consider the derivational formulation of the PIC offered by Chomsky (200 1). When a strong phase is completed, the next phase down (i.e. the specifier, the phase head, and the complement of the phase bead) becomes inaccessible to further operations.
(60)
Suppose that there is a projection
Brazi lian
Portuguese
(Uriagereka
1 997).
FP in between CP and TP in Suppose,
furthermore.
that the
embedded subject raises to Spec, FP after raising to Spec, Tp15. On the view of the PIC in (60). when CP is completed. the next phase down will be spelled-out. The next phase down is the entire TP projection. H�nce, when an English embedded CP is completed. the entire TP projection is inaccessible, including
in Japanese. on the other hand. have no overt possessor, yet all the same require obligatory affectedness and possession between the matrix subject and the subject of the passive participle.
1 5 Some preliminary evidence for BP subject-raising to a position higher than Spec. TP comes from quantifier float above TP-oriented adverbs. While quantifier-float is known to be a dubious diagnostic for A-traces (see Bobaljik 2003), the following BP sentence
(C. Rodrigues. personal FP:
communication. January 2004) may indicate that the subject occupied embedded (i)
? As vitimas me
parcccm que todas inesperadamentc resolveram testemunhar
the victims CI-l Sg seem-pi
a favor dos
that ail-pi unexpectedly
decided- 3pl witness-inf
policiais.
in favor of.the-pl policemen "The victims seem to me to have unexpectedly all decided to witness in favor of the policemen." On the hypothesis that
inesperadamente is a TP-Ievel adverb. and that there is V-to-T raising in BP.
the presence of todas on the left of the adverb may diagnose movement through embedded Spec. FP. Further research
300
is clearly needed.
Andrew N ev ins
the embedded subject in Spec, TP, as we saw i n the disc ussion around (34). In Brazilian Portuguese, on the other hand, the embedded subject moves to Spec, FP. When the CP ph ase is completed, though the TP projection is inaccessi ble, FP is still accessible. Hence the embedded subject is av ai lable for further raising to the matrix spec, TP (possibly through the edge of spec, vP. if it is a strong phase when introducing parece).
Hyperraising, then, is only allowed in languages that manage to avoid the embedded subject being trapped by the PIC. On the account developed here, it has nothing to do with the (un)availability of Case-assignment in the embedded clause, nor with the Activity Condition. To conclude this section: I have shown that the existence of accusative subjects in Russian contravenes the Activity Condition. The Activity Condition does not seem to be operative with extraction from certain types of vPs (and, perhaps, certain types of CPs). The generalization that there is no A-movement from a Case-position is wrong, but most of its effects can be captured by independent principles (the Single Case Constraint, and the Phase Impenetrability Condition). 6.
Obligatory Pied-Piping for Internal Experiencers
This section concludes with a few more puzzles. There is one more derivation that, at first blush, looks like an instance in which case-valued DPs cannot undergo EPP movem ent Consider the experiencer argument of an athematic subject verb, and the illict derivational step of preposition-stranding to satisfy the EPP: .
(6 1)
tOEPP seem [to [EddieoBdl [cpthat California is in political trouble]
(62)
* EddieoBL to seem [to �QRLt] £Cpthat CA. is in political trouble]
The fonnulation of the EPP in (27) should allow for satisfaction by a non-nominative subject. Prepositions can be stranded under A-movement, as is known from pseudopassives: (63)
EddieEPp,Nom was lawyer.
vpass
spoken to tEddie on the train by a maniacal
There is no violation of The Single Case Constraint in (62), and no appeal to strong-phase locality can block the derivational step in (62). The Activity Condition can rule out (62), however, as the experiencer already bears oblique Case1 6. However, Brent DeChene (personal communication, October 2003) has pointed out that there are two structurally analogous configurations to (62) that 16 The Activity Condition would have trouble, of course, with Ihe corresponding Icelandic examples, in which the experiencer is dative and not contained within a preposition, since raising them is are perfectly grammatical. This reinforces that suggestion that preposition stranding is wbat
is problematic in English.
30 1
Derivations without the Activity Condition
should allow perfectly licit EPP movement of the experiencer, without contravention of the Activity Condition. Both involve passivized verbs. Compare (63) with (64), and (65) with (66): (64)
* EddieNom was said [to �] £Cpthat California is in trouble.]
(65 )
* This charityNom has been given [to [tcharity]] a book about adverbs1 7.
(66)
This charityNom has been donated [to [tcharity ]] by nearly everyone.
There is no constraint against A-movement with p-stranding of the experiencer/goal in a passivized verb when a non-argument follows. However, when an argument follows, A-movement with p-stranding is impossible. The relevant configuration seems to involve an "internal gap". Kuno' s (1973) formulation of the Clause Nonfinal Incomplete Constituent Constraint roughly captures this set of facts, though nonfmal must be appropriately delimited to refine the generalization, and its source remains puzzling. To demonstrate that (non)finality is relevant, consider a contrast in A' movement of the experience with p-stranding, which obtains between internal and external (i.e, extraposed) gaps:
(67)
(68)
* Who does Luisa seem [to [twho ]] to be a genius? Who does Luisa seem to be
a
genius to twho?
Though an explanation is lacking for the "internal-gap" restriction on experiencer preposition-stranding (possibilities involving remnant movement suggest themselves), it should be clear that the impossibility of (62) does not constitute an argument for the Activity Condition, as it is related to a much broader class of phenomena unrelated to case-valuation. One potential problem arises with the verb surprise, which disallows A-movement of the experiencer (69)-(7 1 ) but allows A' -movement (72): (69)
TOEPP surprise Mary that John won
(70)
It is certain lit to surprise Mary that John won
(7 1 ) (72)
* Mary to surprise tMIlIY that John won Who did it surprise twho that John won?
There is a possibility, however, that the experiencer is assigned case by a null preposition, similar to McGinnis' ( 1 998) analysis of strike in John struck Mary as a complete fool. The obligatory pied-piping of internal experiencers developed in this section would ban (7 1), a case of preposition stranding. A 17 There is no Case problem in this example; pied-piping is what's relevant. Consider To ,hat charity had been donated a book about metathesis, and 10 this charity has been given a book about adverbs.
302
Andrew Nevins
pied-piping derivation is disallowed because PP-experiencers cannot satisfy the EPP: (73)
* [To Eddie] seemed t that Zak would survive
Aside from its postu lat ion of null prepositions (also argued for by Hornstein 2003 for the object ofpromise in subject control uses), the restrictions on movement of the experiencer of surprise can be understood. A' movement in (72) arguably involves p ied-p iping . A fin al question arises concerning the difference between strike and surprise with respect to A-movement across the experiencer. Contrast (74)-(75) and (76):
(74) (75) (76)
John struck Mary [xp tJohn as a fool]
John seemed to Mary ri P tJohn to be a fool] * John surprised Mary that tJohn was so foolish
.
The Phase Impenetrability Condition will rule out (76) 18, as the subject has A-moved from out of a CPo Return ing to our main conclusion, there is a generalized ban on preposition -stranding of internal experiencers that requires investigation, but in no way supports the Activity Condition. 7.
Concluding Remarks
I will summarize the argumentation of the entire paper. DPs move because there
is an EPP, indep end ent of Case-valuation I9. The Inverse Case Filter does not subsume EPP effects, and can be abandoned with no loss of empirical coverage; on the contrary, EPP-only movement affords the theoretical vocabulary to describe many contrasts in quantifier scope. EPP-only movement in Russian shows that · a stru cturally-Case marked DP may move only to satisfy the EPP. The putative generalization that A-movement of structurally-Case-marked DPs is impossib le falls under the Activity Condition, suggesting that the latter should be abandoned. A number of configurations that the A ctivity Condition rules out can be ruled out instead by a morphological constraint (only one Case-valuation is possib le) , a.nd two syntax-internal constraints (Phase Impenetrability t taken as an axiom, and the Ban on Improper Movement, derived as a consequence of parallel cyc lici ty in subtree construction). Hyperraising is assimilated to Copy18
II is worth considering why the complement of surprise should contain more structure than strike.
Howard Lasnik (personal communication. November 2003) suggests that it the additional structure of the complement may come from factivity: (i)
It surprises me Ibat it's raining. #but it's actually nOl raining
(ii)
It strikes me Ibat it's mining, but it's actually not mining
19 Though I have assumed the EPP is axiomatic within phrase-marker consuuction, for an C and phase
explanation involving the need for an intervening XP in the linearization of the head
below, under an interpretation of the Linear Correspondence Axiom, see Richards (2002).
303
Derivations without
the Activity Co ndi tio n
Raising, and argued to be the result of Kase-peeling: moving a DP out of its Kase-phrase. A few remaining puzzles concerning the impossibility of moving DPs from within PPs are shown to be equally problematic for the Activity Condition and the model developed here, hence providing no argument for the former.
References Anderson, Stephen. ( 1 992). Amorphous Morphology. Cambridge University Press. Aygen, Gulsat. (2002). Finiteness. Case. and Clausal Architecture. Doctoral Dissertation. Harvard University. Bailyn, John Frederick. (2004). Generalized Inversion. Natural Language and Linguistic Theory 22, pp. I-SO.
Belletti, Adriana. (1 988). The Case of Unaccusatives. Linguistic Inquiry 17. pp. 375-4 1 6. Bobaljik, Jonathan. (2003). Floating Quantifiers: Handle with Care. In Lisa Cheng and Rint Sybesma. eds. The Second Gl01 International Sta1e-of-the-Article Book. Mouton de Gruyter. pp. 107-148. Boeckx, Cedric. (2003). Islands and Chains: Resumption as Stranding. John Bejamins. Boskovic, Zeljko. (2002). A-Movement and the EPP. Syntax 5.2. Chomsky, Noam. ( 1 98 1 ). Lectures o n Government and Binding. Foris: Dordrecht. Chomsky. Noam ( 1 993). A minimalist program for linguistic theory. In Kenneth Hale and Samuel Jay Keyser (eds.), The View from Building 20: Essays in Linguistics in Honor ofSylvain Bromberger MIT Press. Chomsky, Noam. (200 1 ). Derivation by Phase. In Michael Kenstowicz (ed.), Ken Hale: A Life in Language. MIT Press. Cottins, Chris. ( 1 997). Local Economy. MIT Press. Coopmans, Peter. (1 989). Where Stylistic and Syntactic Processes Meet: Locative Inversion in English. Language 65, 728-5 1 . Ferreira, Marcelo. (2000). Hyperraising and N u l l Subjects in Brazilian Portuguese. Ms., Universidade Estadual de Campinas. Fujii, Tomohiro. (2003). Licit and Illicit Long Subject-to-Subject Raising. In Yukio Otsu (ed.), The Proceedings of the Fourth Tokyo Conference on Psycholinguistics.
pp. l 09- 133. Hiluzi Syobo. Grohmann, Kleanthes, K., John Drury. and Juan Carlos Castillo. (2000). No More EPP. In the Proceedings of WCCFL 19. Somerville. MA: Cascadilla Press. Halle, Morris and Alec Marantz (1 993). Distributed Morphology and the Pieces of Inflection. In Kenneth Hale and Samuel Jay Keyser (eds.). The View from Building 20: Essays in Linguistics in Honor of Sylvain Bromberger,
Cambridge. Mass.: MIT Press. Heycock, Caroline. 1 994. Layers ofPredication. New York: Garland. Holmberg, Anders. 2000. Stylistic Fronting. Linguistic Inquiry 3 1 : 445-483.
Hornstein, Norbert. (2003). On Control. In Randall Hendricks (ed.) Min imalist Syntax. Blackwell. Iatridou, Sabine. (1 993). On Nominative Case Assignment and a Few related Things. In J. Bobaljik and C. Phillips (eds.), MITWPL 19. pp. 1 75-1 96. Kuno, Susumu. (1 973). Constraints on Internal Clauses and Sentential Subjects. Linguistic Inquiry 4.3, pp. 363-385. Lavine. James and Robert Freidin. (2002). The subj ect of defective T(ense) in Slavic. Journal ofSlavic Linguistics 10( 1 -2). http://www .facstaff.bucknell.eduljlavine/publicationspage.htm Lasnik, Howard. (1 992). Case and Expletives. Linguistic Inquiry 23 . pp. 3 8 1 -405.
304
Andrew Nevins
Lasnik,Howard. (200 I ). On a Scope Reconstruction Paradox. http://cognet.mit.edullibrary/bookslchomsky/celebration/essays/1asnik.html Legate, Julie. (2003a). Identifying Phases. Paper presented at the EPPlPhases Workshop at MIT. http://john.ling.yale.eduljlegate/phasews.pdf Legate, Julie. (2003b). Some Interface Properties of the Phase. Linguistic Inquiry 34(3). Lightfoot, David and CHene Rodrigues. (2003). Subject Inversion and Clitic Climbing. In I. Calgri, L. Meroni, and G. Tesan (eds.) University of Maryland Working Papers in Linguistics 1 2. Martin, Roger. ( 1999). Case, the Extended Projection Principle, and minimal ism. In S. Epstein and N. Hornstein (eds.), Working Minimalism . MIT Press. McGinnis, Martha. (1 998). Locality in A·Movement. Doctoral Dissertation, MIT. Distributed by MITWPL. Nevins, Andrew and Pmnav Anand (2003). Some AGREEment Matters. In G. Garding and M. Tsujimura (eds.), The Proceedings of WCCFL 22. Cascadilta. Noyer, Rolf. ( 1992). Features, Positions, and AffIXes in Autonomous Morphological Structure. Doctoral Dissertation, MIT. Distributed by MITWPL. Oehrle. Richard T. and Hiroko Nishio. ( 1 98 1 ). Adversity. In K. A. Fanner and C. Kitagawa (eds.), Proceedings of the Arizona Conference on Japanese Linguistics. Coyote Papers, University of Arizona, Volume 2. Postal, Paul. (2002). A Paradox in English Syntax. Ms., NYU. http://www.nyu.edulgsasldeptllingulpeoplelfaculty/postal/paperslskeptica1/ Potsdam, Eric and Jeffrey T. Runner. (2001 ). Richard returns: Copy Raising and Its Implications. Proceedings of Chicago Linguistics Society. Rezac, Milan. (2003). The fine structure of Cyclic Agree. Syntax 6.2. van Riemsdijk, Henle (1 978). A case study in syntactic markedness: The Binding Nature ofPrepositional Phrases. Lisse: The Peter de Ridder Press. Richards, Norvin (2002). Why there is an EPP. Paper prcsented at Movement and InteIpretation Workshop. Meikai University. Rizzi. Luigi. 1 982. Issues in Italian Syntax. Dordrecht: Foris. Rodrigues, Cilene (2003). Doctoral Dissertation. University of Maryland, College Park. SchUlZe, Carson. (2001). On Korean "Case Stacking": The varied functions of the particles IaJ and luI. Th e Linguistic Review 18(3): 1 93-232. Sauerland, Uli. (2003). Intermediate Adjunction with A·Movement Linguistic Inquiry 34, pp.308-3 14. Sigurdsson. HaUdor. ( 1 992). Verbal syntax and Case in Icelandic. lnstitue of Linguistics. University of Iceland. Soltan, Usama. (2003). Locative Inversion: A Phase-based Analysis. Ms, University of Maryland, College Park. Svenonius, Peter. (2000). Quantifier Movement in Icelandic. In Peter Svenonius (ed.), The Derivation of VO and OV, John Benjamins. Ura, Hiroyuki. ( 1994). Varieties o/Raising and the Feature·Based Bare Phrase StroclUre Theory. MITOPL 7. Distributed by MITWPL. Ura, Hiroyuki. (2000). Checking Theory and Grammatical Functions in Universal Grammar. Oxford Univeristy Press. Uriagereka. Juan. (1997). An F position in Western Romance. In K. Kiss (ed.) Discourse Configurational Languages. Oxford: Oxford University Press. Williams, Edwin. (2003). Representation Theory. MIT Press.
305
Derivations without the Activity Condition
Department of Linguistics
E39-245 MIT 77 Massachusetts Avenue Cambridge, MA 02 1 39 USA
[email protected]
306
and Philosophy
* Phase edge and extraction: a Tagalog case study
Andrea Rackowski and Norvin Richards MIT
In this paper. we examine evidence for the phase theory of movement (Chomsky 2000, 2001). in the context of Tagalog, arguing in particular that Tagalog has overt morphology that signals movement of arguments to check EPP on the head of the lIP phase. We show that this morphology interacts with extraction in ways which Chomsky's theory leads us to expect, and develop a theory of the Tagalog facts which also accounts for the effects of Huang's (1 982) CED.
1.
Introduction
According to the phase-based theory of syntax (Chomsky 2000, 200 I ). in order for some element to be extracted out of a phase, it must be located at the edge of that phase, either by merging into that position or moving there. Movement to the edge of the phase is accomplished by an EPP feature that forces some argument within the domain of the phase head to move to. check it. In this paper, we examine evidence for this theory in the context of Tagalog. In particular, we argue that Tagalog has a process of 'object' shift for specific arguments. We argue that. just as in Germanic languages. specific arguments are forced to move to the edge of the vP phase in order to receive the correct semantic interpretation. In Tagalog, this movement is signaled by morphology on the verb which agrees with the shifted argument (in Case, we will claim; see Rackowski 2002 for arguments). and also by the marker ang (or si for proper names) on the shifted argument. The shifted argument is underlined in the following examples� and the agreement morphology on the verb is in bold (Maclachlan 1992): (1)
Agent um -
a. .
