This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
]
The LP constraint in (74) states that a subject must precede the category which selects it as a subject. Like the LP constraint in (72), the LP constraint in (74) is another instantiation of the head-final property of Korean. In (54d,e) John-ul is the subject of the VP New York-ey isstako 'to stay in New York' but does not precede the VP. Thus, these sentences violate (74). It is important to note that the LP constraint in (74) does not apply in the case of the control construction. The constraint in (74) applies only when the whole SYNSEM value of the controller and the subject of the VP
Argument Composition, Long-Distance Scrambling
193
complement is the same. As shown in (68b), however, the controller and the understood subject of the verbal complement of the control verb are just coindexed, and the whole SYNSEM value cannot be the same because of the case mismatch. Hence, in the control construction, (74) does not apply, and the controller is allowed to linearly follow the verbal complement, Finally, the subject of the raising verb can occur between the complement VP and the raising verb, as shown in (54c). This is a consequence of the flat analysis shown in (73).17 3.2.1.3. Accusative ha-causative construction. As discussed in section 3.1.4, the scrambling possibilities of the accusative ha -causative construction are not the same as those of the control or raising construction. On the one hand, that variant of ha shares properties of a raising verb in that the controller cannot occur after the governed verb (e.g., (63)), but on the other hand, it also behaves like a control verb in that an argument within the VP complement can scramble with arguments outside the VP (e.g., (64)). On our approach, all the properties of the accusative ha -causative fall out naturally if we assume that this type of ha has the lexical description in (75), which is a revision of (60b):
(75)
This descrition resembles the output of the lexical rule in (67). In our account, however, there is no lexical entry corresponding to an input for (67), and therefore (75) must simply be listed separately. The lexical entry in (75) licenses the structure in (76b) for the sentence in (76a) (= (59b)): (76) a. Mary-ka John-ul ku chayk-ul ilkkey hayssta. M-Nom J-Acc the book-Ace read caused 'Mary caused John to read the book.'
194
Chan Chung
In (76b), John-ul and the subject of the verb ilkkey 'read' have the same SYNSEM value, represented as SNP. Thus, by the LP constraint in (74), John-ul must precede ilkkey. The sentence in (63) is unacceptable since it violates this LP constraint. Also in (76b), ku chayk-ul 'the book' must precede ilkkey because of the LP constraint in (72), which states that a complement must precede the category which subcategorizes for it. Our analysis correctly predicts that all permutations of three NPS and a complement verb are possible as long as these two LP constraints are observed.
3.2.2. ANALYSES OF S-COMPLEMENT CONSTRUCTIONS In this section, we will extend the idea of scrambling via argument composition further, proposing that scrambling out of an s-complement also can be accounted for by the same mechanism. In Korean, arguments of an embedded clause also can scramble with arguments of the main verb, as shown in (77) and (78):18, 19 (77) a. Nay-ka [s Mary-ka ku chayk-ul ilkesstako] sayngkakhayssta. I-Nom M-Nom the book-Ace read thought 'I thought Mary read the book.' b. Ku chayk-ul nay-ka Mary-ka ilkesstako sayngkakhayssta. the book-Ace I-Nom M-Nom read thought (78) a.
Mary-ka motun salam-hanthey [s John-i ku mwuncey-lul M-Nom all people-Dat J-Nomthe problem-Ace haykyelhaysstako} malhayssta. resolved told 'Mary told all the people that John resolved the problem.' b. (?)Ku mwuncey-lul, Mary-ka, motwun salam-hanthey, John-i the problem-Ace M-Nom all people-Dat J-Nom haykyelhaysstako malhayssta. resolved told
Argument Composition, Long-Distance Scrambling
195
c. ? Mary-ka, ku mwuncey-lul, motwun salam-hanthey, John-i M-Nom the problem-Ace all people-Dat J-Nom haykyelhaysstako malhayssta. resolved told In our analysis, arguments of the head of the embedded clause can be attracted to the COMPS lists of the matrix verb. To this end, we posit the following lexical rule which is similar to that in (67):
(79)
The input description in (79) takes a sentence as its complement. When the element represented by GD is the empty list, the input matches a lexical entry that takes only a sentence as its complement (e.g., sayngkakhayssta 'thought' in (77)). When H] is realized as {NP[dat]}, the input entry matches a lexical entry that takes both an Np[dat] and an s as its complement (e.g., malhayssta 'told' in (78)). In contrast, the output description of (79) takes a verb as its complement in addition to S, and the subject and complement of the complement verb (represented by NP[nom] 0 and GO) are attracted and appended to the COMPS lists of the matrix verb. See section 4.1 for motivation of the idea to let the embedded subject be attracted to the matrix verbs' COMPS list, rather than to the SUBJ list. Instances of scrambling as in (77b) and (78b,c) are licensed by output entries of the lexical rule. For example, (80a) is the structure of (77b)
196
Chan Chung
licensed by (80b):20 (80) a.
Note that the rule in (79) allows arbitrarily many arguments of the embedded clause to scramble out. This is in fact possible in Korean, as shown below:21 (81) a. Nay-ka [s Mary-ka ku chayk-ul John-hanthey pilyecwuesstako] I-Nom M-Nom the book-Ace J-Dat lent sayngkakhayssta. thought 'I thought that Mary lent the book to John.' b. Ku chayk-ul John-hanthey nay-ka Mary-ka pilyecwuesstako the book-Ace J-Dat I-Nom M-Nom lent sayngkakhayssta. thought c. John-hanthey ku chayk-ul nay-ka Mary-ka pilyecwuesstako J-Dat the book-Ace I-Nom M-Nom lent sayngkakhayssta. thought
3.23. ADJUNCT SCRAMBLING In previous sections, we focused only on scrambled arguments. In Korean, however, an adjunct that modifies a verbal expression can also scramble with arguments, as shown in (82b,c).
Argument Composition, Long-Distance Scrambling
197
(82) a. Ku kongwen-eyse Mary-ka John-ul mannassta. the park-at M-Nom J-Acc met 'Mary met John at the park.' b. Mary-ka ku kongwen-eyse John-ul mannassta. M-Nom the park-at J-Acc met c. Mary-ka John-ul ku kongwen-eyse mannassta. M-Nom J-Acc the park-at met In this section, we will discuss how to handle scrambled adjuncts. We have already assumed a flat structure for a sentence. Hence it is natural to posit that an adjunct is also licensed by the flat structure. In order to account for scrambling among complements and adjuncts in German, Kasper (1994) proposes to analyze adjuncts as sisters to complements, and hence to the head verb: i.e., an adjunct is contained in a flat structure. In this flat structure analysis, we need the following HeadSubject-Complement-Adjunct schema for Korean:
(83)
The schema in (83) states that zero or more adjuncts can appear in a clause and that if at least one adjunct is present, it is a sister to the arguments (subject and complements) and to the verb which subcategorizes for the arguments.22 A special property of (83) is that the value of an adjunct modifying a predicate is cospecified in the valence value of the predicate, i.e., as generally assumed, an adjunct still selects what it modifies, but the list of the modifiers is also specified as the value of the new valence feature ADJT on the modified predicate.23 The proposal to make adjuncts part of the valence specification is justified in that there seems to be no syntactic difference between arguments and adjuncts in Korean. Both can scramble within a clause, scramble out of an s- or vp-complement, be afterthought expressions in a simplex clause, and be extracted out of an s- or vp-complement to be afterhtought expressions. However, an adjunct differs from an argument in its semantics: i.e., an adjunct does not have any specific semantic role in a predicate which is modified by the adjunct, and it is a
198
Chan Chung
semantic head in that the CONTENT value of an adjunct is identical to that of the mother node. These facts will be discussed one by one shortly. In this section, we focus on the case where an adjunct scrambles out of a VP or s complement. As pointed out in Saito (1985), an adjunct generally can scramble out of a complement clause or a VP complement, as shown in (84) and (85):
(84) a.