-
B-um-ili ang bata ng tela sa palengke para sa nanay Nom.asp-buy ANG child CS c10th DTmarket P DTmother 'The child bought cloth at the market for mother. •
Theme -in
b.
8-in-ili-0 ng bata ang tela sa palengke parasa nanay asp-buy-Acc CS child ANGc10th DT market P DTmother 'The child bought the cloth at the market for mother. '
• The authors would like to thank Geitara Banzon for her help with the facts of Tagalog; thanks. too, to Alec Marantz, Martha McGinnis. David Pesetsky, and the audience at the Workshop on EPP and Phases for helpful comments. Responsibility for any remaining errors is entirely ours.
MIT Working Papers in Linguistics 49. 307-328 Perspectives on Phases 0 2005 Andrea Rackowski and Norvin Richards
Ph ase edge and extraction:
A Tagalog case study
Locative -an c.
B-in-ilh-an
ng bata ng tela ang palengke parasa nanay asp-buy-Dat CS child CS cloth ANG market P DTmother 'The child bought (the) cloth atthe market for mother. '
Benefactive ;-
I-b-in-ili ng bata ng tela sa palengkc ang nanay Obl-asp-buy CS child CS cloth DTmarket ANGmother 'The child bought (the) cloth at the market for mother:
d.
The ang-marked phrase has been known by a number of names in the literature on Tagalog, including 'subject', 'topic" and 'pivot' - here we will refer to it as the 'subject', though this term should be regarded with caution. I Since extraction can only occur from the edge of the phase, the morphology exemplified above correlates with the possibility of extraction; only 'subject' DPs may be extracted in Tagalog, as shown in (2); (2)
a.
Sino [ang binigy-an ng lalaki ng bulaklak _ ]? who ANG gave-Dat CS man CS flower 'Who didthe mangivethe flowerto?'
b.
* Sino lang i-binigay ng lalaki ang bulaklak who ANG Obi-gave C S man ANG flower
c.
*Sino [ang nagbigay ang lalaki ng bulaklak _ ]? who ANG Nom-gave ANG man CS flower
]?
This morphology also interacts in interesting ways with long-distance extraction; embedded clauses must become the 'subjects' of higher verbs in order to allow extraction out of them. We argue that this follows from a requirement that embedded clauses enter into Agree relations with higher heads in order to allow the extraction of arguments out of them. The resulting theory will capture the facts of Tagalog wh-extraction. and will also offer a new account of the CED and its effects in more familiar languages. 2.
'Subjects'
Consider again the examples in ( l ), repeated here as (3):
(3)
A gent -um-
a.
B-um-iIi ang bata ng tela sa palengke para sa nanay Nom.asp-buy ANG child CS cloth DTmarket P DTmother 'The child bought cloth atthe market formother:
I For some discussion of the syntactic properties
associated with 'subjects' in Tagalog, see
Schachter ( 1 976. 1 996), Guilfoyle, Hung, and Travis ( 1 992), Kroeger ( 1 993), Richards ( 1 993, 2000). Maclachlan ( 1 996), Aldridge (2002),
308
Rackowski (2002), and references cited there.
Andrea Rackowski and Norvin
Richards
Theme -in b. ng bata ang tela sa palengke parasa nanay B-in-ili-0 asp-buy-Acc CS chil d ANG cloth DT market P DTmother 'The child bought the cloth at the market for mother. ' Locative -an c. B -in -ilh-an n g bata ng tel a ang palengke para sa nanay asp-buy-Dat CS child CS cloth ANGmarket P DTmother 'The child bought (the) cloth atthe market for mother! Benefactive. id. J-b-in-ili ng bata ng tela sa palengke ang nanay Obl-asp-buy CS child CS cloth DTmarket ANG mother 'The child bought (the) cloth at the market for mother. ' One interesting property of 'subjects' that is evident from these examples is that all 'subjects' are obligatorily specific in Tagalog. That is, the sentence in (3a) cannot mean 'A child b ought cloth at the market for mother. ' In contrast, direct objects are obligatorily non-specific in sentences where the external argument is the 'subject' (3a) and are optionally sp eci fic in sentences where some other argument is the ' s ubject' (3c,d). The effects of , subjecthood' on specificity are particularly clear in the c ase of pronouns; being obligatorily specific, pronouns are ungrammatic al as ordinary objects in Tagalog. If an object pronoun is present, the form of the verb which makes the pronoun a 'subject' must be used, as illustrated in (4).
(4)
a.
b.
Sinampal-0 ako ng mandurukot asp.slap-Ace I.ANG CS pickpocket 'Nthe pickpocket slapped me. ' ... S-um-ampal ko ang mandurukot. Nom.asp-slap me ANGpickpocket 'The pickpocket slapped me.'
This pattern is strikingly reminiscent of object shift in Germanic languages, where sp eci fi c d ire ct obj ects are disallowed in VP-intemal positions. In Icelandic, for exampl e, spe cific objects shift out of VP and nonspecifics do not. Pronouns obligatorily shift. Shifting is to a position to the left of VP-adjoined adverbs and negation. (5)
a.
b.
Nemandinn las b6kina ekki students-the-NOM readbook-the-ACC not 'The students didn't read th e book. '
(Thrain sson 200 I)
Hann las ekki baekur he readnot books 'He didn't readbooks . '
309
Phase ed ge and extraction: A Tagalog case study
c.
?* Hann
he d.
e.
las baekur ekki read books not
( Diesing 1996)
Nemandinn las hana ekki students-the-NOM read it not 'The students didn't read it.' '"
Nemandinn las ekki hana students-the-NOM rcadnot it 'The students didn't read it. '
(Thrainsson 200 I )
Chomsky 200 1 argues that object shift occurs as the result of an EPP feature on v that is present only when it has an effect on semantic outcome. There is an effect on semantic outcome because the position at the edge of the vP is assigned a specific interpretation, while everything internal to vP is assigned a non specific interpretation. According to this theory, any specific argument must therefore raise to the edge of vP in order to receive the correct interpretation. The process is diagrammed in (6). (6)
vP ------
OBJspcic fic ---SUBJ
______ VP v -----V
We claim that Tagalog 'subjects'. like the object-shifted phrases in Icelandic, are DPs which have entered into an Agree relation with v, allowing them to raise into the edge of the phase and triggering Case-agreement morphology on the verb: (7)
( 8)
Lu-Iutu-io ng lalaki ang adobo. Asp-:cook-AccCS man ANG adobo 'The man will cook the adobo. ' vP
------
DPoo
-----DP EA _______ VP V[SPP]
------V
310
too
Andrea Rackowski and Norvin Richards
As in Icelandic, direct objects may fail to undergo this process, in which case they wi II receive a non-specific interpretation, and verbal agreement will register the external argument:
(9)
Magluluto ang lalaki ng adobo Nom-asp-cook ANG man CS adobo 'The man wi l l cook adobo'
3. Multiple Shift in Icelandic and Tagalog The mechanics of Icelandic object shift become more complicated when there is than one internal argument: If there are two objects, the higher one must shift first in order for the lower one to be allowed to shift. For instance. if a specific indirect object shifts, a specific direct object also may do so, but without shifting the higher argument the lower one cannot move, as shown in ( 1 I a). more
Eg Icina Mariu baekumar ekki lend Maria(DT) books(ACC) not • I do notlend Maria the books:
( 1 0)
I
(I I)
a.
?* Eg hina baekurnar I
b.
ekki Mariu. lendbooks(ACC) not Maria(DT)
Eg hina Mariu ekki baekumar. I lend Maria(DT) not books(ACC) 'I do not lend the books to Maria.' (Collins and Thrainsson 1 996)
The ordering of shifting between the indirect and direct objects has been observed by Richards 1 997 and Bruening 200 I, as being simply a case of the movement obeying Superiority/Shortest. Since the 10 begins in a position higher than the DO, it must be the first argument to Agree with the higher attracting head but, once it has done so, the DO is free to raise up and tuck-in below it (Richards J 997). This process is illustrated in ). ( 1 2). ( J 2)
____ 10 · _______ DO
_______ -------
----
tlO
_______
tDO
In this section we will argue that the locality conditions on this shift are just as strict in Tagalog as we have just seen them to be in Icelandic; if more than one argument shifts, the closest ope must move first, with the lower argument tuCking-in below it (as in ).
31 1
Phase edge and extraction: A Tagalog case study
( 1 2) above). 2 On the face of it, it seems counterintuitive to say that the process of making DPs into the 'subject' in Tagalog is subject to strict locality conditions. What we have seen so far suggests that almost any DP in the sentence can become the 'subject'. In a sentence containing both a benefactive and a direct object, for instance. the benefactive and the direct object are both possible 'subjects': ( 1 3)
a.
I-pinagluto ni Romeo ng adobo ang babae Obl-asp-cook CS Romeo CS adobo ANG woman 'Romeo cooked (the) adobo forthe woman'
b.
Niluto-0 ni Romeo ang adobo para sa babae asp-cook-Acc CS Romeo ANG adobo for DT woman 'Romeo cooked the adobo for a woman'
How can both ofthese DPs be equally accessible to Agree? Rackowski (2002) offers arguments (reviewed in the next section) that the examples in ( 1 3) actually have different argument structures; ( 1 3a) involves an instance of Pylkkanen's (200 1 , 2002) high applicative construction, while the benefactive in ( I 3b) is in a PP below the direct object.3 The structures are given in ( 1 4): ( 1 4)
a.
High applicative
vP
� v
ApplP
� DPbcn /'---.... Appl VP � V
b.
DPoo
vP �
v
Prepositional benefactive
VP
� DPDO � Vroot PP /'--... P
DPbcn
Considerations of locality lead us to predict that in struchlres like these, only the highest internal argument - the benefactive in ( 1 4a) and the direct object in ( 1 4b) - will be able to become the ·subject'. This prediction is correct: it is ungrammatical to move the DO past the benefactive in a clause l ike (14a), as 2 In what follows we are asswning that there is no tucking-in below a thematic specifier (see Chomsky 200 I, Rackowski 2002 for discussion).
3 See also Nakamura (1996) for a similar claim. 3 12
Andrea Rackowski and Norvin Richards
evidenced by the impossibility of a DO-subject clause with an in situ applicative benefactive, ( 1 5a):
( I S)
a.
* Niluto-0 ni Romeo ng babae ang adobo. asp-cook-Ace CS Romeo CS woman ANG adobo ' Romeo cooked the adobo for a woman.'
b.
I-pinagluto ni Romeo ng adobo ang babae. Obi-cook CS Romeo CS adobo ANG woman ' Romeo cooked (the) adobo forthe woman.'
The proposed account also allows us to capture a surprising fact about the behavior of the direct object when the benefactive argument is the 'subject'. We have seen that in ordinary transitive clauses, the direct object may either undergo object shift or not. If it does, it controls Case-agreement morphology on the verb. and receives a specific interpretation; if it does not. it receives a non-specific interpretation, and the external argument controls Case-agreement on the verb. The two options are illustrated again in (16):
( 1 6)
a.
Niluto-0 ni Romeo ang adobo asp-cook-Acc CS Romeo ANG adobo 'Romeo cooked the adobo'
Object Shift
b.
Nagluto si Romeo ng adobo Nom-asp-cook ANG Romeo CSadobo 'Romeo cooked adobo'
No
Object Shift
In example ( I 5b) above, however, the direct object may be either specific or non-specific, a kind of optionality we have not yet dealt with. This surprising fact is now recognizable as just another case of tucking in below an already-shifted object, as in Icelandic ditransitives. The DO is optionally specific in a benefactive-subject clause because once the features of v have been checked by .the raising" of the benefactive, the DO is free to raise to that position (if it is specific) and tuck-in below the benefactive, as shown in ( 1 7). Wh ile object shift of the DO will presumably have effects on the specificity of the DO, it apparently has no effect on the morphology on the verb, which continues to Agree with the benefactive argument. What the specificity facts in ( 1 Sb) show us, then, is that the verbal morphology is controlled by the first argument to undergo object shift (or, equivalently in this case, by the structurally highest argument):
31 3
Phase edge and extraction: A Tagal og case study
vP
( 1 7)
�
DPsen � DPDo� DPEA � V[ePPl
ApplP
� � appl
VP
�
V
The important thing to note about multiple argument shift is that in both Icelandic and Tagalog it is constrained by exactly the same superiority condition: The higher argument must shift first in order to license shifting of the lower one. This analysis correctly predicts the impossibility of applicativized benefactives in a DO-subject clause, since that construction would violate superiority. It also allows US to explain the optional specificity of the DO in a benefactive-subject clause, since object shift of the DO, being the second instance of object shift, has no morphological effect. 3.1 Variable binding and clause structure
We mentioned above that Rackowski (2002) offers the structures in ( 1 9) for the vPs (prior to Merge of the external argument) in the examples in ( 1 8) : ( 1 8)
( 1 9)
a.
I-pinagluto n i Romeo ng adobo ang babae Obl-asp-cook CS Romeo CS adobo ANGwoman 'Romeo cooked adobo for the woman'
b.
Niluto-0 ni Romeo ang adobo para sa babae asp-cook-Acc CS Romeo ANG adobo for DT woman 'Romeo cooked the adobo for a woman' .
a.
vP �
High applicative
v ApplP
� DPben � Appl VP � V
314
DPoo
Andrea Rackowski and Norvin
b.
vP /'-..... v YP
Richards
Prepositional benefactive
/'-.....
DPDO/'-..... VrootPP /'-..... P DPben These structures are crucial to our claim that promotion to 'subject' status is constrained by locality in Tagalog; only the highest internal argument may undergo the Tagalog equivalent of Object Shift. One of Rackowski's (2002) arguments for these structures is based on the facts of pronominal-variable binding in Tagalog. Pronouns may be bound in Tagalog by quantifiers which c-command them; thus, for instance, a quantificational external argument may bind a pronominal embedded in a non 'subject' while the reverse is not true (word order is irrelevant in all of the
DO,
following examples; Tagalog scrambling does not affect quantifier-variable binding): (20)
a.
b.
* Nagmamahal
ang kanyangiama ng bawat anakj. Nom.asp-pag-Iove ANG poss. father CS every child 'Heri father loves everychildj: (Richards 1 993)
Nagmamahal ang bawat ama; ng kanyangj anak. Nom.asp-pag-Iove ANG every father CS poss. child 'EveD' father; loves hiSt child.'
Crucially, these binding relations are not disrupted by promotion to ' subject' status; (20b) is well-fonned, even if the direct object becomes the 'subject': (2 1 )
Minamahal-0 n g bawat amaj
ang kanyan&j,anak
asp-love-Ace CS every father ANGposs. • Every fatherj loves hisi child'
child
The contIast between (20a) and (2 1 ) is the interesting one. In both ofthese examples, a non-'subject' quantifier attempts to bind a pronoun embedded in the 'subject', and binding fails in (20a) and succeeds in (21 ). Apparently binding relations of this kind can always succeed if the Merged position of the quantifier c-commands that of the pronoun; the grammaticality difference between (20a) and (2 1 ) therefore arises because the subject asymmetrically c-commands the object. With this in mind, we can consider pronominal variable binding relations between internal arguments. Rackowski (2002) discovered that the conditions on these relations reflect the structures in ( 1 9); a 'subject' benefactive argument, which can only arise from the structure in ( 1 9a). may not be bound by a direct object:
315
Phase edge and extraction:
(22)
A Tagalog case study
* I-binantay
ko ng bawat anakj ang kanyangjmagulang. Obl-asp-watch I CS . every child ANG poss. parent. ' I watched everychildi for hisiDarents. '
A prepositional benefactive argument, by contrast, ought to have the structure in ( I 9b), and is thus correctly predicted to be susceptible to binding by the direct object: (23)
B-um-antay ako ng bawat anakj Nom.asp-watch ANG.! CS every child [para sa kanyangi magulang]. P DTposs. parent '! watched everychildi for hisiparents. '
Note that the ill-formedness of (22) crucially cannot be attributed directly to the fact that the benefactive argument is a 'subject'; as (2 1) shows, 'subjects' are susceptible to binding by non-'subjects'. The structures in ( 1 9), on the other hand, offer a straightforward explanation for the asymmetry; the DO is Merged in a position c-commanding the benefactive argument in (23), but not in (22).
3.2 Summary
In the past sections we have argued that promotion to 'subject" in Tagalog is syntactically akin to Object Shift in the Germanic languages. We have seen that Tagalog 'subjects' have the specific reading associated with Object Shifted arguments in Germanic, and that promotion to 'subject' is constrained by considerations of locality which are identical to those in Germanic. The 'subject' in Tagalog triggers agreement for Case on the verb, and we have seen that this morphology is sensitive to the first argument to occupy the specifier of vP; multiple arguments may shift, we have claimed, but only the first (and therefore highest) of them will control agreement on the verb. 4. Tagalog, English, and the Syntax of Extraction Knowing what we now know about the syntax of case-agreement in Tagalog, we can tum to the facts of wh-extraction. We will see that a number of English and Tagalog conditions on A-bar movement can be made to follow from fairly basic assumptions.