Mary-ka [s John-i nayil kkaci ku swukcey-lul M-Nom J-Nom tomorrow by the homework-Ace machyeya-hantako] malhayssta. finish-must said/told 'Mary said that John had to finish the homework by tomorrow.' b. (?)Nayil kkaci Mary-ka John-i ku swukcey-lul tomorrow by M-Nom J-Nom the homework-Ace machyeya-hantako malhayssta. finish-must said/told
(85) a.
Mary-ka John-hanthey [VP nayil kkaci ku swukcey-lul M-Nom J-Dat tomorrow by the homework-Ace machilako] malhayssta. finish said/told 'Mary told John to finish the homework by tomorrow.' b. (?)Nayil kkaci Mary-ka John-hanthey tomorrow by M-Nom J-Dat ku swukcey-lul machilako malhayssta. the homework-Ace finish said/told c. (?)Mary-ka nayil kkaci John-hanthey M-Nom tomorrow by J-Dat ku swukcey-lul machilako malhayssta. the homework-Ace finish said/told
Here, (84b) and (85b,c) are slightly marginal compared with the (a) counterparts, but they are quite acceptable. On our approach, this fact is accounted for by the flat structure analyses, under the assumption that an adjunct of a complement verb can also be attracted to the ADJT list of a matrix verb when arguments are attracted. To this end, we need to revise the argument composition lexical rules in (67) and (79). For example, the
Argument Composition, Long-Distance Scrambling
199
rule in (67) is revised into (86):
(86)
(Here a = 1 or 3.) The rule in (86) differs from (67) only in that, in the former, the adjunct (ADJT) list of a complement verb is attracted to the ADJT list of a matrix verb as well. On this approach, (84b) has the structure in (87a):
(87)
a.
200
Chan Chung
In (87), the adjunct PP nayil kkaci 'by tomorrow' takes the complement QDv as its MOD value. Note that the schema in (83) is nonspecific as to whether the modified category is the head verb or a complement verb. The structure in (87) instantiates the latter case, where the modified category is a complement verb. In this case, if we directly apply the semantic principle of Pollard and Sag (1994) given in (88), we do not get the correct interpretation. (88) In a headed phrase, the CONTENT value is token-identical to that of the adjunct daughter if the DTRS value is of sort head-adjunctstruc, and to that of the head daughter otherwise. (Pollard and Sag, 1994:56) The daughters of the s category in (87) constitute the head-subj-compadjunct-structure which contains an adjunct daughter. Then, by (88), the semantic content of the whole sentence in (87) must be identical to the semantic content of the widest-scoping adjunct daughter, which can be roughly represented as follows: by-tomorrow' (necessarily' (finish' (Mary', the-homework'))). However, this is obviously not the correct interpretation of the whole sentence. This problem can be solved by a slight revision of (88) to (89): (89) In a headed phrase, the CONTENT value is token-identical to that of an adjunct daughter if the head daughter is the value of MOD of the adjunct daughter, and to that of the head daughter otherwise. In (87), the adjunct daughter does not take the head daughter as its MOD value, and hence the CONTENT of the whole sentence is the same as that of the head daughter, which is roughly represented as follows: said' (Mary', by-tomorrow' (necessarily' (finish' (Mary', the-homework')))). On our approach, the argument composition lexical rules play a crucial role in the determination of the correct meaning mentioned above. Here the important point is that the value of the CONTENT of the input entry is token-identical to that of the output entry: i.e., only the valence value of the input entry differs from that of the output value. (90) shows the structure of (84a) which is licensed by the lexical entry to which the argument composition lexical rule does not apply. Here the structures of [Tjv:[3] and 0V: GO are shown in (91) and (92), respectively.24
Argument Composition, Long-Distance Scrambling
(90)
(91)
(92)
201
202
Chan Chung
If we apply the argument composition lexical rule to (92), we obtain the following output entry:
The lexical description in (93) is the same as the head verb mv in (87b) and licenses the flat structure in (87a), whereas the one in (92) is the same as the matrix head verb EV in (90) and licenses the hierarchical structure. However, the CONTENT value of both is identical so that the structures in (87a) and (90) have exactly the same interpretation, which can roughly be represented by the CONTENT value QD in (92). A potential problem for our analysis is asymmetrical distribution of adjuncts in s- and vp-complement structures. The flat structure in (87) predicts that an adjunct which modifies the matrix head verb may occur at any place in a sentence. This prediction is borne out in the case of the vp-complement construction. The sentences in (94) show that the PP ilcwuil ceney 'one week ago' which modifies the matrix verb malhayssta 'told' can be placed rather freely: (94) a.
Mary-ka ilcwuil ceney John-hanthey ku project-lul M-Nom one week ago J-Dat the project-Ace kkuthnaylako malhayssta. finish said/told 'One week ago Mary told John to finish the project.' b. Mary-ka John-hanthey ilcwuil ceney ku project-lul M-Nom J-Dat one week ago the project-Ace kkuthnaylako malhayssta. finish said/told c. (?)Mary-ka John-hanthey ku project-lul ilcwuil ceney M-Nom J-Dat the project-Ace one week ago kkuthnaylako malhayssta. finish said/told d. Mary-ka John-hanthey ku project-lul kkuthnaylako M-Nom J-Dat the project-Ace finish ilcwuil ceney malhayssta. one week ago said/told
Argument Composition, Long-Distance Scrambling
203
Even though (94c) is slightly marginal compared with the others, it is quite acceptable when a short pause is given after the adjunct, or stress is given to the adjunct.25 As shown in (95), however, in the s-complement construction, an adjunct modifying a matrix verb never occurs between the embedded subject and its predicate verb: (95) a. Mary-ka ilcwuil ceney John-i ku project-lul M-Nom one week ago J-Nom the project-Ace kkuthnayya-hantako malhayssta. finish-must said/told 'One week ago Mary said that John had to finish the project.' b. *Mary-ka John-i ilcwuil ceney ku project-lul M-Nom J-Nom one week ago the project-Ace kkuthnayya-hantako malhayssta. finish-must said/told c. *Mary-ka John-i ku project-lul ilcwuil ceney M-Nom J-Nom the project-Ace one week ago kkuthnayya-hantako malhayssta. finish-must said/told d. Mary-ka John-i ku project-lul kkuthnayya-hantako M-Nom J-Non the project-Ace finish-must ilcwuil ceney malhayssta. one week ago said/told The sentences in (95b,c) cannot be ameliorated by any pause or stress and present a sharp contrast with (94c). Therefore, the prediction seems not to be borne out that an adjunct which modifies the matrix head verb may occur at any place in a sentence. In our analysis, however, sentences such as (95b,c) can be ruled out by the interpretive principle mentioned in note 21. The principle is repeated here as (96): (96) Interpretive Principle Suppose (i) that Y is an NP[NOM], (ii) that X is the first verb following Y, and (iii) that Z is any constituent which occurs between Y and X. Then Z cannot be a "semantic dependent (semantic argument or functor)" of a verb superordinate to X. For example, the structure of semantic dependency for (95b) is as follows:
(97)
204
Chan Chung
Thus, (95b,c) are ruled out by (96) and do not present counterexamples to the flat structure analysis. Now, let us consider some matters that are relevant to LP constraints on adjuncts. First, an adjunct cannot occur after a category that is modified by the adjunct. The relevant examples are in (98): (98)
*Mary-ka [s John-i ku swukcey-lul M-Nom J-Nom the homework-Ace machyeya-hantako nayil kkaci] malhayssta. finish-must tomorrow by said/told 'Mary said that John had to finish the homework by tomorrow.'
To account for this fact, we need to assume another LP constraint in (99) which is another instance of head-finality:26 (99) SYNSEMIH < [ADJT < . . . CD ... >] The constraint in (99) simply states that an adjunct precedes a category which is modified by the adjunct. Another matter that deserves discussion is scrambling of an adverb phrase (ADVP) which is derived from a stative verb ("true adjunct" in terms of Saito, 1985).27 This kind of adjunct scrambles out of the complement clause or VP only with great difficulty, if at all, as shown in (100): (100) a.
Mary-nun [s John-i yelsimhi enehak-ul M-Top J-Nom hard linguistics-Ace kongpwuhaysstako] sayngkakhanta. studied think 'Mary thinks John studied linguistics hard.' b. ??/*Yelsimhi Mary-nun John-i enehak-ul kongpwuhaysstako hard M-Top J-Nom linguistics-Ace studied sayngkakhanta. think
(101) a.
Mary-ka John-hanthey [VP yelsimhi enehak-ul M-Nom J-Dat hard linguistics-Ace kongpwuhalako] malhayssta. study said/told 'Mary told John to study linguistics hard.' b. llYelsimhi Mary-ka John-hanthey enehak-ul kongpwuhalako hard M-Nom J-Dat linguistics-Ace study malhayssta. said/told
Argument Composition, Long-Distance Scrambling
205
However, considering the examples in (102), we cannot simply assume that an ADVP does not scramble out of an s or VP at all: (102) a. (?)Enehak-ul yelsimhi Mary-nun John-i kongpwuhaysstako linguistics-Ace hard M-Top J-Nom studied sayngkakhanta. think 'Mary thinks John studied linguistics hard.' b. Enehak-ul yelsimhi Mary-ka John-hanthey kongpwuhalako linguistics-Ace hard M-Nom J-to study malhayssta. said/told 'Mary told John to study linguistics hard.' In (102), the ADVP scrambles out of the s- or vp-complement together with the object NP of the embedded verb. Such examples are notably more acceptable than (l00b) or (l0lb). Thus, in our analysis, it seems that the unacceptable (b) sentences in (100) and (101) are licensed by syntax but ruled out by certain processing factors whose exact nature is presently unknown to us. That is, ADVPS are considered the same as other kinds of adjuncts and there is no specific syntactic constraint on ADVP scrambling. In this section, we showed how the account of verbal complexes can be extended to the account of so-called long-distance scrambling. The crucial mechanisms in our analysis of long-distance scrambling are argument composition and lexical rules. Kiss (1994) extends the argument attraction mechanism in Hinrichs and Nakazawa (1989, 1994) to account for German modal, control, and raising verb constructions. The suggestions in this section share some of the same basic ideas. However, our proposals differ from his in several respects. The main differences are as follows: (i) arguments of the complement verb which are attracted to the nonauxiliary matrix verb are all sisters to the matrix and complement verbs (cf. Bouma and van Noord, 1998, for the same suggestion on the Germanic verb clusters); (ii) different scrambling possibilities among different vp-complement constructions are accounted for by the applicability of an argument composition lexical rule, i.e., whether the matrix verb has only the input entry of the lexical rule or only the output entry, or both of them; (iii) scrambling out of a complement clause is also accounted for via argument compostion; and (iv) the argument composition analysis extends to scrambling of adjuncts. In the next section, we discuss what theoretical predictions can be made with our approach to long-distance scrambling.
206
Chan Chung
4. THEORETICAL PREDICTIONS The goal of this section is to show what theoretical predictions our theory can make. To this end, we discuss some phenomena that have been raised in previous studies of scrambling: i.e., long-distance passivization and weak crossover effects. 4.1. Long-Distance Passivization Korean exhibits so-called long-distance passivization phenomena, where an object of an embedded verb is passivized and appears as the matrix subject, while only the matrix verb bears passive morphology. Examples are given in (103) and (104) (see Kiss, 1992, Nagai, 1991, and Yatabe, 1993, for discussions on long-distance passivization in German and Japanese): (103) a. Nay-ka Mary-hanthey ku cengchayk-ul sihaynghalako cisihayssta. I-Nom M-Dat the policy-Ace carry out ordered 'I ordered Mary to carry out the policy.' b. Ku cengchayk-i naey-uyhayse Mary-hanthey sihaynghalako the policy-Nom I-by M-Dat carry out cisi-toy-essta. order-Pass-Past (Literally) 'The policy was ordered by me for Mary to carry out.' ('It was ordered by me for Mary to carry out the policy.') (104) a. Motun salam-i hyun cengpwu-ka ku cengchayk-ul all people-NoM current government-Nom the policy-Ace sihaynghalilako sayngkakhayssta. will carry out thought 'All people thought the government would carry out the policy.' b. Ku cengchayk-i motun salamey-uyhayse hyun cengpwu-ka the policy-Nom all people-by current government-Nom sihaynghalilako sayngkak-toy-essta. will carry out think-Pass-Past (Literally) 'The policy was believed by all people that the government would carry out.' ('It was believed by all people that the government would carry out the policy.') In (103b), the object NP ku cengchayk-ul 'the policy' within the VP complement is passivized, while the passive morpheme -toy is realized on the matrix verb, e.g., cisi-toy-essta 'was ordered.' The same long-distance passive can also affect arguments of the s-complement as shown in (104b). This kind of passivization is problematic for Phrase Structure Grammar in general, because under standard assumptions an object can be passivized only when it is an argument of a verb on which the passive morpheme is realized.