4.1 Some assumptions We will crucially assume, following much work on locality, that a probe must Agree with the closest available goal, where a potential 'goal' is taken to be anything that is capable of moving (following Chomsky 2000, 200 I , we assume that all and only phases are in principle capable of moving), and that dominates the feature the probe is seeking. The definition of 'closest' we will use is given in (24):
316
Andrea Rackowski and Norvin Richards
(24)
A goal G is the closest one to a probe P ifthere is no distinct goal G' such that the mother ofG' dominates the mother of G.
As noted by Fitzpatrick (2002). defining 'closest> in terms of domination allows us to caprore both the effects of Shortest Attract and of Chomsky's {I 964) A over-A condition. This is demonstrated in the trees below; in both of these trees, G' is closer to P than G is, because X', the mother ofG', dominates X, the mother of G: (25)
8.
'Shortest Attract'
b. 'A-over-A condition'
/'--.. /'--..
/'--..
P /'-...
P
X'
X'
/'--..
/'--..
G' /'-...
G'
/'--..
X /'--..
X
/'--..
G
G
If we assume that phases are always in principle capable of moving. then we also derive the effects of Phase Impenetrability, as a special case of the A-over A condition; in a tree like the one in (26), for example, the probe C cannot Agree with the whP Goal, because vP is a closer potential Goal that can move (since it is a phase) and dominates a wh-feature (namely. the wh-feature that is also dominated by the wh-phmse):
(26)
CP
-------
C [+wh]
TP
_______
T'
�
v'
�
We do expect phrases in the highest specifier position of a phase to be accessible to Agree, if we accept Kayne's ( 1 994) claim that the specifier of 8 phrase is not dominated by that phrase (only by a segment of that phrase). In a tree like (27), for example. the whP is not dominated by vP; whP and vP both have the same mother (namely T') and are therefore equally close to the probe C:
317
Phase edge and extraction: A Tagalog case study
(27)
CP
-------
C TP [+wh] _______ T'
�p wh
v'
------v
We will also assume that once a probe P has Agreed with a goal G, P can ignore G for the rest of the derivation. It will be unimportant for our purposes why this is so (see Richards 1 998, Hiraiwa 2001 for theories about this), but some version of this assumption seems to be needed to deal with the facts in (28). once the Copy Theory of movement is assumed:
(28)
a.
? you gave :lj! to whom?
b.what
? didyou give� to=rm?
In the derivation of a multiple-wh question like the one in (28), the step in (28b) is somewhat problematic on standard assumptions; the copy of what is an XP bearing a wh-feature that is closer to the Probe than whom, yet it faHs to interfere with Agreement. We assume that this has to do with the fact that C has already entered into an Agree relation with what, allowing C to ignore this chain for the rest ofthe derivation. Finally, we will make some (fairly standard) assumptions about which heads are capable of entering into Agree relations. In particular, we will assume that v is responsible for checking Case on the direct object, and can also have EPP features which allow it to attract other Active phrases to its edge, making them accessible for Agree by higher probes. We will make similar assumptions about interrogative C, which also has a feature that it needs to check (namely [+wh]), and could in principle attract other Active goals as well; though this will play no role in the account. It will not be necessary in our account for declarative C to Agree with anything at all. The assumptions discussed above are summarized in (29): (29)
318
a.
A probe P must Agree with the closest Goal G that can move.
b.
A goal G can move if it is a phase.
c.
A goal G is the closest one to a probe P if there is no distinct goal G' such that the mother of G' dominates the mother of G.
d.
The specifier ofXP is not dominated by XP (Kayne 1 994).
Andrea Rackowski and Norvin Richards
e.
f.
g.
Once a probe P Agrees with a Ooal G, P can ignore 0 for the rest of the derivation (Richards 1 998, Hiraiwa 200 1). v
has a Case feature that is checked via Agree, and can also bear
EPP fearures which move Active phrases to its edge.
[+wh] C has a [+wh] feature which is checked via Agree (and sometimes Move).
We will see in the next two sections that these assumptions, most of which are fairly widely held in some form, yield a version of Huang's (I 982) CED (see Ceplova 200 1):
(30)
Only those CPs and DPs that Agree with a phase head on independent grounds (e.g., direct objects, complement clauses, and phrases in Spec CP) are transparent for wh-extraction.
We will also see that the Tagalog facts follow straightforwardly, given the picture of Tagalog syntax developed by Rackowski (2002); in fact, the Tagalog data will lend further support to our account of the English facts. 4.2 English: CED
Let us first consider the formation of a well-formed long-distance wh-question in English. (3 1 )
[cp Who do you [liP think [cp that we should ["p hire
_
]]l]?
The heads that participate in the derivation are the ones in (32); we will sketch the derivation as though movement begins once the tree has been completed, simply for ease of presentation:
who]]]]
(32)
In the first relevant step in the derivation, the v head of the embedded clause Agrees with who, and who moves to the specifier of vP, as is standardly assumed:
(33)
[ q+wh) [
V
[ C [
"'--,to])]]
.. _ vi...._
In the second step. v of the matrix clause Agrees with the complement CP, just as it would with a direct object; for the time being, we will simply assume. contra much of the literature, that CPs and DPs have similar requirements with respect to Case. We will present evidence from Tagalog in the next section that this Agree relation does in fact exist; we ask readers who are skeptical about what kind of licensing relation could exist between v and CP to bear with us for now:
319
Phase edge and extraction: A Tagalog case study
(34)
[C(+wh) [
I
[ C l (..:..:.w:r.:.t0::..:.-.- v..:. u
I
..:..:. wh9 =...1Jw llw J
__
As we saw in the last section, once a probe has Agreed with a particular goal. it can disregard that goal for future Agree relations. Matrix v can therefore now Agree with phrases dominated by
CP,
CPo
since it has already Agreed with It can also agree with
could, for example, Agree with the embedded
vP.
It
who.
since who is in the specifier of the embedded vP and therefore, as we also saw in the last section, not fully dominated by who and embedded vP are thus equally available to Agree with matrix v; v Agrees with who, which moves to Spec vP:
vP.
v
(3 5) Finally, matrix
v
[ C [who
I
!
C Agrees with who, causing it to move to its scope position:
[C[+wh] [ who v [ C [wile
(36)
wile)]]]
I
v
I
whe]]]]
The long-distance wh-question in (3 1 ) can thus be derived successfully 4 given our assumptions about how locality works. Let us now move on to consider an ill-fonned example:
(3 7)
*
[cpWbo do you we'll
[vp
[vpthink[cpthat (epif we ("p hire_II,
regret it]]]]?
The first two steps of this derivation are unproblematic, proceeding along lines that are familiar from the previous derivation. adjunct clause will Agree with
(38)
[cp q+wh) [vp
v
First the v of the embedded
who. causing it to move to Spec
[ep C lep if Il..:J who n. [vp .
vP:
v]]]]
The next relevant step is that matrix v Agrees with its complement before:
CP, as
4 The successive-cyclic movement path of the wh-pbrase in our derivation is somewhat unorthodox; it is stopping in Spec vP but not in Spec CPo This is straightforwardly consistent with some of the kinds of evidence that are standardly used to motivate postulating successive-cyclic movement, but others will need to be considered carefully. The effects of successive-cyclic wh-movement on Irish complcmcntizers, for example, would need to be reinterpreted; Irish complementizers would have their forms determined not by wbetber they had hosted a wb-phrase in their specifiers. but by
whether they had entered into Agree relations with v beads which had hosted wh-phrases. Panial
wh-movement, in tbis theory, would have to analyzed in tenns of the Indirect Dependency Approach
1 994, 2000) rather than the Direct Dependency Approach (van Riemsdijk 1983, McDaniel 1 989); thanks to Robert Frank for raising this problem, and to Ben Bruening for solving it.
(Dayal
320
Andrea Rackowski and
N orvin
Richards
Finally, the matrix v must attract a wh-phrase, in order for this wh-phrase to move to its edge and so escape Spell-out. As before, v is entitled to ignore the complement CP, since it has already Agreed with it. However, even given this, the closest moveable phrase that dominates a wh-feature is now the adjunct CP;
general principles of locality therefore prevent attraction of who , which is dominated by this CP and therefore further away from the Probe. We have no
theory to offer of why English rejects phrases like if we hire who as possible wh phrases, but whatever conditions on pied-piping rule this out will also doom the current derivationS . The assumptions we made in the previous section, then, derive the effects of Huang's (1982) CEO. They do so on the basis of another assumption,
however, which some might find questionable; that CPs and OPs behave alike with respect to attraction by v, in that complement CPs, but not adjunct or subject CPs, enter into checking relations with v to check Case or something like it. Of course, if v is going to agree with any CPs at all, these are the ones we would expect it to agree with. Still, we might prefer to find some kind o f independent evidence for the Agree relations that we are positing.
4.3 Tagalog In previous sections we argued that Tagalog is a language in which the Agree relations that v enters into have a morphological consequence. Recall from section Error! Reference source not found. above that Tagalog has verbal morphology which, we argued, signals the Case of the OP which has undergone (possibly covert) movement to the edge of the vP phase. In (40), the Case agreement morphology is in boldface, and the OP this morphology agrees with is underlined:
(40)
a.
Nagbigay ang magsasaka ng bulaklak sa kalabaw Nom-gave ANG fanner CS flower water-buffalo 'The fannergave a flowerto the water buffalo'
h.
I-binigay ng magsasaka ang bulaklak sa
DT
kalabaw
Obi-gave CS farmer ANG flower OT water-buffalo • Althe fannergave the flower to the water buffalo '
5
There do in fact appear to be languages which exploit clausal pied-piping to circumvent CEO violations: Basque (Ortiz de Urbina 1 990, 198)
(i)
[Mikeli zer esan ondorenljoan zen etxetik? Mikel-Oat what say after go AUX home-from '[After saying what to Mikel) did he leave home?'
Thanks to Jeong-Me Yoon for pointing this out to us.
32 1
Phase edge and extraction: A Tagalog case study
c.
Binigy-an ng magsasaka ng bulaklak ang kalabaw gave-Dat CS fanner CS flower ANGwater-buffalo Althe fannergave althe flowerto the water buffalo' •
The agreement morphology on the verb may agree either with a DP argument, as above, or with a CP argument. In (4 1 ). for example, the verb sabi 'say' bears either Nominative morphology that agrees with its DP 'subject' or Accusative agreement that agrees with its CP complement: (4 1 )
a.
Nagsabi ang kalabaw na masarap ang bulaklak Nom-said ANGwater-buffalo that delicious ANG flower 'The water buffalo said that the flower was delicious'
b.
Sinabi-0 ng kalabaw na masarap ang bulaklak said-Ace CS water-buffalo that delicious ANG flower 'Althe waterbuffalo said that the flower was delicious'
Wh-extraction in Tagalog imposes constraints on the kind of agreement morphology the verbs in the clause may bear, of a kind that the theory outlined in the previous section leads us to expect. Long-distance extraction in Tagalog always requires intervening verbs to Agree with the CP from which extraction takes place. This is shown for long-distance extraction of the adjunct kailan in (42)-(44): (42)
(43)
a.
Kailan [ sinabi-0 ng sundalo when said-AccCS soldier r na uuwi ang Pangulo e ]] ? that Nom-will-go-home ANG President 'Whendidthe soldiersay that the President would go home?'
b.
... Kailan [ nagsabi ang sundalo when Nom-saidANGsoldier [ na uuwi ang Pangulo e ]]? that Nom-will-go-home ANG President 'Whendidthe soldiersay that the President would go home?'
a.
b.
322
Kailan [i-pinangako ng sundalo when Obi-promised CS soldier ang Pangulo e ]}? [na uuwi that Nom-wiII-go-home ANG President 'Whendidthe soldierpromise that the President would go home?'
* Kailan [ oangako ang sundalo when Nom-promised ANG soldier [na uuwi ang Pangulo e ]1? that Nom-will-go-home ANGPresident 'Whendidthe soldierpromise that the President would go horneT
\ ,
Andrea Rackowski and Norvin Richards
( 44)
a.
b.
Kailan [ pinaniwala-an ng sundalo when believed-Oat CS soldier [ n a uuwi aDg Pangulo e ] ] ? that Nom-will-go-home ANG President 'Whendidthe soldierbelieve that the President would go horn eT ang sundal o When Nom-believedANG soldier r na uuwi aDg Pangulo e ]]? that Nom-wiU-go-home ANG President 'Whendidthe soldierbelieve that the President would go horneT
* KailaD [ naniwala
The same constraint is shown to hold in cases of long-distance relativization in (45); here we have given only the well-fonned examples, but choice of any other kind of agreement on the verbs would make the sentences ill-formed:
(45)
a.
b.
c.
ang kalabaw [ na sinabi-0 ng goro ANGwater-buffalo that said-Acc CS teacher ng lalaki ng bulakl ak e ] [ na bibigy-an that will-give-Dat CS man CS flower 'the water buffalo that the teacher said that the man would givea flowerto' ang kalabaw [ na i-pinangako ng gura ANGwater-buffalo that Obi-promised CS teacher ng lalaki ng bulaklak eJ] [ na bibigy-an that will-give-Dat CS man CS flower 'the water buffalo that the teacher promised that the man would give a flower to' ang kalabaw [ na pinaniwala-an ng gura ANGwater-buffalo that believed-Dat CS teacher
[na bibigy-an
ng lalaki ng bulaklak ell
that will-give -Dat CS man CS flower
'the water buffalo that the teacher believed that the man would give a flower to' Note .that the higher verb is demonstrably not agreeing with the extracted wh phrase. This is perhaps clearest in (45), where the extracted operator bears Dative case, but the higher verb takes whatever form is appropriate for agreement with the complement clause (Accusative for sab; say' , Oblique for pangako 'promise', and Dative for paniwala 'believe'). This is not because the first two verbs lack a Dative-agreeing form: •
(46)
Pinangaku-an niyaako promised·Dat CS-he ANG-me bibigy-an niyang. bulaklak ang kalabaw] [na that will-give-Dat CS·he CS flowerANGwater-buffalo "He promi sed � that he will give a flower to the water buffalo . '
32 3
Phase edge and extraction: A Tagalog case study
Extraction out of a complement clause therefore seems to require v to Agree with the complement clause. This, of course, is what the theory developed in the previous section predicted; in order for v to Agree with a wh-phrase in the complement clause, v must first Agree with the complement clause itself. thereby making it transparent and making the embedded wh-phrase accessible to Agree. Next let us consider local extraction. Here there are two cases to consider: OP-extraction, shown in (47), and non-OP-extraction, shown in (48) (49). There are two differences between the OP-extraction case and the non DP-extraction case. One is that OP-extraction involves a cleft construction, while non-OP-extraction does not; we will not discuss this difference in this paper (though see Richards (2004) for some discussion). The other difference is that OP-extraction imposes restrictions on the verb of the clause; it must agree with the extracted OP, as (47) shows:
(47)
B.
Sino lang binigy-an ng lalaki ng bulaklak _ ]? who ANG gave-Dat CS man CS flower 'Who did the man give the flower toT
b.
* Sino [ang i-binigay ng lalakiang bulaklak who ANG Obi-gave CS man ANG flower 'Who did the man give the flower to?'
c.
*Sino lang nagbigay ang lalaki ng bulaklak _ ]? who ANG Nom-gave ANGman CS flower 'Who did the man give the flower to?'
]?
No such restrictions are imposed by extraction ofnon-DPs, as (48)-(49) show; any form of the verb may be used in these cases. This is what we expect; extraction has to take place via the edge of vP, and this verbal agreement is agreement for Case, so Case-bearing extracted phrases ,\\,:il1 necessarily agree with the verb, while non-Case-bearing extracted phrases will not:
(48)
(49)
324
a.
KaHan binigy-an ng lalaki Dg bulaklak aDg kalabaw? when gave-Dat CS man CS flower ANG water-buffalo 'When did the man give a flower to the water buffalo?'
b.
Kailan i-binigay ng lalaki ang bulaklak sa kalabaw? when ObI-gave CS man ANG flower DTwater-buffalo 'When did the man give the flower to the water buffalo?'
c.
Kailan nagbigay ang lalaki ng bulaklak sa kalabaw? when Nom-gave ANG man CS flower DTwater-buffalo 'When did the man give a flower to the water buffalo?'
a.
Sa aling kalabaw i-binigay ng lalakiang bulaklak? to which water-buffalo Obi-gave CS man ANGflower 'To which water buffalo did the man give the flower?'
Andrea Rackowski and Norvin Richards
b.
Sa aling kalabaw nagbigay ang lalaking bulaklak
A related fact that also follows from the theory is exemplified in (50). We saw in section 2 above that use of Nominative agreement on a transitive verb typically requires the direct object to be non-specific (50a). In the theory developed above and in Rackowski (2002), this is because Nominative agreement indicates that the object has not undergone object shift to the edge of vP. As (SOb) indicates, however, Nominative agreement does not have this effect on the object when the 'subject' is extracted (Adams and Manaster-Ramer 1 988, Voskuil I 993): (50)
a.
K-um-ain ang kalabaw ng bulaklak Nom-ate ANG water-buffalo CS flower 'The water buffalo ate a/*the flower'
b.
Sino ang k-um-ain ng bulaklak? who ANG Nom-ate CS flower 'Who ate a/the flower?'