Argument Composition, Long-Distance Scrambling
207
Note that the subject of the embedded clause also can be the subject of the long-distance passive, as shown in (105): (105) Hyun cengpwu-ka motun salam-eyuyhayse current government-NoM all people-by ku cengchayk-ul sihaynghalilako sayngkak-toy-essta. the policy-Ace will carry out think-Pass-Past (Literally) The current government was thought by all people to carry out the policy.' Long-distance passivization phenomena are naturally accounted for by our theory of long-distance scrambling. According to that proposal, the object NP ku cengchayk-ul 'the policy' of the embedded verb in (103b) and (104b) can be attracted to the COMPS list of the matrix verb. Then the attracted NP can be promoted to subject by the passive lexical rule in (106), while the passivization morpheme is realized on the matrix verb: (106)
The passive lexical rule in (106) takes as input a base form verb (e.g., cisiha 'order' in (103a)) and gives as output a passive form verb (e.g., cisitoy 'be ordered' in (103b)). The subject of the input is removed from the SUBJ list, and its index is reassigned to the PP complement of the output. In addition, the nondative (i.e., nominative or structural) NP complement of the input is placed in the SUBJ list of the output entry. For instance, the structure of (103a) can be analyzed as a flat structure as in (107): (107)
208
Chan Chung
The application of the passive lexical rule in (106) to the head verb, [5]v in (107), yields (108) as output: (108)
fsuBJI COMPS (pp[dat]fl,
1 HV, pp[uyhayse]m} I
Then the lexical entry in (108) licenses the passivized sentence in (103b) as follows:
(109)
In our theory, examples such as (105) can also be accounted for in terms of the passive rule in (106). The subject of the embedded verb is attracted to the COMPS list of the matrix verb: i.e., the embedded subject NP is a complement NP in the matrix verb in the output entry of the argument composition lexical rule, and thus the matrix verb feeds the passive rule. For example, (110) is the lexical entry for the matrix verb sayngkakha 'think,' which corresponds to output of the argument composition lexical rule:
(110)
The description in (110) feeds the passive rule in (106), and its output car be either (Ilia) or (lllb), since either Np[nom]^ or NPt^r]^ can be th< subject of the passive verb:
(111) a.
Argument Composition, Long-Distance Scrambling
209
The description in (Ilia) licenses the sentence in (104b), while (lllb) licenses the sentence in (105). 4.2. Amelioration of Weak Crossover Effects Our unified account of long-distance scrambling and clause-internal scrambling in terms of the flat analysis also predicts some facts about the weak crossover (WCO) effect. In Korean, clause-internal scrambling ameliorates WCO effects as shown in (112): (112) a. *[Kui-uy/proi emma-ka] nwukwUj-lul salangha-ni! he-Gen/pro mother-Nom who-Acc love-Q 'WhOj does his; mother love?' b. NwukwUf-ul [ku^uy/^prOf emma-ka] salangha-nil who-Acc he-Gen/pro mother-Nom love-Q In (112a), the w/z-operator nwukwu-lul 'who' neither o-commands nor precedes the NP dominating a pronoun coindexed with the operator, which induces WCO effects.28 In contrast, in (112b), the operator precedes the NP dominating a pronoun, resulting in improved acceptability. Thus scrambling in Korean can ameliorate WCO effects. In Korean, the amelioration of the WCO effect also occurs with longdistance scrambling (Cho, 1994), as shown in (113) and (114):29 (113) a. ^KUf-uy/prOf emma-ka] [s Mary-ka nwukwurlul he-Gen/pro mother-Nom M-Nom who-Acc ttaylyesstako] sayngkakha-ni! hit think-Q (Literally) 'Whoi did hiSj mother think Mary hit?' b. Nwukwiij-ul [kUj-uy/^prot emma-ka] Mary-ka who-Acc he-Gen/pro mother-Nom M-Nom ttaylyesstako sayngkakha-ni! hit think-Q (114) a. *[KUj-uy/proi emma-ka] [s Mary-ka nwukwunkttj-lul he-Gen/pro mother-Nom M-Nom someone-Acc ttaylyesstako] sayngkakhanta. hit think-Q (Literally) 'HiSj mother thinks Mary hit someonej.'
210
Chan Chung
b.
Nwukwunka^-lul [kui-uy/(^proi emma-ka] Mary-ka someone-Acc he-Gen/pro mother-Nom M-Nom ttaylyesstako sayngkakhanta. hit think
In the (a) sentences, nwukwu-lul 'who' and nwukwunka-lul 'someone' neither c-command nor precede the NP dominating the pronoun, which induces a WCO effect. In contrast, in the (b) sentences, the operator precedes the NP dominating the pronoun, and thus the WCO effect is ameliorated. This observation shows that, in Korean, the WCO effect is ameliorated by scrambling regardless of whether the type of scrambling involved is clause-internal or clause-external.30 This kind of data shows that long-distance scrambling in Korean cannot be treated the same as long-distance scrambling in Hindi, because long-distance scrambling in that language does not ameliorate the WCO effect (e.g., (115b)), while clause-internal scrambling does (e.g., (115a)) (Mahajan, 1990).31 (115) a. kis-kOi uskiit bahin pyaar kartii hEt who his sister loves 'WhOj, his; sister loves tj?' b. *kis-koi uskiii bahin-ne socaa [ki raam-ne ti dekhaa thaa]. who his sister thought that Ram seen be-past 'WhOj, his; sister thought that Ram had seen tj?' If the behavior with respect to WCO effects is considered to be diagnostic of different kinds of dislocation, then the above evidence suggests that long-distance scrambling and clause-internal scrambling in Korean are not qualitatively distinct phenomena. Then there is no reason to assume another mechanism such as A-bar movement in GB or SLASH feature percolation in GPSG/HPSG for long-distance scrambling. In our theory, this lack of distinction between long-distance and clause-internal scrambling is captured since we analyze both as licensed in terms of argument composition, resulting in flat constituent structures. In this section, we discussed what theoretical predictions are made by our theory of long-distance scrambling. The phenomena discussed above suggest that, in Korean, clause-internal scrambling and long-distance scrambling are syntactically indistinguishable. Our theory of long-distance scrambling can naturally account for this fact because both types of scrambling are licensed through flat structures. This unified account is possible due to argument composition. The examples of long-distance passivization in section 4.1 suggest not only that both types of scrambling have the same properties, but also that the arguments of the embedded verb are attracted (or raised) to the COMPS list of the matrix verb.