Again, this is what we expect. If the 'subject' is to be extracted, it must be on the edge of vP. and hence object shift of the object is impossible. As we know from other domains, when movement of this kind is ruled out by the syntax, the semantic consequences that it would have had are suspended (Adger 1 994, Chomsky 2001 ). The Danish facts in (5 1 ) are another instance of this. Object shift of pronouns is normally obligatory in Danish, as (S I a-b) show; being by nature specific, they must move into the domain where they can receive a specific interpretation. However, when Holmberg's Generalization makes object shift impossible, as in (5 I c), it can simply fail to occur, with no semantic consequences for the unshifted pronoun: (5 1 )
a.
Peter kebte den ikke Peter bought it not
b.
* Peter kabte ikke den Peter bought not it
c.
Peter harikke kebe den Peter hasnot bought it
Something similar seems to be happening in the Tagalog example in (50b); the direct object can receive a specific interpretation even without object shift, since object shift is ruled out by the syntactic conditions imposed by whMextraction.
325
Phase edge and
extraction: A Tagal og case study
5. Conclusion In this paper we have shown that Tagalog provides unique evi,dence for the phase-based theory of movement. In partic u l ar. Tagalog voice morphology indicates that some element has moved to the edge of the phase, and this in tum offers support for the hypothesis that clauses from which some element is extracted must themselves move to the edg e of a ph ase in order to allow extraction. In addition, the theory presented here, in conjuncti on with some fairly basic assumptions, accounts straightforwardly for the Tagalog morphological and speci fic i ty requirements on extraction, and also offers a new way of deriving Huang's ( 1 982) Using general principles of locality, along with previously established generalizations about the nature of syntactic relations between a single Probe and multiple Goals, we are led to a new expectation about the nature of successive-cyclic wh-movement: Probes along the path of movement ought to Agree, not just with the moving wh-phrase, but with the clauses out of which extraction takes place. We have seen two types of evidence for this conclusion. First, only those clauses with which v might be expected to be able to Agree are transparent for extraction; this is Huang's ( 1 982) CED. Second, in languages like Tagalog, we have argued that Agree with v has overt morphological consequences, and we have seen that in cases of long-distance wh-extraction, the morphology signalling this Agree relation must be controlled by the clause from which wh-extraction takes place.
CEO.
Bibliography Adams. K., and A. Manaster-Ramer ( 1 988). Some questions of topic/focus choice in Tagalog. Oceanic Linguistics 27.79- 1 02. Adger, D. (I 994}. Functional heads and inlerprelalion. Doctoral dissertation, University of Edinburgh.
- Aldridge. E. (2002). Wh-movement in Seediq and Tagalog" In A. Rackowski and N. Richards (eds.), MITWPL 44: Proceedings ofAFLA 8. MIT Working Papers in Linguistics, Cambridge, MA. 1 -1 6. Bruening, B. (2001). QR Obeys Superiority: Frozen Scope and ACD. Linguistic Inquiry 32: 233-273. Ceplova, M. (200 1 ). Minimalist islands-restricting P-features. Ms., MIT. Chomsky, N. ( 1 964). Current issues in linguistic theory. The Hague: Mouton. Chomsky, N. (2000). Minimalist inquiries. In R. Martin, D. Michaels, & J. Uriagereka (eds.), Step by step: essays on Minimalist syntax in honor ofHoward Lasnik. MIT Press, Cambridge, MA. 89- 156. Chomsky, N. (2001 ). Derivation by phase. In M. Kenstowicz (ed.), Ken Hale: a life in language. MIT Press, Cambridge, MA. I -52. Collins. C .• and H. Thrainsson. ( 1 996). VP-Intemal Structure and Object Shift in Icelandic. Linguistic Inquiry 27: 391 -444.
Dayal, V. ( 1 994). Scope marking as indirect wh-dependency. Natural Language Semantics 2. 1 37- 1 70. Dayal, V. (2000). Scope marking: cross-linguistic variation in indirect dependency. In U. Lutz et al (eds.) Wh-scope marking, pp. 1 57- 1 93. John Benjamins, Amsterdam. Diesing, M. ( 1 992) . Indefinites. Cambridge, MA, MIT Press.
326
Andrea Rackowski
and Norvin Richards
Diesing, M. ( J 996). Semantic Variables and Object Shift. In H. Thrainsson. S. D. Epstein, S. Peter (eds.), Comparative Germanic Syntax. ll. Dordrecht, Kluwer: 66-84.
Fitzpatrick, J. (2002). On Minimalist approaches to the locality of movement. Linguistic Inquiry 33.443-464.
Guilfoyle, E., H. Hung. and L. Travis. ( J 992). Spec of IP and Spec of VP: two subjects in Austronesian languages. Natural Language and Linguistic Theory 1 0.3754 14.
Hiraiwa, K. (200 1 ). Multiple Agree and the Defective Intervention Constraint in Japanese. MITWPL 40 : the proceedings of HUMIT 2000. MIT Working Papers in Linguistics, Cambridge, MA. 67-80. Huang, C.-T. J. ( 1 982). Logical relations in Chinese and the theory of grammar. Doctoral dissertation, MIT. Kayne, R. (1 994). The antisymmetry ofsyntax. MIT Pres s, Cambridge, MA. Kroeger, P. ( 1 993). Phrase Structure and Grammatical Relations in Tagalog. Stanford: CSLI Publications. Maclachlan, A. E. (1 992). Morphosyntax of Tagalog Verbs: The Inflectional System and Its Interaction with Derivational Morphology. McGill Working Papers in Linguistics 6: 65-84.
Maclachlan, A. E. ( 1 996). Aspects of ergativity in Tagalog. Ph.D. dissertation, McGill University, Montreal. McDaniel, D. ( 1 989). Partial and multiple wh-movement. Natural Language and Linguistic Theory 7. 565-604.
Nakamura, M. ( 1 996). Economy 0/ chain formation. Ph.D. dissertation, McGill University, Montreal. Ortiz de Urbina, J. ( 1 990). Operator feature percolation and clausal pied-piping. In L. Cheng and H. Demirdash (cds.). MIT Working Papers in Linguistics 1 3 , 1 93208. Cambridge. MA: MITWPL. Py lkkanen, L. (200 1 ). What applicativc heads apply to. In M. Fox, A. Williams & E. Kaiser (eds.), Proceedings of Ihe 24th Annual Penn Linguistics Colloquium. Penn Working Papers in Linguistics. 7 . 1 . Department of Linguistics . University of Pennsylvania. Pylkklinen, L. (2002). Introducing Arguments. Ph.D. dissertation, MIT, Cambridge, MA. Rackowski. A. (2002). The Structure o/ Tagalog: Specificity. Voice. andthe Distribution ofArguments. Ph.D. dissertation. MIT. Cambridge, MA. Richards, N. ( 1 993). Tagalog and the typology of scrambling. Honor.; thesis. Cornell University. Richards, N. ( 1 997). Whal moves where when in which language? Ph.D. dissertation. MIT, C ambridge, MA. Richards. N. ( 1 998). The Principle of Minimal Compliance. LinguistiC Inquiry 29.599629.
Richards, N. (2000). Another look at Tagalog subjects. In I. Paul. V. Phillips, and L. Travis (cds.). Formal Issues in Austronesian Linguistics. Dordrecht: Kluwer. Richards, N. (2004). Zazaki wh -movement. In M. Kenstowicz (ed .), MIT Working Papers on Endangerd and Less-Familiar Languages. volume 6: Studies in Zazaki Grammar. Cambridge, MA: MIT Working Paper.; in Linguistics.
van Riemsdijk. H. ( 1 983). Correspondence effects and the empty category principle. In Y. Otsu et al. (eds.) Studies in Generative Grammar and Language Acquisition, pp. 5- J 6. Tokyo, JP. Schachter, P. ( 1 976). The subject in Philippine languages: topic. actor. actor-topic, or none of the above. In C. Li (ed.) Subject and Topic. New York: Academic Press.
Schachter, P. ( 1 996). The subject in Tagalog: still none o/the above. UCLA Occasional Papers in Linguistics 1 5. Los Angeles: UCLA Department of Linguistics.
327
Phase edge and extraction: A Tagalog case study
Thrainsson, H. (2000). Object Shift and Scrambling. In M. Baltin and C. Collins (eds.). A Handbook of Contemporary Syntactic Theory. Blackwell. Voskuil. J. ( 1 993). Verbal affixation in Tagalog (and Malay). McGill Working Papers in Linguistics 8.75-106. Department of Linguistics. McGill University. Montreal.
Department of Linguistics and Philosophy 32-0808 MIT Cambridge. MA 02 1 39
[email protected], [email protected]
328
Asymmetries in Prosodic Domain Formation* Michael Wagner Massachussetts Institute o/Technology
This paper presenrs an asymmetry in the prosody of Dutch. English, and German: if a functor precedes its complement. it is phrased sep
arately; jf it follows its complement or part of it, the two are phrased
together and the functor is prosodically subordinated and 'affixed'to
the preceding prosodic constituent The prosodic asymmetry is first shown for the relation between predicates and their complements. and is then generalized to modification. A recursive algorithm is pre sented that derives the correct prosodic structure and nuclear stress in the three languages. exploiting only the syntactic asymmetry of projection and the left-right asymmetry of linear order.
1.
Modifiers vs. Arguments
The prosody of a sentence is influenced by semantic, syntactic, and phono logical factors, a fact that a fact that poses a challenge for modular theories of grammar. Data concerning relative prominence (stress), the distribution of accents, and facts about prosodic phrasing have shaped the development of generative theories ever since the outline of the transformational cycle in the grammar of English in Chomsky et al. ( 1 957) and Chomsky and Halle (1 968) (SPE). The key issue is the question of what type of syntactic or semantic information is necessary to assign the correct prosody. One crucial factor that has been proposed to play a role is the distinc tion between modifiers and arguments (Gussenhoven. 1984; Kritka. 1984; Selkirk, 1 984; Jacobs , 1992; Truckenbi-odt, 1 993; Zubizarret a. 1998). This paper presents prosodic asymmetries that a show a close parallel between -This paper is a development of some ideas presented at the workshop on phases. The current version benefited from a stay at the Department of Linguistics at Stanford University in 2003. I ewe much to discussions with David Adger. Karlos Arregi. Asaf Bachrach. Corrien Blom, Lev Blumenfeld,
Edward Flemming,
Morris Halle, Brent Henderson. Irene Heim. Florian Jaeger.
Lahiri. Winnie Lechner. Alec Marantz, Shigeru Miya Neeleman. Jon Nissenbaum, David Pesetsky. Chris Potts. Norvin Richards. Andres Salanova. Lisa Selkirk, Michal Starke, Donca Steriade. Huben Truckenbrodt. Roben Truswell. Susi Wunnbnmd. Also. I wish to thank the audiences at the phase work-shop at MIT. at Trend Paul Kipamcy. Angelika Kratzer. Aditi
gawa.
Ad
2003 in Santa Cruz. and at the Phonology- and Syntax Reading Group in Stanford in June 2003.
Also thanks to Iris Mulder for help with the typesetting. All errors are my responsibility.
MIT Working Papers in linguistics 49. Perspectives on Phases
329-367
© 2005 Michael Wagner
.
Asymmetries in Prosodic Domain FormatioD_
predicates and modifiers that has gone unnoticed in earlier studies. Two sim· pie
principles. 'Sister�Matching' and 'Subordination' are proposed. They prosody between sister constituents, and thus recursively map
negotiate the
syntactic tree structures to prosodic structures. The local decisions based on sister�constituents account for the
global distribution of accents in general and the
location of nuclear stress
in particular.
1.1. The Modifier-Argument Contrast Modifiers in Germanic differ from arguments with respect to their prosody relative to
1 984;
a following predicate (Gussenhoven. 1 984; Krifka. 1 984; Selkirk , 1992; Truckenbrodt, 1 993). Consid er first the case of ar
Jacobs .
guments (the relevant predicates are in italics, the relevant arguments are
underlined). I
(1)
Arguments a.
[ Sie hat ] [ einen Tango getanzt.] she has
a.acc Tango danced
'She danced b.
a tango.'
[ Sie ist ] [ in den Garten she is
getanzt.]
in the.acc garden danced
'She danced into the garden.'
( I a) and PP arguments (1 b) are subordi S ubordin ation means that they are realized with a reduced pitch range
Predicates following DP-arguments nated.
-
.
if not altogether deaccented. They are prosodically 'affixed' to the preced
I put the functor I underline the argument of the n:todifier, i.e. the modified
ing domain.2. Consider now the case of modifiers (again, in italics, and constituent):
(2) Modifiers
a. [ Sie hat J [ den ganzen Abend ] [ getanzt. ] she has the.acc entir� evening danced 'She danced the entire evening.'
IThe onbogmphic accent marks the location of high tone accents-non-final accents are not always realized, due to rhythmic effects. as will be discussed later. 20ne could say that the arguments 'incorporate' into the prosodic domain of the verb that
selects them; or converseJy, that the verb becomes affixed to the prosodic domain of the argument.
Indeed, structures that involve actual incorporation or affixation in German show similar prosodic patterns. I will not explore this point further in this paper.
330
Michael Wagner b. [ Sie hat J [ im Gtinen
she has
in.dat garden
J [ getan zt] . danced
'She danced in the garden '. Modifiers in (2) are mapped to their own accentual domain, separate from the accentual domain of the verb.3 The difference between modifiers and arguments outlined so far is only one half of the pattern, however. 1.2.
Linear Order Effects in Predication
Compl ements do not always induce the subordination of the predicate that selects them: When the predicate precedes the complement, two separate accentual phrases are derived. Consider the cases below in (3) that minimally differ from ( 1 ) in that the pred icate has raised to second position and thus preced es their argum ents. a. [ Sie tdnzte ] [ einen Tango ] . she danced a tango
(3)
b. [ Sie tdnzte ] [ in den Garten ]. into the garden he danced The examples contrast with ( 1 ) in that the predicate can have an accen t in this environment, without requiring narrow focus on the verb. Consider also arguments that follow the final verb in German. Complement clauses usually follow the selecting verb, in constructions often described as 'extraposition' (4a); but they can also precede them (4b).4 a. [ Er hat versprochen ] [ zu schweigen ] . he has promised to be silent
(4)
3The following examples were presented in Cinque ( 1 993 , 250. 254) with nuclear stress on the adjunct-this. however, is not the appropriate prosody in neutral contexts. Unless the modifier is rocused, nuclear stress in these examples falls on the verb. as was pointed o ut already in Truckenbrodt (1993). Jacobs ( 1999), and Ishihara (2001). (l)
a.
...[dass Karl] rein Bllch) (mit Muehe] [l�en leann]. Karl a book with problems read
that
b. ...[class Fritz] [gut) [k6chen kannJ. that Fritz wen cook can.
The word order in (I) actually seems to require marked focus on the predicate. The more neutral order is: 'dass Karl mit Muehe ein Buch lesen kann' . The example may then actually not be relevant for establishing the neutral stress pattern, since both word order and focus slrUcture are not neutral. 4lt is hard to test whether both (4a. b) are possible with neuual focus. In section 2 I will discuss ample examples iIIusuating the same asymmetry. with cross-linguistic data involving only all-new sentences with wide focus.
33 1
Asymmetries in Prosodic Domain Formation b. [ Er hat zu schweigen versprochen. ] he has to be silent promised 'He promised to be silent.'
Whi Ie the selecting predicate is subordinated and accentless in
( 1 ) and (4b), it is accented in (3) and (4a), where it precedes its argument. The following generalization captures this asymmetry descriptively5 (5) Prosodic Asymmetry •
•
When a functor A precedes complement B, a sequence of two prosodic domains that are on a par: A B. The last domain counts as the nuclear domain. When a functor A/ollows (an element from) complement B, A is subordinated: 13 A (unless A is focused or B is old information)
This generalization points in a similar direction as the approach in Steedman . (200 1 ) , who relates prosodic structure to the direction of composition in a categorial grammar framework. Note that it is sufficient if a part of the complement precedes a pred icate, in order for the predicate to subordinate. Some complements can op tionally follow a final predicate. In those cases, the predicate (here: gewarlel 'waited') fails to subordinate (a), as expected; the predicate subordinates even when only part of the argument precedes (b). a case of partial extraposition: (6)
a.
[ Hrer habe ich oft gewdrtel ]. [ auf den Zug ] [ der mich nach Hause bringt. ] Here have I often waited on the train that me to home takes
b. [ H{er habe ich oft [ [ auf den Zag 1 gewartel ) ]. [ der mich nach Hause bringt. on the train waited that me to home takes Here have I often 'I
waited here often for the train that would take me home. '
The generalization in (5) captures the data outlined so far. The subordina tion of predicates following their complement has the effect that both in the predicate -< argument order and in argument -< predicate order, the nuclear stress resides on the argument. This is because-as observed in Newman ( 1 946)-the last accent is perceived as the nuclear stress. I will assume here that 'Nuclear stress' simply refers to the last of a sequence of accents within a given domain, which is perceived by native speakers as the main prominence.6 SCompare observations about complement prominence in Schmerling (1976). Hayes and Lahiri ( 1991). Zubizarreta (1998). I can not review Iheir respective proposals in detail here. I may also related to cases of cyclic subordination in wh-movement discussed in Ishihara (2003). t7rbe generalization about the asymmetry recalls incisive observations on semence stress in Bierwisch (1968) and Bresnan (197 1 ) . which will be discussed later on.
332
Michael Wagner 1.3.
Linear Order EtTects in Modification
A closer look at the prosody of modifiers again reveals an asymmetric pattern depending on linear order-in fact, a pattern that is exactly parallel to the asymmetry observed between predicates and arguments. When a modifier precedes the modified constituent. two separate prosodic constituents are derived, as was illustrated in (2); however. when it follows the modified constituent. the modifier is subordinated. Consider as a first example the case of postposed sentential modifiers: (7) Postposed Adjuncts a. [ Sie hat ] [ getanzt, den ganzenAbend ]. she has danced the. ace entire evening 'Sie hat danced the entire evening. ' b. [ Sie hat ] [ getanzt . im Ganen ] . she has danced in.dat garden 'She danced in the garden' . Constituents following the final verb such a s those i n (7) are often called 'afterthoughts' , as if they were simply added after a sentence as some external addendum. However, they are modifiers integrated into the phrase structure of the sentence, as cases of VP coordination illustrate: (8)
[ Sie hat getanzt, ] den ganzen Abend, [ und Cocktails getrunken ] . she has danced the entire evening and cocktails drunk 'She danced and drank cocktails the entire evening.'
Example (8) also illustrates that it is sufficient if part of the argument of the modifier precedes it in order to trigger its prosodic subordination. analogous to the case of the arguments of predicates (6). The prosodic relation between modifiers and modified constituent is similar to that between predicates and arguments. Earlier treatments (e.g. Krifka . 1 984; Gussenhoven, 1984; Jacobs , 1992) focused on the contrast between modifiers and arguments, and thus missed out on the parallel between modi fiers and predicates.7 71l is not obvious whelher it is legitimate to call a phrase like 'the entire evening' a modifier. and group it with complement taking predicates, since maximal projections are not usually laken
to be able to take complementS and project (but see Starke (200 1) for a different view, discussed further in section 4). We can make sense of this if we adopt the analysis of adverbials proposed in Alexiadou ( 1997) and Cinque (1999). who view adverbials as specifiers of functional projections. It is not 'the entire evening' itself then, which is Ihe modifier, but a functional head that takes the OP as itS specifier.
333
Asymmetries in Prosodic
Domain Formation
1.4. A Complication: Rhythmic Effects Rythmic effects sometimes blur the generalization about the
asymmetry. in to be omitted. The accent is then effec tively 'switched-off ' . This hap pen s in situations where an accen t falls too close to a followi ng accent. The notion 'close' seems to depend also on speech rate. Consider the fo llowin g example. where both a rendition with two accents (a) or only one accent on the direct object (b) seem possible in a wide-focus, all-new context: that they allow pre-nuclear accents
(9)
a.
[ Sie tanzte ] [ Tango ] . she danced tango
b. [ Sie tanzte
Tango
].
she danced tango
Modifiers are again similar to predictes. In ( l Oa), it is possible to omit the accent on the manner adverb • gut' in ( l Oa) without changing the information
structure of the sentence. This seems harder in the case of a longer adverb, that places more unstressed material between its main stress and the following verb ( l Ob). ( 1 0)
Sie glaubt, dass er she believes that he a.
[ gut k6chen kannl well cook can.
.
h. [ ausgezeichnet ] [ k6chen kann. ] excellent
cook
can.
The extreme case of rhythmiC deaccenting is to drop all but the last accent. It is the rhythmi c component then
that the present approach holds responsible
for the pattern discussed in Chomsky ( 197 1 , 200): There is a rendition of the question in ( 1 1a) that only places an accent on 'shirt' with
wide focus on the
Dp'8
(1 1 )
a.
Was it an ex convic t with a red shirt that he was warned to look out for?
b.
No, it was an Automobile salesman.
-
Following ideas going back to Gussenhoven( 1 99 1 ) , Hayes
and Lahiri ( 199 1) , In
Ghini ( 1 993) , I view rhythmic effects as a phonological restr uc turin g . Bne idea is mat at least in fast
speech this is poss ible wimout presupposing any of me
deaccented material as being 'given'.
334
Michael Wagner
particular, rhythmic clash removal can erase prosodic boundaries and thus impoverish the prosodic structure. The rhythm principles have to be stated such that they never erase the domain that provides the last accent, but always the pre-nuclear domain. In the course of the paper, I will pOint to the effects of rhythm where necessary, but I will not discuss its grammar in any detail.9
1.5.
Summary and Outlook
Predicates and Modifiers, jointly referred to as 'functors', are mapped to a prosodic domain independent of their complement when they precede it; they are prosodically subordinated and 'affixed' to their complement when they follow the entire complement or part of it. Assuming the analysis of adverbs and adjectives proposed in Cinque ( 1 999) , functors are generally the pro jecting element, be they predicates or modifiers. This close match between prosody and syntax breaks down when looking at cases of secondary predi cation and modification, as we will see. I will argue that the generalization about subordination has to be stated based on the asymmetry of syntactic projection. The basic idea is that material that is higher in the structure but follows material that is lower in the structure is subject to prosodic subor dination. When looking at two sister constituents, the sister that projects syntactically counts as 'higher' . If it precedes the non-projector, the two are mapped to their own prosodic domain and are on a par. If it follows the non-projector, it is prosodicaUy subordinated. It is effectively affixed to the preceding prosodic domain. For the representation of prosody, I use the bracketed grid notation introduced in Idsardi (l992) , where unmatched brackets are used to delimit prosodic constituents. In the syntactic representation, I indicate which sister . is the non-projecting element in the tree structures by interrupting the link to the spine of the tree. Branches that are subject to prosodic subordination are drawn with thinner lines than branches that are not. Two principles, sister matching and subordination, negotiate the relative prosody between sisters:
( 1 2) Two Basic Cases a.
Head Intial Structure: Sister Matching
b.
Head Final Structure: Subordination
90ther reasons for lacking accents, apart from rhythm. relate to the backgrounding oflinguistic
material
as 'given' (Schwarzschild . 1999). I will not discuss these issues in this paper.
335
Asymmetries in Prosodic Domain Formation
(x(X (x(x a f3
/�
f3
a
(x (x(x f3 a
We can relate grid-sturcture with accent placement by stipulating that each and only top-line grid marks are realized as pitch accents. Newman ( 1 946) observes that the last accent in a sequence of accents is perceived as the most prominent. The subordination principle has therefore the effect that the complement is more prominent than the functor even when it follows IO Whether or not functors bear an accent when they precede their comple it ment does not change prominence relations and is decided based on rhythmic considerations. The proposed system maps trees to prosodic grid structures. The Nuclear stress is determined indirectly: The last element that is not subordi nated is perceived as the nuclear stress within any given domain. We return to the details of the prosodic mapping in section 5 , after confirming the basic asymmetry witha wider range of facts. I I -
2.
Prosodic Asymmetry: Predication
This section presents evidence confirming the generalization about prosodic asymmetry in (5) with respect to predicates and their complements. I present data from different dialects of West-Germanic (Dutch, English, German), involving infinitival or nominal complements. 2.1.
Predicates and Infinitival Complements
West Germanic languages differ in their prosody; However, once linear or der is taken into account, most of the apparent prosodic differences actually reduce to syntactic differences. This section looks closely at sequences of predicates. 1 2 Consider first the case of Dutch: l&rhis could just be stated as a telelogica1 constraint on the syntax-phonology mapping (e.g. Truckenbrodt ( 1 995). Schwanschild (1999)). requiring complements to be more prominent than functors. l iThe generalizations about the asymmelry of prosody outlined in this paper may ultimately relate directly to the asymmetry of syntax (cf. Haider (1993) and Kayne (1994» . It is beyond the scope of this paper to discuss the consequences of the prosodic facts for syntactic theory and vice versa. 121n presenting the cluster data I am tacitly assuming that they form constituents. These may have been derived via head movement-if we allow for head-movement in the first place. I will not explore the possibilities in detail. The numbers in the examples indicate Ihe path of selection
336
Michael Wagner
( 1 3) Dutch Predicate Cluster: Final Stress
... r dat hij ] [ wilde that he
wanted.to
] [ helpen ] [ verven}. help.to paint
The predicates are ordered according to their embedding, starting with the highest predicate. The actual output for ( 1 3) contains less accents than given here: The accent on 'helpen' is dropped, as indicated in (l4a). This seems to be due to rhythmic restructuring, which gets rid of clashes. One indication that this is the correct characterization of the data is example ( 14b). If a preposition separates the last two predicates, they are separated enough to both maintain their accents.
( 1 4)
a. . ..[ dat hij ) [ wilde helpen ] [ verven] b. ...[ dat hij ] [ wilde helpen ] [ met verven]
( 14a,b) suggest that nuclear stress in the Dutch predicate clusters is final, pre final predicates may also bear an accent. Predicate 2 in ( 1 4a) is in an accented position as indicated in ( 1 3), but gets rhythmically deaccented. The sentence in ( 1 3) can also be pronounced with only one accent on the last predicate. The rhythmic nature of accent-placement in the pre-nuclear domain is further evidenced by (15a vs. b) and ( l 5c vs. d) respectively. (15) Hij zei dat hij .. . he said that he ... . a . .. .wilde verven. wanted.to paint
b. .,.wilde helpen verven. wanted.to help.to paint
c . ...wflde kUnnen helpen verven. wanted.to be.able.to help.to paint d. . .. wilde mogen kUnnen helpen verven. ... wilde mogen kunnen helpen verven.
, ... wants to be allowed to be able to help to paint. '
The following example illustrates that rhythm is evaluated rather globally. In a matrix sentence in Dutch. the first predicate is considered for the computation between the
predicates. starting from the highest predicate • J '. to the one selected by it ·2'
an d so forth. Th e tree-representation encodes projecting constituents b y uninterrupted lines.
Predicates that receive an accent are indicated by a bold-faced branch. All sentences presented
involve sentence wide focus.
337
Asymmetries in Prosodic Domain Formation
of rhythm on the predicate sequence, and is not part of the cluster:
al though it has risen to first position
( 1 6) Hij wflde mogen he
wants
kunnen helpen verven. to be allowed to be able to h elp to paint.
"He wants to be allowed to be able to help to pain t
.'
One way
to make sense of this pattern is to say that the mapping to prosody p lac es accents on each predicate, which are then rhythmically organized in the later eurythmic component resulting in the omission of certain accents. Consider now the German counterparts of the Dutch predicate clusters: ( 1 7)
German Predicate Cluster:
...dass er malen helfen .. .that he
Initial Stress
wallte. paint help want.
3
2
1
Main stress in German falls on the first p redica te. No accents can occur in the
post�nuclear domain unless one of the predicates is contrastively focused-in
which case it would receive main prominence. There are secondary stresses after the nuclear one, but these cannot be realiz ed as pitch accents. This is true independent of the number of predicates that follow: ( 1 8) [ Er sagte dass er ] [ malen helfen kOennen dUrfen wal lte ] he said that he paint help can be.allowed wanted
.
The two languages also differ in the linear order of predicates, apart from the linear location of main word stress: while the p redicates in Dutch are ordered according to embedding, the order in German is the exact inverse. The two d iffe rences initial vs. final stress. embedding vs. inverse order. conspire to keep main stress on the most deeply embedded predicate according to the path oj selection. The following paradigm shows three of the possible orders of a par� licular predicate sequence in German. When predicates are ordered according to embedding as in (a), this order is often taken to involve 'extraposition· . Different orders are possible. however. in s o�call ed 'restructuring' environ� ments (e.g. in b.c).13 ,
IJ-rhere are many syntactic differences between 'exuaposition' an d 'restructuring' thal l will nOl address in this paper-restructuring derives what appear to be monoclausal constructions
338
Michael Wagner
' ... weil er ihr. . . , . . .because he...
( 1 9)
b. [versprach] [zu schweigen zu versuchen].
f ...::2 - d j - d �2
c. [zu schweigen zu versuchen versprach.]
j�2� 1
a. [verspnich] [zu versuchen] [zu schweigen].
be silent try promise , ...promised her to try to be silent.' The example ( 1 9a) is similar to predicate clusters in Dutch ( 1 5), in that main stress is rightmost and secondary accents precede the main one. The fact that the median predicate does not necessarily lose its accent rhythmically as in the Dutch example ( 1 3) (although it may in fast speech), maybe due to the fact that there is unstressed phonological material-the prepoSition-intervening, preventing a clash. Remember the similar pattern in the Dutch example in ( 1 4). where also a preposition separated two predicates. The prosody in Dutch extraposition constructions is generally equal to that of the German cases. Note also that the facts are equivalent in the relevant constructions in Englishl4 :
(20) He wanted to be able to help to succeed. That Dutch and German indeed do not differ in their prosodic systems be
comes apparent in predicate clusters that are attested in both languages: 15 (2 1 )
• facilitate scrambling between clauses, and allow pronouns that are arguments in lite lower clause to be affixed on the matrix verb in second position. etc. (d. Wurmbrand. 2003). Restructuring does not always result in a different word order between me predicates (Haider.
that, e.g.
1994).
14Since English does not allow cluster formation and has a rather fixed word order between predicates. only the evidence from OP's preceding predicates discussed in the next subsection serves to show subordination in this language. A look at complex nominals, however, shows further evidence for the asymmetry at work: (1)
a
[ a trainer ] [ of shiyers ] [ of vampires ]
b. [ a trainer ] [ of vampire slayers c. [ a vampire slayer trainer. ]
1
Each predicate preceded by its complement is subordinated. a... expected. ISHere, the OP argument preceding the cluster is made 'given' (old information). in order to prevent subordination of the cluster.
339
Asymmetries in Prosodic Domain Formation
a. . ..dat Jan M anegiven kan ) gezfen3 hebben 2. that Jan Mary could seen have
b. ...weil
sie ihn hat) malen 3 wollen2.
because she him has paint
wanted
In this example, the second predicate 'hebben ' is preceded by its comple ment 'gezien ' , which effects its subordination. The modal is unstressed, which is unsurprising since it in a position that loses stress via the rhythmic principles
that disallow clashes. The next example illustrates again that it is in fact sufficient if a subconstituent from the complement precedes (this word order is not accepted by all native speakers of Dutch): (22)
a. . . .dat Jan Marie gezfen3 kanl hebben2.
'that Jan could have seen Mary. '
b. ...weil er es kaufen3 wirdl konnen2' because he it buy will can
3
»"" 1
2
The distribution of accents so far follows the generalization in (5) plus rhyth mic deaccenting. Consider the following cases of particle climbing, which
further illustrates that each predicate preceded by part of its complement (the particle) is subordinated. Pre-nuclear predicates may lose the accent to avoid clashes:
(23) Climbing up the Cluster (cf. Evers , 2001 , and ref. therein) Het labyrinth waar we hem niet over... a.
WIllen hoeven laten
ii81 denken.
b. WIllen hoeven iiaJ laten denkeo. c.
iii] zullen hoeven laten denken.
1 -<2-<3-<5-<4 1 -<2-<5-<3-<4 subordination 5 -< 1 -<2-<3-<4
about will need let think
'The labyrinth about which we won't let him reflect • The correlation between prosody and syntax in predicate clusters and extra position constructions was already observed in Bech (1955157). While the
generalization that nuclear stress falls on the most deeply embedded con
stituent is already predicted based on the approach based on major and minor
340
Michael Wagner projection lines in Cinque ( 1 993 , 269ft) and also in the approach pursued in Jacobs ( 1 99 1 . 1 992) . the linear asymmetry observed here is not: predicates are subordinated exactly when their complement or a subconstituent from their complement precedes.
2.2. Predicates and Nominal Complements Predicates preceding their complement can receive an accent in English (a). This is also true for DP-complements (b).
(24)
a. She wanted to h el p (0 succeed.
b. She wanted to help to paint the house.
The case of an infinitival complement preceding its selector is unattested in English. but consider DP-complements in contexts that show OV order in English:
(25) What did she want to change before moving in? She wanted to have the walls painted. Integration with subjects is in generally possible in English, both with unac cusative (a) and unergative (b) verbs) 6• Subordination is also observed when there is more than one predicate (c):
(26)
a. [Gasolfne evaporated]. b. [the deanla telemarketer called]
c. [The dean was expected to come.]
Subordination of predicates following arguments can also be observed in Dutch and German. even when multiple predicates follow an argument. 17 16It has been reported. however, that unaccusatives tend to phrase with the subjects whereas unergatives don', (Selkirk , 1995; Hoskins . 19%). This at least does not hold for all unergative and unaccusative predicateS, as illustrated here. r7 At this point, we can look at evidence that the asymmetry outlined in (5) also applies to accentual domains that are pre-nuclear. Consider the case of a complement of a predicate that is not the lowest predicate in a sequence:
(I) Sie hat Marla vezsprochen zu bleiben. she has Mary promised to stay
The argument Maria is selected by versprochen • promise' •
which then takes a second argument stay'. This example illustrates that indeed only those predicates subordinate that are preceded by an element from its complement domain-not all predicates in a cluster blindly subordinate to 8 preceding DP argument. The nuclear stress falls on the rightmost accentual phrase, provided by the predicate 'bleiben'. This example illustrateS that the distribution of secondary accentual phrases obeys the same principles and shows the same asymmetry. It is not simply guided by rhythmic principles. zu bleiben 'to
341
Asymmetries in Prosodic Domain Formation
(27)
a.
.
.. dat hij [een muur6 wildel mogen2 kunnen3 helpen4 vervens·]
that he a
wall
want
allow
can
'he says that he wants to be allowed to
help
paint
be able to help to paint a
wall.'
b.
weil er [ein Bfld6 malen helfen koennen diirfen wollte.]