Argument Composition, Long-Distance Scrambling
211
5. SUMMARY AND CONCLUSIONS The main proposals of this paper are summarized as follows: (i) clauseinternal scrambling and long-distance scrambling are syntactically indistinguishable, and thus they must be licensed in a uniform way; (ii) both types of scrambling result from the relative lack of linear precedence constraints among the nonhead constituents at flat clausal structures; (iii) the crucial mechanism involved here is argument composition, which is utilized to account for the verbal complex system; (iv) the coexistence of verbal complexes and long-distance scrambling in languages like Korean is accounted for in a principled way because both are licensed by the same mechanism. Our analysis is similar in spirit to the A-movement approach in GB (Yoshimura, 1989), since argument composition can be regarded as an extension of raising. However, the A-movement approach seems to be problematic under the GB framework. Under the standard assumptions, A-movement (e.g., raising and passivization in English) is triggered by certain syntactic factors (Bayer and Kornfilt, 1991). The problem is that there do not seem to be any syntactic factors triggering scrambling in Korean. Based on facts about the verbal noun construction, however, Lee (1990) proposes that case alternation and word order variation are derived from the interaction between (i) the scope of an aspect morpheme and (ii) different licensing conditions for nominative, accusative, and genitive case. (See Miyagawa, 1991, for similar claims in Japanese.) That study suggests that scrambling is allowed as long as a constituent is within the scope of an appropriate Case assignor, and thus that scrambling is a Case-driven phenomenon. However, it is still an open issue how long-distance scrambling and its effects on the WCO effect and passivization can be handled in the Case-driven theory. Yatabe (1993) proposes an analysis of Japanese long-distance scrambling which is similar to our theory. That is, scrambling out of an embedded s or VP is accounted for by either "pseudo-raising" (A-movement in terms of GB) which ameliorates the WCO effect or extraposition (A-bar movement in terms of GB) which does not ameliorate the WCO effect. In the analysis of Korean, however, there is no motivation for the distinction because the amelioration of the WCO effect occurs in the Korean counterpart of Yatabe's extraposition example, as shown in (116): (116) a. *Kui-uy/proi emma-ka [s John-i etten chinkwUj-hanthey he-Gen/pro mother-Nom J-Nom which friend-Dat yok-ul haysstako] sayngkakhayss-ni*! curse-Ace did thought-Q (Literally) 'Which friendj did hisj mother think John cursed?'
212
Chan Chung
b. Etten chinkwu-hanthey kui-uy/^prolemma-ka John-i which friend-Dat he-Gen/pro mother-Norn J-Nom yok-ul haysstako] sayngkakhayss-nil curse-Ace did thought-Q Note that in Yatabe's theory, pseudo-raising allows only the NP with accusative case to be raised and thus the examples in (116) cannot be instances of pseudo-raising. However, the WCO effect is ameliorated even in this case in Korean, which suggests that the distinction between pseudo-raising and extraposition cannot be motivated by this kind of data in that language. Finally, let us compare our argument composition analysis with the verb projection raising (VPR) analysis proposed by Haegeman and van Riemsdijk (1986). They propose the reanalysis rule in (117) to account for verbal complexes in the Germanic languages: (117)
Reanalysis If the representation of a sentence contains the line XV q V r Y, where 0 < / < 2 and Vr is a VR (verb raising) verb, then add the line XV X Y to that representation.
Here, the superscript i in Vq allows the verbal category Vq to be instantiated as V°, V, or VP, depending on the language. The rule is intended for stating that when there is a sequence of verbs, Vq and Vr, and when Vr is a VR verb, Vq and Vr are reanalyzed as Vx, as in (118):
(118)
In structure (118), where Vr is the head, the internal 0-role of Vq is assumed to percolate up to Vx, while both the internal and external 0-roles of Vr are assumed to percolate up to Vx. Then the percolated internal 0-role of Vq is assumed to be "associated with" that of Vr. Here, the notion of "associated with" is not clearly defined, but we can regard it as having the same effect as argument composition or structure-sharing. In our account, (i) the COMPS list of Vq is attracted to the COMPS list of Vr by the lexical structure of Vr, and (ii) the SUBJ list and COMPS list with the attracted elements percolate up by the Valence Principle. We may say that argument composition and the Valence Principle correspond to the internal 0-role association and 0-role percolation, respectively. Despite the apparent similarity between the VPR and our analyses, there is at least one crucial difference, which bears on the applicability range of argument composition and 0-role association. In Haegeman and
Argument Composition, Long-Distance Scrambling
213
van Riemsdijk's (1986) framework, the exhaustive association of internal 0-roles between the governed verb and the governing verb occurs only when two verbs are reanalyzed as a verbal complex. As discussed in sections 3 and 4, however, such internal 0-role association also seems to be required to account for long-distance scrambling in Korean where the governed verb and the governing verb do not form a verbal complex. That is, the notion of 0-role association cannot be extended or generalized into the account of this phenomenon because syntactic material can intervene between the governed and governing verbs in the long-distance scrambling case (e.g., (51), (69c), and (70c)). In contrast, we can use the same argument composition mechanism to account for the same phenomenon by simply changing the selection mechanism, i.e., by lexically specifying that the governed verb is selected by the governing verb's COMPS list, rather than by the GOV list of an auxiliary verb. The flexibility of our system is due to the fact that argument composition is based not on configuration but on structure sharing within a lexical entry.
ACKNOWLEDGMENTS Many parts of this work are based on my dissertation at Ohio State University. I thank Carl Pollard, Peter Culicover, Robert Levine, Robert Kasper, Craige Roberts, and Andreas Kathol for their valuable comments and suggestions. This work has also benefited from the comments of two anonymous reviewers and from discussions with the audience at the HPSG workshop held at the University of Tubingen in 1995. All remaining errors are, of course, mine.
NOTES :
The distinction between passivizable and nonpassivizable auxiliaries seems to result from semantic constraints on cz-passivization. According to Li and Thompson (1975), the passive construction in topic-prominent languages such as Japanese and Korean carries a special meaning (e.g., the "adversity" passive in Japanese). The cz-passive in Korean also seems to have a semantic constraint: i.e., the verb or verbal cluster to which the passive ci attaches must not be a verb denoting a certain state, in terms of Dowty (1979). For example, the following sentences show sharp constrasts when compared with the sentences in (7b) and (8b): (i) a. *Ku namwu-ka Mary-eyuyhayse caluko isse
the tree-Norn M-by
cut
ci-essta.