...
because he a picture paint help can be. allowed wanted
The post-nuclear stresses, both in Dutch and German, cannot receive an ac centual realization unless the relevant predicate is focused or the preceding argument constitutes given information. A focused predicate, however, would receive nuclear stress. All predicates following the direct object are subordi nated, that is. they are at least severely reduced in pitch range, if not altogether deaccented. The asymmetry relevant here is again that in a neutral context,
two
accentual phrases are derived when the predicate precedes the complement while
one accentual domain is derived in cases the predicate follows the
complement. Relevant data were already noted in Newman
( 1 97 1 ) . though only looking at the main (28)
( 1946); Bresnan
stress:
a. He had plans to leave. (argument --< functor) b. He had plans to leave. (functor
--<
argument)
For English is is reported that the verb phrases together with afollowing argu ment, contrasting with Dutch and German in the directionality of phrasing. But note that uncontroversially, English differs from Dutch and German in that the verb can bear an accent. The evidence for the phrasing of verb and direct object often given is the application of the rhythm rule:
(29) Evidence for Phrasing Kenesei and Vogel ( 1995) a. .. .in English: Rhythm Rule: [They managed] [to outclass] [Delaware's canteen]. b. .. .in German: No Accent on Verb [Sie haben Delaware's Kantfne Ubertroffen]. In Dutch and German, the verb does not receive an accent due to subordi nation; in English, on the other hand, two adjacent accentual phraSes (verb and object) are derived that are then phrased together into a higher prosodic domain-this domain allows rhythmic restructuring if there is a clash. Both Dutch and German show a similar prosody to English when the verb precedes a direct object:
342
Michael Wagner (30) Sie UinZle Tango she danced tango
Considering the evidence discussed, we can conclude that the three languages have a very similar if not identical mapping from syntax to prosody. and all show the asymmetry outlined in (5).
2.3. A Second 1)rpe of Predication There are a number of cases that appear to contradict the asymmetry outlined in (5). As is well known, the subject in an intransitive sentence sometimes phrases separately from the verb, e.g. when it encodes information that counts as 'given' in the context. The following context is set up to facilitate wide focus in the embedded clause-but with a backgrounded or 'given subject. t
(3 1 ) What did you say the dean did? I just said that [The dean] [arrived].
The same is true in OV contexts. with the direct object preceding the predi cate(s) in German and Dutch. If the direct object encodes ' given' information, the predicate(s) form phonological domain of their own: (32)
a . ...dat hij [de muuf6 wildel mogen2 kunnen3 helpen4 vervens.l that he a wall want allow can help paint b. ...weil er [das Bild6 malen helfen konnen dUrfen wollte.] that he the picture paint help able allowed wanted •
.,.that he wants to be allowed to be able to help to paint a waH,'
These examples contradict the generalization in (5), since predicate clusters preceded by their argument are not subordinated but receive nuclear stress. But these sentences differ in their syntactic structure. Cases with nuclear stress on the verb in Dutch were argued in Neeleman and Reinhart ( 1 998) to involve scrambling of discourse-old information, e.g. Counterexamples even exist where the argument does not constitute discourse-old information. Consider the following example:
(33) Why did they close the factory? a.
[ The factory ] [ went bankrupt ]
b. [Gasoline evaporated]. c. [A w6rkerl [evaporatedl.
343
Asymmetries in Prosodic Domain Formation The factory' is given in (33a) and thus the verb receives main stress. A
DP containing new information (as in 33b) shows the normal pattern. 'A worker' in (33c) does not have to be 'given' but still nuclear stress falls on the verb. The indefinite is arguably treated as a member of a set inferred from the background (the workers of the factory). As in (a), the verb receives an independent accent. Also. unlikelihood/unexpectedness of the predicate may play a role. 1 8 19 Similar contrasts as in the case of unaccusative verbs can be con structed with unergative verbs. In (34b) , the subject is again an indefinite that is interpreted as a member of set made salient by the discourse (i.e. a partitive relating to a set in the background). (34) Why did they interrupt the play? a. [A child was crying]. b. [An actor] [was crying]. The stage-level, individual-level distinction is directly reflected in prosody (Schmerling ( 1976); Diesing (1 992); Kratzer ( 1995b», and here discourse new-/oldness is not at stake. Individual-level predicates resist subordination: (35)
a. [your eyes are red]. b. [Your eyes] [are blue].
The cases where predicates that are preceded by their complement are not sub ordinated all seem to involve operators (focus. partitive, generic. topic ...) that add further structure. I argue that this further layer of structure is responsible for the different prosody. The idea is that the predicates really take an empty category as their complement. either a trace or an empty operator, which is bound by the c-commanding actual argument which functions as the speCifier of a functional projection (or alternatively as the complementlrestrictor of the operator as in b): (36) Predication without Subordination a.
�
Argument
../"-... F � e
Predicate
b.
�
�
Argument
F
."".-....
Predicate
IBntis type of factor was in fact claimed in Bolinger ( 1972) to be sufficient to explain intonation patlems-dlis is clearly incorrect. given the fact that e.g. the differences that only relate to linear order but leave information structure unaffected (e.g. the comparison between Dutch and German predicate sequences) would not be explicable in terms of expectedness or similar considerations. 19A relevant factor may also be the thetie/categorical distinction Krifka (e.g. 1 984 . with re.'Ipect to prosody.).
344
Michael Wagner
The same treatment is applied to secondary predicates. Consider the case of depi ctive predicates which also fail to subordinate (cf. Pylkkanen (2002) on the structure of dep icti ve predicates). ,
(37)
Subject Oriented Depictive
�
�
sue
�
Voice
yp /""'-...
see Sue;
XP
d .� lire
Peter
saw Peter ured• .
The counterexamples to (5) are then taken to be o fone kind: cases i n which the predicate is not really the projecting element, and there is no oven argument in complement position. The generalization about prosodic asymmetry can
be restated in terms of projection
(38)
as follows:
Prosodic Asymmetry •
When a projector A
precedes its complement B, a sequence of
two prosodic domains that are on a par:
A B. The last domain
counts as the nuclear domain. •
Wh en projector A follows an element from its complement do B. A is subordinated: B A (unless A is focused or B is old
main
information) There are two syntactic ways to construct a predicate-argument relation, so the functor-argument relation is not sufficient to capture the prosodic asym 20 metry. The generalization about prosodic asymmetry is syntactic in nature, and relates to the asymmetry of projection.
2.4.
Summary
All evidence taken together shows that the asymmetry observed in (38) applies to all three languages. There is a single prosodic system, apparent prosodic differences reduce to syntactic differences. lD-rhe generalization could be stated in terms for functor/argument if one adjusts the types of the VP and the depictive such that the VP takes the depictive as a comple men t. Within the framework of Heim and Kratzer ( 1 998). the operation of predicate modification is used to combine predicates to form new predicates. preserving the type. Pylkklinen (2002) uses
predicate modification to combine VPs with depictives to form complex predicates. It is not
clear. however. that it makes sense to type shift the VP such that it takes the depi ctive as a complement. A discussion of the semantics of predication would be necessary at this point. but has to be postponed to futu re work (1banks to Robert Truswell for brining up the issue).
345
Asymmetries in Prosodic Domain Formation
3.
Prosodic Asymmetry: Modification
This section shows evidence that modifiers are subordinated when they follow the modified constituent, unless they are focused. Again. there is a class of modifiers that fail to obey the generalization. These are analyzed as cases of 'secondary modfiication' • with distinct syntactic properties.
3.1 .
Scrambling in Vitro
Determiner Phrases, especially definite descriptions. can be modified by re strictive modifiers. They are close in meaning to restrictive relative clauses. The following example refers to a man that was seen yesterday. This type of modifier in first position may have a colloquial ring to it. but is nevertheless acceptable to all speakers I consulted. A determiner phrase in first position with several restrictive modifiers allows various word orders-a nice testing 21
ground for the effects of word order permutation on prosody.
21The discussion of the modifier here is at odds with the discussion in Barbiers ( 1995 , 132). (1 ). Since the conflicting time reference in the main clause renders this sentence reportedly ungrammatical, Barbiers analyzes the adverb in first position 8.-' a sentence adverbial. The derivation of the following sentence involves then scrambling of a DP to a specifier position of an adverbial. and subsequent movement of the adverbial phrase including the DP. a constituent he labels 'Pseudo-OP', to first who looks at sim ilar modifiers in Dutch. Consider the example in
position. (1)
Die man gisteren vertelde (*vandaag) de waarheid. the man yesterday told today the truth
The star that Barbiers attributes to this example vanishes in a context that makes this sentence plausible. at least for two Dutch native speakers 1 consulted. Similarly. Gennan native speakers tend to reject its German equivalent. but only if one does not motivate it by providing a context. Quite generally, there is no problem in introducing a DP modifier with conflicting time reference in the first positi on in German. if the example is pragmatically plausible. I owe the following example to Paul Kiparsky (p.c.) (its Dutch equivalent was a1so.accepted by two Dutch n ative speakers).
(2) [Der B6mbe gestern] wird (morgen] [einen Krfeg auslosen). the bomb yesterday will tomorrow a 'The bomb yesterday will unleash a
war
war tomorrow. '
unleash
Also. it is quite possible to coordinate two DPs, both modified. with conHicting time adverbials.
(3) Die Bombe gestem und das Attentat heute werden in Zukunft viele Probleme the bomb ye.<;terday and the assassination today will in future many problems verursachen. cause Moreover, many adverbials that are sentential adverbiaJs (e.g. probably, un!orlufUllely) are
strictly prohibited to occur together with a DP in first position (and again the same seems to hold
346
Michael Wagner
(39) Restrictive DP-modifiers a.
[Gestern I 1 [dieser Mann2]
war rein Marsmensch].
b. [Dieser Mann2 gestern 11 war [ein Marsmensch ]. 'That man yesterday was a Martian.' Depending on the order of the DP 'der Mann' relative to the modifier, there are either two (a), or one (b) accentual phrase in the first position preceding
the verb in V2 positi on .
subordinated.22
A m odi fier that follows the modified constituent is
It is not the case that all 'extraposed' material in complex DP struc subordinated. Complements that follow the head noun, e.g.. do not subordinate. Consider the following examples, which involves both a tures is
post-posed DP modifier and an argument (example (without prosody) due to Drosdowski , 1995 , 798):
(40) [ seine H6ffnung gestern my
hope
] [ au f Hilfe ] .
yesterday
for help
While the DP modifier 'yesterday' subordinates. since it follows part of the modifiee, the complement of the head noun remains accented. true in Dutch). I
take it then that there is not Pseudo-DP ([DP AdvJ) movement to first position.
All adverbial that do coOccur wilh a
DP in first
position are in fact DP modifiers-al least in
German and possibly also in Dutch, and show Ihe relative asymmetry in their prosody depending
on word order with respect to the modifiee. This asymmetry may not be testable in Dutch, where sentences like Ihe following are at least highly marked if not unacceptable (judgement Barbiers , 1 995 , 134):
(4) *Gisteren de krant meldde dil voorval niet. yesterday the paper reported
this incident net
Again, this particular example sounds odd out of context in German as well, so it may be nOl thc
best test; but even cases like the acceptable one in (39a) were at least very marked for my Dutch informants.
llA similar asymmetry was pointed out to me by Henk van Riemsdijk. In indefinite DPs, it is
possible to let an adjective follow the NP-but only if the determiner is
asymmetry can be observed:
(1)
'doubled'.
Again. the
a. Albert will ein schnclles Auto (kaufen).
b.
Notc that
Albert will Albert
ein Auto (kaufen), ein schnelles.
wants a
car buy
the post-DP modifiers
a
fast.one
in fact is postposed after sentence final verbs-the syntax of
this construction would need some further discussion.
347
Asymmetries in Prosodic Domain Formation 3.2.
Repetitive Again
The adverb 'again' in English has many readings but not all of them are available in each syntactic position. Stechow ( 1 996» argues that the different readings arise by placing the same lexical item 'again' in different structural positions. Consider a case of wide-scope 'again' : (4 1 ) Wide-Scope 'again' a. [ Again ]. [ somebody opened an emergency door ]
b. [ Agmn ]. [ somebody opened an emergency door ] (focus on adverb) The relevant reading of 'again' is the one where it has widest scope. and expresses that there was another event in which somebody opened an emer gency door. In other words, 'again' takes scope over the entire proposition 'somebody opened an emergency door' . Positioning the adverb in this position yields two intonational do mains. the one containing the adverb and the one containing its complement. where the nuclear prominence is judged to be within the complement. which may include various accentual phrases. Deaccenting and subordinating the complement proposition is only possible when the adverbial itself is focused. Consider now a different linear position , in which the same scope is readily available: (42) Wide-Scope 'again'
a. [ [ Somebody opened the door], again ] b. [ [ Somebody opened the door] again ]
When the adverbial follows the modified sentence, it is subordinated, and seem to get affixed to the last accentual phrase in the intonational phrase of the complement proposition, as in (a), The pronunciation in (b) is available when the fact that somebody has opened an emergency door was just mentioned or is assumed to be salient information in the present conversation. Consider the following two contexts, which facilitate one or the other pronunciation. (43) Why is the guard so upset?
a. [ [ Somebody opened an emergency door], again ] (neutral) b. [ [ Somebody opened an emergency door] again ] (possible only if the fact that somebody opened an emergency door is taken to be part of the background) Compare the following context:
348
Michael Wagner
(44) Yesterday there was an alarm. because somebody opened gency door. But why is the guard so upset now? a.
an
emer
[ [ S6mebody 6pened an emergency door]. again ] (dispreferred in this context)
b.
[ [ Somebody opened an emergency door] again ) (preferred in this context)
We conclude that the expected asymmetry is indeed attested.23 If we analyze 'again' as an argument-taking predicate that leaves the type ofthe complement constituent unchanged, we can link this asymmetry to the generalization in
(5 ) . 24 3.3.
Frame-Setting Locatives
Locatives have different readings. again depending on where in the structure the locative is placed. Consider the following wide-scope locative, which Maienbom
(200 1 ) characterizes as setting the
frame for the proposition ex
pressed by the rest of the sentence. The presence of an additional sentential
locative modifier forces the frame-setting reading (45), since two locatives of
the same
type
would be contradictory:
(45) In Italy. Lothar bought his suits in France. (Maienborn , 200 1 . 1 97) The intended reading is one where the frame setting locatives narrows the time under consideration to 'the time when Lothar was living in Italy', while the the lower locatives specifies where the events of buying suits took place. There are two possible linearizations of frame-setting locatives, and again the prosodic asymmetry is observed:
(46)
a. [ In Italy.
1 [ Lathar bought.his suits in France ].
2JThe data in German is more involved. It would take too long to discus.'! it here. but see some relevant discussion in the next section. 24ft seems that some similar modifiers like 'for the third time' have adifferenl prosodic pattern. Even without mentioning previous door openings. the following sentence receives preferentially stress on the adverbial:
( I) Why is the guard so upset? S6mebody opened
an
emergency door for the thfrd lime.
Is this because it is a 'heavier' modifier (Lakoff, 1 972)1 The more material is in the modifier, the more a context may be made salient that involves a focus-so the problems of controlling information sttuchlring surface again in this case; third. note that there is a scalar item in this adverbial: 'third'. Scalar items invoke a set of linguistic alternatives, a 'Horn scale'-this may motivate focus on the adverbial here.
349
Asymmetries in Prosodic Domain Formation
b. [ [ L6thar bought his suits in France ]
in Italy ].
Frame-setting locatives contrast with lower locatives, which receive an accent
sentence finally (4 6b). How they can get an accent despite of the fact that they
are modifiers apparently following the modifiee will be discussed in section 3 .6. Lower Locatives cannot occur in sentence-initial position:
(47) a. *[ b.
[ In France, ]
[ L6thar bought his suits
*[ ( I n italy ] [ I n France. ]
] in Italy J.
[ L6thar bought his suits
J.
Similar asymmetries exist in German. This type of postposing of an adverb after a final verb (as in 7) may in fact be restricted to frame-setting locatives in German. 2S
(48)
a. [ In Frankreich hat ] [ Peter immer ] [Tango getanzt ].
b.
[ Peter hat immer ] [ [ Tango getanzt ] in Frankreich ]. Peter has always
Tango danced
in France
Frame-setting locatives show the relevant asymmetry both in English and German. Jacobs (200 1) , Maienborn (2001 ) note that frame-setting modifiers are similar in syntactic position and their semantics to topics. The same
parallel may apply to the intonational realization of these modifiers and topics. I will not explore this in any detail here. The crucial point here is the observed prosodic asymmetry.
3.4. Accented vs. Unaccented Adverbs The proposal regarding modifiers and their prosody forward in this paper may
throw some new light on a difference between 'accented' and 'unaccented '
adverbials in the middle field in German. Lower type adverbs. such as manner adverbials, receive an accent of their own (a). Higher adverbs. such as the equivalents of 'unfortunately', 'probably' , 'always'. and wide-scope 'again' often do not receive an accent, even in neutral contexts.
(49)
a. Sie hat haeufig gut geschlafen. she has often
well slept
(stress on 'gut' can be rhythmically omitted) b. Sie hat leider/wahrscheinlich
geschhifen.
she has unfortunately/probably slept 2SHow lhe verb gets into place in second position will not be discussed here. The prosody of the modifier 'immer' will be discussed in lhe next section.
350
j Ii ·
Michael Wagner
One might attribute this fact to some inherent property of these adverbials however, in first position, ' unstressed' adverbials are stressed and receive an accent:
(50)
a. Wahrscheinlich hat sie geschhifen. probably
has she slept hat sie gehkht.
Leider
b.
unfortunately has she lauged And moreover:
'stressed ' adverbs are de-stressed in neutral V2 sentences,
where the main predicate has risen to second position instead of an auxiliary.
(5 1 )
Why did she want to buy a new bed? a. Sie schlief haeufig schlecht. b.
... wei I sie haeufig schlecht schlfef..
These facts relate to the on the generalization in
(5). Manner adverbials
modify the main predicate and a subordinated when they follow it. Higher sentential adverbials modify the entire sentence. and a subordinated when they follow (part of) their complement-as they do when they are not initial. Of course, as usual focus can override these generalizatlons.26
3.5.
Subordinated PPs
Consider two linearizations of locative PPs
(5 2)
a. Maria wollte
in German: ein Geige umtauschen.
in der Stadt
Mary wanted in.the supermarket a violin exchange b. Maria wollte eine Geige in der Stadt umtauschen. Maria wanted a violin
in.the city
exchange
The locative is subordinated only if it is preceded by an element from the event that it modifies
(52b); if it
precedes the entire event, it is not subordinated
(52a). just as expected. This case has to be distin guished from resultatives, discussed in Krifka ( 1984) , Uhmann ( 1991): 26
Soh (2001) argues that adverbs
that are heads
are
unstressed and adverbs that are phrases
are stressed in Shanghai and Hokkien Chinese. One test for the SIaIUS of Ibe adverbs is modifi
ability. It seems that modifiability crosscuts Ibe class in Gennan at least in one direction, since
bolb
('sehr wahrscheinliclt') and 'well' ('sehr gut') allow Ibe addition of a modifier. true that 'higher' adverbs lend nOI 10 be modifiable (consider Ibe con trast with
'probably'
However, it is
subject oriented '(*sehr) clevererweise' (cleverly, not modifiable) and the manner adverb 'auf
(sehr) clevere Weise' (ceverly,
modifiable)-but note also the syntactic differences
(compound vs. PP). The modifier in fact appears inside of Ibe
in realization PP. Further inquiry is needed here.
35 1
Asymmetries in Prosodic Domain Form ati on
(53)
a. b.
Maria hat gas Auto in die Garage gefahren. Maria has the car in the garage driven *
Maria hat
in die Garage gas Auto
Maria has the car
gefahren. in the garage driven
The subordinated PP in (53a) is a resultative predicate. and cannot precede the direct object (53b). The PP is the comp lement of the embedding verb Jahren. Accordi ng to Krifka ( 1 984) . the PP is interpreted as pan of the predicate. This is responsible for the subordination in this example. When the direct object is pronominalized in either of the two t ypes of ex amples. it is presumably moved to a highe r projection and therefore will not interfere with the prosody of the lower domain . Then the difference between the two types of PP emerges:
(54)
a. Maria woHte sle In der Stadt umtauschen. Maria wanted it in.the city exchange b. Maria wollte es in die Garage fahren. Maria wanted it in the garage drive
Note the two different patterns with respect to the last predicate in the sen· tence: Since the locative PP in (54a) im Supermarkt modifies the entire event i ncl uding the predicate umtauschen, nuclear stress now falls on the predic ate. The goal PP in die Garage in (54b), however. is the complem ent of the predi· cate Jah ren . thus 'fahren' is subordinated. This pattern lends further support for the the asymmetry outlined in (5). 3.6. A Second 1Ype of Modification This section illustrates a second type of constructing a modification structure. analogous to the case of secondary predication (section 2.3). In order to derive the correct prosody for the modifiers discussed so far , we have to analyze the modifier as the projecting element. This view of adverbials may not be surprising. at least if we take adverbials to involve functional projections, and consider adverbs constituents in their specifier. following Alexiadou ( 1 997); Cinque ( 1 999). Modifiers involve projecting heads that select an argument-the modifiee. Whether the adverbs are the realization of this projecting f�ature or are in the specifier of the mod i fiying projections is an independent issue that will be discussed in the next section. There are modifiers. however. that fai l to subordinate in cases where they follow the modifiee. Consider low sentence final adverbials in English.
(55)
352
a. She sang beautifully.
Michael Wagner b. She arrived on Monday. c.
Lothar bought his suits in France. (cf. example 45)
Pesetsky ( 1 995) and Phillips ( 1 996) show evidence that certain types of ad verbials are really lower in the structure than the material that they modify. Larson (2004) argues that they are base-generated low within the VP. Haider (2000) presents evidence that they are within the VP and modify an empty constituent to their right, rather than being adjoined high. This is of course exactly what is needed to save the generalization in (38): If they are low in the structure, and do not project up, then they are non-projectors occurring to the right of the projecting material. thus they are not expected to undergo subordination: TP
(56)
�
She
VP
�XP
sang
/'-..... beautifully e
These modification structures are then analogous to cases of secondary pred ication, and could be called 'secondary modification' .27 3.7.
Summary
This section presented evidence for the claim that the prosody of modifica tion shows the same asymmetries as the prosody of predicates and arguments, . namely the one outlined in (38). Since earlier approaches compared argu ments and modifiers, rather than predicates and modifiers, these parallels have gone unnoticed.28 27 Tbe nuclear stress in cases of secondary predication compatible with the approach that places nuclear stress on the most deeply embedded constituent along the major projection (Cinque ,
the the line that connects the g-nodes, and cases where two XPs are sisters,
1993, 269). The major projection line, according to Cinque. is
root of the tree with a terminal node, following
follows the recursive side of the tree, which he assumes to be fixed for any given language (in
Germanic, it's the right hand sister). The principle of subordination, cyclically applied, is an implementation of this idea. that does not invoke the concepts of Jl theory. 28A thorough look at all classes of adverbs defined
by their position in the phrase structure (cr.
Jackendoff. 1972; Cinque , 1999; Nilsen, 2002; Frey , 2003 . for different views) is beyond the
scope of this
paper.
353
Asymmetries in Prosodic Domain Formation ..
4.
Specifiers, Heads, and 1\vo-Place Functions
A notorious problem for the assignment of prosody in general (cf. discussion in Cinque , 1993) and for the present proposal in particular is the prosody of specifiers. Consider the nominal adjuncts involving ·with ' . 'With' would have to be analyzed as a modifier taking two arguments:
(57) The w6man with the hat. Given the analysis so far, the modifier should be the main projection of the construction . 'with' combines with 'the hat' forming the modifier 'with the hal', modifying the head noun 'woman' . We would expect then subordination of 'with the hat', since it follows the modifiee-a prediction that is quite apparently false. Why does the principle of subordination fail to apply? Note that even if 'with' is the main predicate, the entire constituent is not a Pp, but an NP. The determiner 'the' selects a noun, not a prepositional phrase.
(58)
NP
�PP
woman
�
with
the hat
Of course. if this is the correct structure and the projector is really the head
noun 'woman', the problem with respect to the prosody disappear: the pro jector precedes the non-projector. so two accentual domains are expected. While 'woman' is the argument of the functor 'with the hat' . the functor remains un subordinated and receives its own accentual domain since it is its sister that projects. A similar rationale applies to coordination. Again, 'and' can be seen as a two-place function. The .prosody is similar to the previous example:
(59) Coordination The woman and the hat. The analysis that I have to adopt in order to assign the right prosody is the following:
(60)
DP
�BP
DP
�
the
354
woman
/'-....
and
the hat
Michael Wagner
This type of analysis was proposed for coordination in
Munn (2000). The
first conjunct projects. and the second conjunct including coordinated is an adjunct to
the
first. Again, the prosody goes with the syntactic projection,
and not with the functor/argument relation. it seems. The preceding two cases involved what looked like two place func tions (with , and), and prosodic problems that arise. This following case presents another case of a two-place functor which creates problems for the present system:
(6 I)
The w6man wore the hat.
Under the assumption that it is 'wore the hat' that takes the subject 'the w6man' as an argument (Marantz. 1984) , the asymmetry
in (5)
would predict
subordination of 'wore the hat' under 'the woman'-a patently false predic tion of the system described so far. But uncontroversially, 'The woman' is not in its base position. The problem can be resolved by positing that it is the specifier which in fact projects, following the proposal in Starke (200 1), who takes specifiers to be complement-taking projectors. It has raised
to
a higher position in
the extended projection, or. if we follow Starke (200 1 ) , the specifier itself projects a higher element of f-seq, say T, or maybe it realizes the functional head that introduces the argument position of the agent in the
first place, say
voice (cr. Kratzer. 1 995a; Pylkldinen . 2002).
(62)
TP
�
�
the
woman
VP
�
wore
the hat
This analysis is also compatible with the cases of modification and coordi nation discussed above. The three examples all posit similar challenges to the present analysis, and
the tentative proposal
is to resolve all of them by
adopting the analysis of speCifiers independently advocated in Starke (200 1).
But a generalization goes uncaptured here. it seems: In all three cases, bi
nary functors taking two arguments are involved. Only one of the arguments seems to be allowed to stay in the basic projection. This m ay relate to a generalization about predicates not tolerating two arguments in their domain
(Anagnastopou}ou and Alexiadou, 200 1 ) . Arregi (2002) offers a different in analogy to the compound rule (e.g.
solution
to the specifier problem:
Libermann and Prince ( 1 977». he
proposes that the reason specifiers do not receive nuclear stress is due to the
355
Asymmetries in Prosodic Domain Formation fact that the co nsti tuent they attach to is internally branching. 'Specifiers' are defined in bare phrase structure (Chomsky , 1 995) as sisters of branchi n g heads. The idea is then that the sister of a head is more prominent, unless the head is branching, in which case the head is more promin ent. This makes the right distinction between comp lemen ts and specifiers. But this solution does not work for cases where the complement o f an embedded predicate is preposed (other relevant cases are discussed in (22), (23) and (27) : (63) Maria hat z.u schweigen versucht zu versprechen.
Maria has to be silent
tried to promise
The condition that a sister constituent be branching is not sufficient. Only in cases where the non-projecting element is an argument of a two-place functor (such as 'with', 'and' , transitive verbs) does the principle of subordination fail to apply, but not in cases where the complement of an embedded predicate is the sister of a branching cluster of predicates. The problem of the correct analysis of specifiers remains to be solved. 4.1-
Some Apparent Mismatches
This section introduces some apparent counterexamples to the pattern de scribed so far. and explains how they can be accounted for. 4. 1. 1. Arguments that Pattern like Modifiers
Some arguments pattern with modifiers and contrast with arguments in that they don't get nuclear stress in what appears to be an out-of-the-blue context. It is important to note that they also pattern with modifiers in that they can
be extraposed without a resumptive pronoun:29
(64) a.
Du
kannst dich auf Maria verlassen.
29Even for the type of argument in (64), it depends on the context whether these arguments are construed as receiving nuclear suess or nol In the following context, the argument receives nuclear stress and consequently cannot be extraposed: ( I) Warum dachte er, dass er nieht zu
lemen braucht? why thought he that he does not learn need 'Why did he think that he doesn't need to learn?' a. Er hat sich auf seinen Nachbam verlassen. b. nEr hast sich verhissen auf seinen Nachbam. he has himself relied on his neighbor 'He relied on his neighbor:
356
Michael Wagner
b. Du kannst dich you cail
verhlssen auf Maria.
yourself rely
on Mary
Other arguments, those that are eligible for nuclear stress assignment, gener� ally cannot occur in this extra posed position without a resumptive pronoun:
(65) Du kannst * (die) gruessen die Maria. you can
(her)
greet
the Mary
The important point to be made her is that precisely those constituent that do not receive nuclear stress-in general adverbials, but also some arguments can also occur to the right of the verb and be subordinated. Arguments that do not receive nuclear stress may be structurally 'higher' than those which do receive nuclear stress, that is, they are not in the complement position of the verb, just like adverbs. This remains to be investigated in detail.
4. 1.2.
Modifiers that Pattern like Arguments
There are also some apparent modifiers that receive nuclear stress. Consider internal locatives (66a). discussed in Krifka ( 1 984 . 14) , which receive nuclear stress. in contrast to external modifiers (66b):
(66)
am
a. Er hat
Schreibtisch gearbeitet.
he has at.the desk
worked
'He worked at the desk (carpenter).' b. Er hat
am
Schreibtisch gearbeitet.
he has at. the desk
worked
'He worked (sitting) at the desk.' Similar examples of internal locatives are discussed in and Maienborn (2001
•
20 1 ): (67)
a. paul hat in Stiefeln geduscht. Paul has in boots
showered
b. Die Spieler haben den T6rschtitzen auf den Schllltern tragen the players have
the scorer
on the shoulders carried
wollen. wanted These locatives show a categorically different pattern from the frame-setting locatives discussed in section (3.3). The solution proposed here is that they modify unrealized constituents in the complement domain of the main pred icate, i n other words. they are in fact part of the complement domain of the
357
Asymmetries in Prosodic Domain Formation predicate. As expected, they are not allowed to foHow the verb. just like true complements. The following can only be only be an external locative:
(68) Er hat gearbeitet, am He has worked,
Schreibtisch.
at the desk
The cases discussed in this section illustrate that the crucial relation relevant for prosody is structural height. The distinction functor vs.
complement
is used here only to determine which of two sisters counts as structurally 'higher' , based on the asymmetry of projection: functors are assumed to project.
s.
Deriving the Asymmetry
How does the syntax-phonology mapping work? The claim proposed here
is that prosody can be derived using exactly one type of syntactic informa tion: the information of which of two sister constituents projects,30 , by a
recursive mechanism foHowing the transformational cycle in SPE, similar to the proposals in Jacobs
( 1 99 1 , 1 992); Cinque (1 993); Arregi (2002) .3 1
I
will illustrate how the syntax-to-prosody mapping works based on the cases involving predicates and their arguments. Relative prominence can be represented by a metrical grid (Liberman,
1975; Libermann and Prince , 1 977). The prosodic foot structure imposed on the grid marks prosodic phrasing. I assume a version of the bracketed grid as outlined in Halle and Idsardi
( 1 995).
grid mark projection:
(69)
Higher grid marks are introduced by
Projection: Project all top-line grid-marks of a constituent to a new top grid-line, and foot them.
Projection as proposed here leaves relative prominence within the projected material intact, contrary to projection in the literature on the metrical grid, where only the head of a foot projects. This is a necessary modification of the theory, since the claim is that subordination is only negOtiated via syntax. 3OFoJlowing (Wagner. 2(02) . where evidence from p�es, comp�unds and derivatives is presented. Johnson (2002) posit" an asymmetric operation MERGE (essentially, the formation of an ordered pair), argues that focus projection and island conditions can be derived from propenies of recursive Merge. This proposal conbaSts with Chomsky (200 ] ) , who assumes a symmetric operation of set-merge. ] assume that the relation between sisters is asymmeuical at least at the interface to phonology. 31 For evidence against the algorithm based on depth of embedding in terms of the number of cycles or node-counting as in Cinque ( 1993) see Truckenbrodt (1993) and Arregi (2002).
358
Michael Wagner
(70) Examples of Projection a.
x
zu
(
x
(
x x
x
schweig e n
(
x
(
x
(
x x
Z u schweig e n
x
(x
b.
(x x x (x
x
x
(
x
(
x x
(x x x (x
z u vers u ch e nz u schweig e n
x
x
(
x
(
x x
zu versu ch e nz u schweig e n
The conventions about what to project when computing the relative promi nence between two sisters constitutes the phonology-syntax interface. Equal ize' is a different version of the stress equalization principle proposed in Halle and Vergnaud ( 1 987). 32 •
(7 1 ) Projection convention for < a.
a,
J3 > . where a projects:
'Sister-Matching' : if a precedes /3 . Project a and Project /3 .
b . 'Prosodic Subordination' : i f /3 precedes
a,
Project 13 ·
Th e two types o f cases that have to be distinguished are the following:
(72) Two Cases a.