be in the process of passive-past
214
Chan Chung
'The tree was in the process of being cut.' b. *Ku namwu-ka na-eyuyhayse caluko siphe ci-essta. the tree-Norn I-by cut want passive-past 'The tree was wanted to be cut by me.' c. *Ku namwu-ka Mary-eyuyhayse calukey toye ci-essta. the tree-Nom M-by cut be led to passive-past 'The tree was led to be cut by Mary.' Here the verbal clusters to which the passive ci attaches represent states, while those in (7) and (8) represent activities. 2 We may assume that the passive morpheme ci subcategorizes for the TVP (VP without the primary object) rather than the whole VP or v, and that the primary object is directly subcategorized for by ci. An example of the analysis is illustrated in (i): (i) [s Malssengmanhun so-ka [vp (ku nongpwu-eyehay) troublesome cow-Nom the farmer-by [VP [-rvp John-hanthey phala chiwe] ci-essta]]] J-Dat sell resolutely passive-past 'The troublesome cow was resolutely sold to John (by the farmer).' A problem with this analysis is that it is hard to find any syntactic evidence for the constituency of the TVP. As discussed in section 3.1.1, any phrasal category can occur after the head verb as an afterthought expression. However, the TVP in (i) cannot be an afterthought expression as shown in (ii): (ii) *Malssengmanhun so-ka ku nongpwu-eyehay ci-essta, troublesome cow-Nom the farmer-by passive-past [TVP John-hanthey phala chiwe]. J-Dat sell resolutely Also note that no parenthetical can intervene between ci and its governed verb, as shown in (iii), again exhibiting a behavior different from that of other VP complement constructions (cf. (3ciii) and (4)): (iii)
*Malssengmanhun troublesome phala chiwe] sell resolutely
so-ka ku nongpwu-eyuyhay [yyp John-hanthey cow-Nom the farmer-by J-Dat hayekan ci-essta. anyway passive-past
3 Pollard and Sag (1987) define the Locality Principle as follows: the SUBCAT elements of lexical sign (word in Pollard and Sag, 1994) specifies values for SYNTAX and SEMANTICS (SYNSEM in Pollard and Sag, 1994) but crucially not the attribute DAUGHTERS. However, Pollard and Sag (1994) assume that elements of SUBJ or COMPS lists are not of sort sign but sort synsem, which does not include the attribute DAUGHTERS. In this framework, the Locality Principle is incorporated into the definition of the features, and thus it does not need to be stated separately. 4 This terminology is taken from No (1991). 5 Hinrichs and Nakazawa (1989, 1994) use the version of HPSG of Pollard and Sag (1987) in which the valence is not divided into subject and complement lists, but instead encoded in a single list (SUBCAT).
Argument Composition, Long-Distance Scrambling
215
6 This is essentially the same as the CONS function in LISP which takes an element and a list to return a list: element + list => list. 7 An exception is the passive auxiliary ci, which will be discussed shortly. 8 Rule (22) excludes the possibility that an auxiliary first combines with the subject and/or complements, not with the governee verb, and licenses an ill-formed sentence in (ia): that is, in the structure in (ib), rule (22) cannot apply to the top node to discharge the GOV value because the sort of the head is not a simple word but a phrase:
(i) a. *Mekko Mary-ka sakwa-lul siphessta. eat M-Nom apple-Ace wanted 'Mary wanted to eat an apple.' b
' *S[SUBJ< > , C O M P S < >,GOV<[3]V>]
9
A similar approach is taken in Yoo (1993:196). Yoo adopts the GOV feature in Chung (1993) for lexical entries of emotion auxiliaries. The lexical entries proposed by her have a minor technical problem as they are stated, since they lead to saying that an emotion verb takes only a transitive verb as its governee. 10 We admit that the assumption of two separate lexical entries for siph weakens the argument. Presently, we have no good explanation why siph causes this case alternation. 11 In Chung (1995), the afterthought expressions are treated differently from scrambling. See Chung (1995: Chap. 4) for detailed discussion on this matter. 12 Some more object-control verbs in Korean are kangyoha 'compel,' malha 'tell,' canglyeha 'encourage,' caychokha 'urge,' pwuthakha 'ask,' helakha 'allow,' and sikhi 'make.' Subject-control verbs are mayngseyha 'vow,' sitoha 'try,' and tonguyha 'agree.' The control verb construction in Korean discussed above is similar to the so-called "optionally coherent" construction in German that has been discussed since Bech (1955). 14 In Korean, the raising-to-subject construction is an instance of the verbal complex construction, which is headed by auxiliaries such as po 'seem,' tusha 'seem, look like,' etc. In this construction, NP arguments can scramble with each other (e.g., (ia,b)). However, the governed verb and the auxiliary cannot be separated by inserting syntactic material (e.g., (ic,d)) or by making the governed verb an afterthought expression (e.g., (ie)). As discussed in section 2, these are the typical properties of verbal complexes. (i) a. Mary-ka ku chayk-ul ilk-ess-na pota. M-Nom the book-Ace read-Past-vform.na seem 'Mary seemed to read the book.'
216
Chan Chung
b. Ku chayk-ul Mary-ka ilk-ess-na pota. the book-Ace M-Nom read-past-vform.na seem c. *Ku chayk-ul ilk-ess-na Mary-ka pota. the book-Ace read-past-vform.na M-Nom seem d. *Mary-ka ku chayk-ul ilk-ess-na hayekanpota. M-Nom the book-Ace read-Past-vform.na anyway seem 'Anyway Mary seemed to read the book.' e. *Mary-ka pota ku chayk-ul ilk-ess-na. M-Nom seem the book-Ace read-Past-vform.na On our approach, the valence value of the raising-to-subject auxiliary po is the same as (23). See Sells (1990) for a similar analysis. 15 In Korean, nominative case can generally be assigned to the subject of a nonfinite clause, as shown in (i): (i) John-un [s [atul-i tolawa-se] kippessta. J-Top son-Norn return (base-form)-because was happy 'John was happy because his son returned.' 16
The lexical rules for long-distance scrambling in (67) may be considered lexical versions of the liberation metarule in Pullum (1982) or the liberation principle in Zwicky (1986). 17 Even though our theory can capture the different scrambling possibilities between control and raising verb constructions by restricting the application of the composition rule to the control construction, it does not explain why such a difference exists. If the difference is caused by an independent reason, we may generalize the lexical rule so that it can apply to all lexical entries which take any kind of verbal expression as a complement. For example, an alternative analysis for the raising-to-object construction in Korean may be possible if we consider Hong's (1990) analysis. According to that proposal, the accusative controller is an embedded topic. This analysis naturally eliminates the unacceptable sentences in (54d,e) and (55), i.e., the embedded topic should precede all of its clause-mate elements. However, we still need the LP constraint in (74) to account for the accusative ha -causative construction, where the causee NP cannot be assumed to be an embedded topic since it allows the complement of the embedded verb to precede the causee (e.g., (64)). A more detailed analysis of this construction is provided in the next section. 18 Another way to analyze (77b) and (78b,c) may be as instances of English-style topicalization (Bratt, 1993). This view is suspicious because no obvious evidence exists which shows that extraction out of an embedded clause differs from the other extractions. See section 4 for detailed discussion on this. 19 We have no explanation at this point for the marginal status of (78b,c). The acceptability improves with short pauses after the NP arguments, which is indicated by commas. 20 In (80b), HP'S case is not specified when assigned case is structural) just for expository convenience. 21 Case marking on arguments can affect scrambling possibilities: i.e., an argument of an embedded verb cannot scramble with an argument of a matrix verb
Argument Composition, Long-Distance Scrambling
217