Head Intial Structure: Sister Matching
/" J3
a
b. Head Final Structure: Subordination
( x(x
(x(x a {3
/�
fj
a
(x
(x(x {3 a
To illustrate how this works, consider first a right-branching structure. (73)
versprach zu versuchen zu schweigen
promised to try
to be.silent
'promised to try to be silent' a. First Step: Create y
x
(
(
x
x x
Z u schweig e n 32The reason why
(x x x (x
x
zu vers u ch e n
I adopt a different version relate to the pre-nuclear rhythmic pattern. The
approaches to stress in modeled
b. Second Step: Create fj
based on the
SPE, Libennann and Prince (1977). and Halle and Vergnaud ( 1 981) are
assumption that in the pre-nuclear domain. prominence is declining. In
terms of relative prominence: 2 3 4 5 I, whereas the present proposal derives a sequence of equal
stresses ahat are rhythmically organized. The output of the algorithm here is similar to the output
(1971) after stress leveling has applied. The last or nuclear accent is special in that it is not subject to rhythm. and is followed by a boundary.
of Libermann and Prince
359
Asymmetries in Prosodic Domain c. Third
Step: <
a,P >
(x (x
x x (x
x
Formation
x
d. Fourth Step: Create r
(
x
(
x
(
x x
x
z u vers u ch e nzu schwei g e n
e. Fifth Step: < r < a,p > > x ( x (
x
(x
(
x
(x
x (
x x x (x
( x (
x
versprach x x
x
x
(
x
(
x
x
versprachzuvers u ch e nz u schweig e n
The representation derived has a crucial property: Three accents. i.e. top level grid marks. are derived, which are essentially on a par. They coun t as the heads of three accentual domains. There are several lines in the grids that would seem superfluous. Why would the simpler version not suffice? (74) versprach zu versuchen zu schweigen 'promised to try to be silent' (
x
(x
x (
x
x x (x
x
x
(
x
(
x x
versprachzuvers u ch e nz u schweig e n
When two complex right-branching structures are put together e.g. in co ordination, the need for further structure becomes apparent. Otherwise. the expectation would be a sequence of accents on a par. ,
(75) Two complex Right-Branching Structures 'promised to try to be silent
and asked to allow to whisper'
(
x
x
x
x
x
(
x
x
x
(x
x
x
(
x
(x
x
(x
x
x
(
x
(x
( x ( x x x( x
x x (x
x (
x x x (x
( x
x
(
x x
(x
(x
x
x x
versprachzuvers u ch e nZ !l schweigenundbatzuerlaub e n z uft ii stern
The additional grouping in (75) is ne�essary. since the pitch level i.s reset at the break between the two predicate sequences. Within each predicate sequence downstep (or declination) between the three accents can be observed. The relative pitch level of the six accents in the structure can be approximated by looking at the left brackets: the highest left bracket in the column represents the relative pitch level. The grouping arises through the brackets that delimit feet at the relevant grid line. For discussion of the phonetic realization of
360
:
'
Michael Wagner hierarchical structures see Ladd ( 1 986, 1 988). linearizations: (76)
1
-<
Consider now two d i ffe re nt
3 -< 2
versprach zu schweigen zu versuchen promised to be silent
to try
'promised to try to be.silent' (
x
(
x
(
x
(
x
x (x
(
x x
x (
(x x x(x
x
versprachz u schweig e nz u vers u ch e n
The next example illustrates the case of a completely inverted structure. (77) 3 -< 2 -< 1 zu schweigen zu versuchen versprach
to be silent
to try
promised
'to promise to try to be silent x
(x
(
x
(
x
x
x
(x
( x
x(x
x x ( x
3
2
J
z u schweig e nz u vers u ch e n v ersprach
The recursive projection mechanism outlined here derives the correct promi nence relations between constituents. The foot structure imposed on the grid marks models intuitions about prosodic domains, serves to mark domains for down-stepping and reset, and captures mismatches in constituency between syntax and prosody. 33 The linear order effect was stipulated here: Ultimately, the very mechanism that fixes linear order should be linked to the prosodic differences. I will return to this problem in the conclusion.34 6.
Conclusion and Outlook
The first pan of the paper presented an asymmetry in prosody: if a functor
precedes an argument it is phrased separately; if itfollows an element from 33 For arguments against a mechanism based on XP-alignment see Wagner
34
Whether or not phonological and synlaCtic derivations apply cyclically
(2004). as was suggested
in
Bierwisch ( 1 968). Bresnan ( 1 97 1 ) . and Adger (2003) is nOl apparenl from the data discussed here. and requires further research.
361
Asymmetries in Prosodic Domain Formation
The asymmetry was infinitival and nominal complements. It was then generalized to cases of modification, again using the asymmetry between functor and argument, si m i lar to approaches to prosody in categorial grammar (Steedman , 200 1). A look at secondary predication and modification provided evidence its argument or the entire argument, it is subordinated.
first illustrated based on predicates and
that the generalization is really syntactic in nature and has to be stated based
on
the asymmetry of projection. rather than on the asymmetry based on the
functor-argument relation. Just like linear order. prosody is by and large deter mined by the architecture of syntactic trees and syntactic relations.
Eurythmic
effects further complicate the picture and by deleting prosodic structure they create prosodic constituents relations alone. 3S
that cannot be
determined based on syntactic
important difference to alternative ways of determined nuclear prosodic subordination observed after the nuclear stress is not an effect of placing the nuclear stress on a preceding constituent, but conversely it is cyclic subordination that causes that nuclear stress to be perceived on that last non-subordinated constituent. A question not touched upon in this paper so far is how the different linear orders between functors and their complements come about. Under an asymmetric view of syntax (e.g. Haider. 1993; Kayne , 1 994) one order is basic and the other order is derived. Remember that predicates that precede their complement receive an independent prosodic domai n and those that follow their complement or part of it are affixed to the prosodic domain of the complement, unless the predicate is focused, in which case it does receive an accent: One
stress
is
that the
,
(78)
a.
[ Sie tanzte] [ einen Tango.] she danced a.acc Tango
b. [ Sie hat ] [ einen Tango getanzt.] she has
c.
a.acc Tango danced
[ Sie bat ] [ einen Tang o she has
a.ace Tango
'She danced
a tango.
[ gettinzt ]Fokus.]
danced
•
Now consider the following pr�sodic pattern ofVP remnant fronting iIi Ger man:
3S The general idea of tying phonological domains more closely to syntactic structure is widely pursued at the moment (Legate . 200 1 ; Marvin . 2002; Arregi . 2002; Adger. 2003; Kahnemuyipour. 2003). Elaborating the differences and parallels would go beyond the scope of this paper.
362
Michael Wagner
(7 9)
a.
[ Peter
[ hat
Peter
der Maria ] [ einen Kuss gegeben
has the Mary
J
a kiss given
b. [ r Einen Kuss gegeben ] hat Peter der Maria. ] a
c.
has Peter the Mary
kiss given
[ [ Einen Kuss gegeben ] hat Peter der [ Marifa]Focus ] a kiss given
has Peter the
Mary
The fronted VP receives an accentual domain, the remaining part of the sen tence is subordinated-unless there is a focused constituent in the remnant material. This, of course, is exactly the asymmetry observed earlier for pred ication and modification. It seems tempting then to analyze the reorderings at stake as cases of remnant movement. A comprehensive syntactic account of
the derivation of linear order in tandem with prosody remains to be developed .
References Adger, David. 2003. Phonology and phasal syntax . Paper presented at GLOW.
1 997 . Aderb pLacement: A case study in antisymmerric
Alexiadou, Artemis.
syntax.
Amsterdam: Benjarnins.
Altmann. Hans, ed.
1988. Intonations/orschungen. Tlibingen:
Niemeyer.
Anagnastopou lou, Elena, and Artemis Alexiadou. 200 1 . The subject in situ generalization, and the role of case in driving computations.
Inquiry 32: 1 93-23 I . Arregi. Karlos. 2002. Focus on
Linguistic
Basque movement. Doctoral Dissertation,
MIT. Cambridge, Ma. Barbiers, Sjef.
tation.
1995. The syntax o/interpretation , volume 14 of HIL Disser
Holland Academic Graphics.
Bech. Gunnar.
1 955157. Studien iiber das deutsche Verbum Infinitum. TIibin
gen: Niemeyer.
1 968. Two cricitcal problems in accent rules. Linguistics 4: 173- 1 78. Bolinger, D. 1972. Accent is predictable (if you are a mind reader). Language 48:633-644. Bresnan, Joan. 197 1 . On sentence stress and syntactic transformations. In Contributions to generative phonology , ed. Michael K. Brame, 73-107.
Bierwisch, Manfred.
University of Texas Press. Cho msky, Noam. ] 97 I . Deep structure, surface structure, and semantic inter pretation . In Semantics: An interdisciplinary reader in philosophy. linguis
tics. and psychoLogy. ed.
D.O. Steinberg and L.A. Jakobovits. Cambridge :
Cambridge University Press. Chomsky, Noam.
1 995. The minimalist program.
Cambridge: MIT Press.
363
Asymmetries in Prosodic Domain Formation
Chom sky. Noam. 200 1 . Derivation by phase. In Ken Hale. a life in language ed. Michael Kenstowicz, 1-52. MIT Press. Chomsky. Noam, and Morris Halle. 1968. The soundpattern ofEnglish. New York: Harper & Row. Chomsky. Noam, Morris HaBe, and Fred Lukoff. 1 957. On accent and junc ture in English. In For Roman Jacobson . ed. Morris Halle. Horace Lunt. and Hugh MacLean. The Hague: Mouton. Cinque. Guglielmo. 1993 . A null-theory of phrase and compound stress. Linguistic Inquiry 24:239-298. Cinque. Guglielmo. 1999. Adverbs andfunctional heads: A cross-linguistic perspective. Oxford: Oxford University Press. Diesing. Molly. 1 992. Indefinites. Cambridge. Ma.: MIT Press. Drosdowski, Gunther, ed. 1 995. Duden Grammatik der deutschen Gegen wartsprache. Mannheim: Duden. Evers, Arnold. 200 1 . Verb clusters and cluster creepers. Utrecht. Frey, Werner. 2003. Syntactic conditions on adjunct classes. In Lang et al. (2003) , 1 63-209. Ghini, Mirco. 1 993. q)-formation in Italian: A new proposal. In Toronto working papers in linguistic , ed. Kerry Dick, volume 4. 41-78. University of Toronto. Goldsmith, John A., ed. 1 995. The handbook ofphonological theory. London: Blackwell. Gussenhoven. Carlos. J 984. On the grammar and semantics of sentence accents. Dordrecht: Foris. Gussenhoven. Carlos. 199 1 . The English rythmn rule as an accent assignment rule. Phonology 8 : 1 -35. Haider. Hubert. 1 993. Deutsche syntax · generativ. TLibingen: Gunter Narr. Haider, Hubert. 1 994. Fakultativ koharente infinitivkonstruktionen im Deutschen. In Zur satzwertigkeit von Infinitiven und Small Clauses. ed. Anita Steube and Gerhild Zybatow. Thbingen: Niemeyer. Haider, Hubert. 2000. Adverb placement convergence of structure and linking. Theoretical Linguistics 26:95-1 34. Halle, MOrris, and William Idsardi. 1995. General properties of stress and metrical structure. In Goldsmith ( 1 995). 403-443. Halle. Morris. and Jean-Roger Vergnaud. 1 987. An essary on stress. Cam bridge: MIT Press. Hayes, Bruce. and Aditi Lahiri. 1 99 1 . Bengali intonational phonology. Nat· ural Language and Linguistic Theory 9:47-96. Heim. Irene, and Angelika Kratzer. 1998. Semantics in generative grammar. Oxford: Blackwell. Hoskins, Steve. 1996. A phonetic study of focus in instransitive verb sen· tences. In Proceedings ofthe ICSLP 96. . •
•
.
364
Michael Wagner Idsardi. William. 1 992. The compulation of prososdy. Doctoral Dissertation. MIT. Ishihara, Shinishiro. 2001 . Stress, focus, and scrambling in japanese. In A few from building e39. MITWPL. Ishihara. Shinishiro. 2003. Intonation and interface conditions. Doctoral Dissertation, MIT. Jackendoff. Ray S. 1972. Semantic interpretation in generative grammar. Cambridge, Ma.: MIT Press. Jacobs, Joachim. 1 99 1 . Focus ambiguities. Journal oj Semantics 8: 1-36. Jacobs, Joachim. 1 992. Neutral stress and the position of heads. In Injorma tionsstrukur und Grammatik ed. Joachim Jacobs, Linguistische Berichte Sonderheft 4, 220-244. OpJaden: Westdeutscher Verlag. Jacobs, Joachim. 1999. Informational autonomy. In Focus. linguistic. cogni tive. and computational perspectives, ed. Peter Bosch, 56-8 1 . CUP. Jacobs. Joachim. 200 1 . The dimension of topic-comment. Linguistics •
39:641-68 1 .
Johnson, Kyle. 2002. Towards an etiology of adjunct islands. Umass Amherst. Kahnemuyipour, Arsalan. 2003. Toward a phase-based theory of phrasal stress. Workshop on Phases. MIT 2003. Kayne. Richard. 1994. The antisymmetry oj syntax. Cambridge, Ma.: MIT Press. Kenesei. Istvan, and Irene Vogel. 1995. Focus and phonological structure. University of Szeged, University of Delaware. Kratzer, Angelika. 1995a. Severing the external argument from its verb. In Phrase structure in the lexicon , ed. Johan Ryck and Laurie Zaring. 1 091 37 . Dordrecht: Kluwer Academic Publishers. Kratzer, Angelika. 1 995b. Stage-level and individual-level predicates. In The generic book , ed. G.N. Carlson and F.J. Pelletier, 1 25- 1 75. Chicago: Chicago University Press. Krifka, Manfred. 1 984. Fokus, Topik. syntaktische Struktur und semantische Interpretation. Universitit Ttibingen. Ladd, Robert. 1 986. Intonational phrasing: the case for recursive prosodic structure. Phonology Yearbook 3:3 1 1-340. Ladd, Robert. 1988. Declination and 'reset' and the hierarchical organziation of utterances. JASA 84. Lakoff, George. 1972. The global nature of the nuclear stress rule. Language 48:285-303.
Lang, Ewald, Claudia Maienborn, and Cathrine Fabricius-Hansen, ed. 2003. Modifying adjuncts. Berlin: Mouton de Gruyter. Larson. Richard K. 2004. Sentence final adverbs and scope. In Proceedings ofNELS 34 , ed. Keir Moulton and Matt Wolff. Stony Brook. Legate, Julie Anne. 200 I . Some interface properties of the phase. Ms. MIT.
365
Asymmetries in Prosodic Domain Formation
Liberman, M. 1975. The intonational system of English. Doctoral Disserta tion, MIT.
Libermann, M., and A. Prince. 197 7. On stress and linguistic rhythm. Lin
guistic Inquiry 8:249-336. Maienborn, Claudia. 200 1. On the pOSition and interp retati on of locative
modifiers. Natural Language Semantics 9: 1 9 1 -240. Marantz, A lec . 1 984. On the nature of grammatical relations. Cambridge,
Ma. : MIT Press.
Marvin, Tatjana. 2002. Topics in the stress and syntax of words. Doctoral
Dissertation, Massachusetts Institute of Technology. Munn, Alan. 2000. Three types of coordination asymmetries. In Ellipsis
in conjunction . ed . Kerstin Schwabe and Ning Zhang, 1-22. Tlibingen :
Niemeyer. Neeleman. Ad, and Tania Reinhart. 1998. Scrambling and the PF-interface. In The projection of arguments, ed. Miriam Butt and Wilhelm Geuder, 309-353. CSLI. Newman, S.S. 1 946. On the stress system of English. Word 2: 1 7 1- 1 87. Nilsen, 0ystein. 2002. Eliminating positions. Doctoral Dissertation, Uni ver sity of Utrecht. Pesetsky, David . 1 995. Zero syntax. experiencers and cascades. Cambridge Ma: MIT Press. Phillips, Colin. 1 996. Order and structure. Doctoral Dissertation, MIT. Pylkldin en Liina. 2002. Introducing arguments . Doctoral Dissertation, MIT. Schmerling, S. F. 1 976. Aspects ofEnglish sentence stress. Austi n : University of Texas. Schwarzschild, Roger. 1 999. Givenness, AVOIDF and other constraints on the placement of accent. Natural lAnguage Semantics 7: 141-177. Selkirk, Elizabeth O. 1 984. Phonology and syntax. the relation between sound and structure. Cambridge, Ma.: MIT Press. Selkirk, Elizabeth O. 1995. Sentenc e prosody: Intonation, stress, and phras ing. In Goldsmith { I 995), 550-569. Soh, Hooi Ling . 200 1 . The syntax and semantics of phonological phrasing in Shanghai and Hokkien. Journal ofEastAsian Linguistics 10:37-80. Starke. Michal. 2001 . On the inexistence of specifiers and the nature of heads. NYU. Stechow, Amim von . 1996. The different readings of 'again' : a structural account. Journal ofSemantics 1 3 . Steedman. Mark. 200 1. The syntactic process. Cambridge, Ma.: MIT Press. Truckenbrodt, Hubert. 1 993. Syntax vs . phonology: Which gets the stress right? Ms. MIT. Truckenbrodt. Hubert. 1 995. Phonological phrases: Their relation to syntax, focus, and prominence. Doctoral Dissertation, MIT, Cambridge, Mass.
,
,
·
366
Michael Wagner
Uhmann, Susanne.
199 1 . Fokusphonoiogie. Eine Analyse deutscher Into
nationskonturen im Rahmen der nicht-linearen Phono[ogie. Tlibingen: Niemeyer. Wagner, Michael.
2002.
Configurational stress in derivatives. compounds,
and phrases. Paper presented at the Workshop on Pertinacity, University of Konstanz. Wagner. Michael.
2004.
Prosody as a diagonalization of syntax. evidence
from complex predicates. In Proceedings ofNELS
34 at Stony Brook . ed.
Mathew Wolf and Keir Moulton. Wurmbrand, Susi.
2003.
Verb-clusters in West-Germanic: The emipirical
domain. In The verb-cluster Sprachbund: lkrb clusters in Germanic and
Hungarian . ed. Katalin lohn Benhamins. Zubizarreta, M. L.
E.
Kiss and Henk van Riemsdijk. Amsterdam:
1998. Prosody, focus. and word order. Cambridge, Ma.:
MIT Press.
Michael Wagner MIT Department of Linguistics and Philosophy
77 Massachusetts Avenue, 32-0962 Cambridge, MA
02 1 39-4307
chae/@mit.edu
367