when they bear the same case marker, and a nominative argument of an embedded verb cannot scramble with an argument of a matrix verb. To account for the former case, Chung (1995) assumes an LP constraint into which the notion of obliqueness is incorporated: coarguments with the same case must be linearized in order of obliqueness: i.e., the less oblique one must precede the more oblique one. To account for the latter case, Chung (1995) assumes an interpretive principle which roughly states the following: Suppose (i) that Y is an Np[nom], (ii) that X is the first verb following Y, and (iii) that Z is any constituent which occurs between Y and X. Then Z cannot be a "semantic dependent (semantic argument or functor)" of a verb superordinate to X. Here semantic dependency involves structure sharing between the functor and its argument in their CONTENT values. The interpretive principle is not affected by argument composition because argument composition changes only the structure of the VALENCE value. See Chung (1995: Chap. 4) for more detailed discussion on this matter, and also see section 3.2.3 in this paper for another example where this principle is utilized. 22 A potential problem for (83) may be the semantic scope relations among more than one adjunct. In Kasper (1994), the scope of multiple adjuncts is captured in the flat structure analysis (i) by the order of adjunct daughters in the adjunct daughter list, which is ordered not by the surface order of the adjuncts but by their semantic scope, and (ii) by a split of the MOD value into SYN and SEM, reflecting the dual nature of adjuncts as semantic heads (functors), but syntactic nonheads. While a more sophisticated version of (83) will have to take these complications into account, the present version will suffice for our purposes. 23 The idea of cospecification of the modifier and modified can be considered as a hybrid of Pollard and Sag (1987) and Pollard and Sag (1994). In the former, a modifier is selected by the modified category, while in the latter, the modified category is selected by its modifier. 24 Here, in order to simplify the CONTENT value, we do not provide the semantics of the modal auxiliary verb, hanta 'must,' which is not directly relevant to current discussion. Also we simplify the CONTENT value of the adjunct nayil kkaci 'by tomorrow' by using the notation of intensional logic: by-tomorrow' (3). 25 Presently, it is not clear why (94c) is awkward. Our conjecture is that the awkwardness may be due to some processing factor. In Korean, when an adjunct immediately precedes a category which is modifiable by the adjunct, speakers tend to interpret the adjunct as modifying the immediately following category. In (94c), the adjunct ilcwuil ceney 'one week ago' immediately precedes the verb kkuthnaylako 'finish,' which is a category modifiable by the PP. Thus, the PP tends to be interpreted as modifying kkuthnaylako rather than malhayssta 'told.' A pause or stress may block this kind of processing interference. This conjecture is supported by (i): (i) Mary-ka John-hanthey ku project-lul ilcwuil ceney kuphi M-Nom J-Dat the project-Ace one week ago in haste kkuthnaylako malhayssta. finish said/told 'One week ago Mary told John to finish the project in haste.'
218
Chan Chung
Here, the adjunct ilcwuil ceney 'one week ago' and the embedded verb kkuthnaylako 'finish' are separated by another adjunct which modifies the embedded verb; consequently, (i) is much better than (94c). 26 The LP constraints encoding head-finality (e.g., (72), (74), and (99)) can be schematized as in (i), where F ranges over all valence features: (i)
[SYNSEM[1] < [VAL|F < . . . , m , . . . >] 27
These adverbs in Korean are usually derived by suffixing -(h)i or -key to a root of a stative verb. 28 See Chung (1995) for more detailed discussion in WCO effects in Korean and Pollard and Sag (1994) for the definition of o-command. 29 Saito (1992) and Yoshimura (1989) make the same observation about Japanese. 30 See (116) in section 5 for the case where an indirect object affects the WCO effect. 31 In our theory, long-distance scrambling in Hindi may be treated as a "genuine" unbounded dependency construction, on a par with English-style topicalization.
REFERENCES Bayer, J., and J. Kornfilt (1991). Against scrambling as move-alpha. NELS 21, 1-15. Bech, G. (1955). Studien fiber das deutsche Verbum Infinitum, Vol. 1 KDVS35, 2. Copenhagen. Borsley, R. (1989). An HPSG approach to Welsh. Journal of Linguistics 25, 333-354. Bouma, G., and G. van Noord (1998). Word order constraints on verb clusters in German and Dutch. In E. Hinrichs, A. Kathol, and T. Nakazawa (eds.), Complex. Predicates in Nonderivational Syntax. Syntax and Semantics, Vol. 30, 43-72. Academic Press, San Diego. Bratt, E. (1993). Clause structure and case marking in Korean causatives. In S. Kuno et al. (eds.), Harvard Studies in Korean Linguistics V, 241-251. Hanshin, Seoul. Cho, J-H. (1994). On scrambling: Reconstruction, crossover, and anaphor binding. In Y-K. Kim-Renaud (ed.), Theoretical Issues in Korean Linguistics, 255-274. CSLI Publications, Stanford, CA. Cho, J. O. (1988). Suffixed verb forms and compound verb constructions. In E. Back (ed.), Papers from the Sixth International Conference on Korean Linguistics, 77-106. Hanshin, Seoul. Chung, C. (1993). Korean auxiliary verb constructions without VP nodes. In S. Kuno et al. (eds.), Harvard Studies in Korean Linguistics V, 274-286. Hanshin, Seoul. Chung, C. (1995). A Lexical Approach to Word Order Variation in Korean. Ph.D. dissertation, Ohio State University.
Argument Composition, Long-Distance Scrambling
219
Dowty, D. (1979). Word Meaning and Montague Grammar. Reidel, Dordrecht. Dowty, D. (1990). Toward a minimalist theory of syntactic structure. In Tilburg Workshop on Discontinuous Constituency, 34-72. Gerdemann, D. (1994). Complement inheritance as subcategorization inheritance. In J. Nerbonne, K. Netter, and C. Pollard (eds.), German in Head-Driven Phrase Structure Grammar, 341-363. CSLI Publications, Stanford, CA. Haegeman, L., and H. van Riemsdijk (1986). Verb projection raising, scope, and the typology of rules affecting verbs. Linguistic Inquiry 17, 417-466. Hale, K. (1982). Preliminary remarks on configurationality. NELS 12. (J. Pustejovsky and P. Sells (eds.).) Heinz, W., and J. Matiasek (1994). Argument structure and case assignment in German. In J. Nerbonne, K. Netter, and C. Pollard (eds.), German in HeadDriven Phrase Structure Grammar, 199-236. CSLI Publications, Stanford, CA. Hinrichs, E., and T. Nakazawa (1989). Flipped out: AUX in German. Chicago Linguistic Society 25. Hinrichs, E., and T. Nakazawa (1994). Linearizing AUXs in German verbal complexes. In J. Nerbonne, K. Netter, and C. Pollard (eds.), German in Head-Driven Phrase Structure Grammar, 11-37. CSLI Publications, Stanford, CA. Hong, K-S. (1990). Subject-to-object raising in Korean. In K. Dziwirek, P. Parrel, and E. Mejias-Bikandi (eds.), Grammatical Relations: A Cross-Theoretical Perspective, 215-226. CSLI Publications, Stanford, CA. Kang, B. M. (1988). Functional Inheritance, Anaphora, and Semantic Interpretation in a Generalized Categorial Grammar. Ph.D. dissertation, Brown University. Kasper, R. (1994). Adjuncts in the Mittelfeld. In J. Nerbonne, K. Netter, and C. Pollard (eds.), German in Head-Driven Phrase Structure Grammar, 39-69. CSLI Publications, Stanford, CA. Kathol, A. (1994). Passive without lexical rules. In J. Nerbonne, K. Netter, and C. Pollard (eds.), German in Head-Driven Phrase Structure Grammar, 237-272. CSLI Publications, Stanford, CA. Kiss, T. (1992). Infinite Komplementation. Neue Studien zum deutschen Verbum Infinitum. Ph.D. dissertation, Bergische Universitat-Gesamthochschule Wuppertal. Kiss, T. (1994). Obligatory coherence: The structure of German modal verb constructions. In J. Nerbonne, K. Netter, and C. Pollard (eds.), German in Head-Driven Phrase Structure Grammar, 71-107. CSLI Publications, Stanford, CA. Kuh, H. (1990). Correlation between Inflection and Word Order. Ph.D. dissertation, Ohio State University. Lee, Y-S. (1990). Case alternation and word order variation in nominal clauses. In H. Hoji (ed.), Japanese /Korean Linguistics. CSLI Publications, Stanford, CA. Li, C., and S. Thompson (1975). Subject and topic: A new typology of language. In C. Li (ed.), Subject and Topic, 457-489. Academic Press, New York. Mahajan, A. (1990). The A/A-Bar Distinction and Movement Theory. Ph.D. dissertation, Massachusetts Institute of Technology, Cambridge, MA.
220
Chan Chung
Miyagawa, S. (1991). Case realization and scrambling. Unpublished manuscript, Ohio State University. Nagai, N. (1991). Complex passive and major subjects in Japanese. Linguistics 29, 1053-1092. No, Y. (1991). Case Alternations on Verb-Phrase Internal Arguments. Ph.D. dissertation, Ohio State University. Pollard, C. (1994). Toward a unified account of passive in German. In J. Nerbonne, K. Netter, and C. Pollard (eds.), German in Head-Driven Phrase Structure Grammar, 273-296. CSLI Publications, Stanford, CA. Pollard, C., and I. Sag (1987). Information-Based Syntax and Semantics. CSLI Lecture Notes, Vol. 13. CSLI Publications, Stanford, CA. Pollard, C., and I. Sag (1994). Head-Driven Phrase Structure Grammar. University of Chicago Press, Chicago, and CSLI Publications, Stanford, CA. Pullum, G. (1982). Free word order and phrase structure rules. NELS 12, 209-220. Ryu, B-R. (1993). Structure Sharing and Argument Transfer: An HPSG Approach to Verbal Noun Constructions. Sfs-Report-04-93, University of Tubingen. Saito, M. (1985). Some Asymmetries in Japanese and Their Theoretical Implications. Ph.D. dissertation, Massachusetts Institute of Technology, Cambridge, MA. Saito, M. (1992). Long distance scrambling in Japanese. Journal of East Asian Linguistics 1, 69-118. Sells, P. (1990). VP in Japanese: Evidence from -te complements. In H. Hoji (ed.), Japanese /Korean Linguistics, 319-334. CSLI Publications, Stanford, CA. Sells, P. (1991). Complex verbs and argument structures in Korean. In S. Kuno et al. (eds.), Harvard Studies in Korean Linguistics IV, 395-406. Hanshin, Seoul. Yatabe, S. (1993). Scrambling and Japanese Phrase Structure. Ph.D. dissertation, Stanford University. Yoo, E. J. (1993). Subcategorization and case marking in Korean. In A. Kathol and C. Pollard (eds.), Papers in Syntax. Ohio State University Working Papers in Linguistics Vol. 42, 178-198. Yoon, J. H. S. (1993). Tense, coordination, and the clausal structures of English and Korean. In S. Kuno et al. (eds.), Harvard Studies in Korean Linguistics V, 436-446. Hanshin, Seoul. Yoshimura, N. (1989). Parasitic pronouns. Paper presented to the Japanese/Korean Linguistic Conference at UCLA. Zwicky, A. (1986) Concatenation and liberation. CLS 22. 65-74.
CONSTITUENCY AND LINEARIZATION OF VERBAL COMPLEXES ANDREAS KATHOL Department of Linguistics University of California at Berkeley Berkeley, California 94720
1. INTRODUCTION
In the descriptive toolbox of modern feature-based theories such as Lexical-Functional Grammar (LFG) and Head-Driven Phrase Structure Grammar (HPSG), one idea that has led to important new ways to think about the nature of linguistic knowledge is the concept of informational identity, that is in the HPSG case token-identity of information from possibly rather disparate sources. Analyses based on this concept have, for instance, yielded significant descriptive and explanatory advances in the area of nonfinite complementation. On the common transformational approach, any dissociation between syntactic and semantic "subjecthood," such as in subject-to-subject raising constructions, is achieved by means of transformational relationships among syntactic tree representations. By contrast, it has become part of most nonderivational accounts of English to think of English raising constructions in terms of a (partial) identity between the raising verb's subject and the understood subject of the VP complement. In Pollard and Sag's (1994) rendition of raising as structure sharing, the subcategorization information of the subject-raising verb consists of an NP subject and a VP complement. That VP is an unsaturated verbal projection, which means that it still requires a subject to yield a full Syntax and Semantics, Volume 30 Complex Predicates in Nonderivational Synta.
221
Copyright © 1998 by Academic Press All rights of reproduction in any form reserved. 0092-4563/98 $25.00
222
Andreas Kathol
sentence. This subject is not provided by means of some inaudible element, but rather, the coreference tag " Q] " in (1) indicates token-identity with the subject of the embedding raising verb. In other words, the "understood subject" of the VP complement is realized syntactically as the subject of the higher predicate. (1)
[ . . . | SUBCAT ( CDNP, VP [ . . . | SUBCAT < CD >] ) ]
In the same lexicalist spirit, extending the domain of application for structure sharing further, Hinrichs and Nakazawa's concept of argument composition (first introduced in Hinrichs and Nakazawa, 1990 has had a great impact on the analysis of nonfmite complementation in head-final languages. The basic idea is to extend the linkage between the valence of the embedding and embedded predicates to all arguments, not just the subject. As the description in (2) illustrates, the whole valence value is shared among the raising verb itself and the verbal constituent subcategorized for by the latter. The entire list of selected arguments then consists of the singleton list containing the verbal complement appended (notated as " °") to the list of arguments "attracted" from the lower predicate. (2)
[. . . | SUBCAT H o]
Ignoring the verbal complement, this means that the raising verb becomes an intransitive predicate if and only if the selected predicate is intransitive and so on for predicates of all arities, as exemplified for transitives in (3a). Moreover, in a language such as German which possesses impersonal predicates—that is, without a syntactic subject even in the finite forms—structure sharing of the valence information entails that the raising verb too will become impersonal if and only if the embedded verb is. This situation simply corresponds to the case in which the list of attracted arguments is empty, as shown in (3b): (3)
a. [ ... | SUBCAT El° ]
b. [... I SUBCAT Q] < > °] As Kiss (1992) shows, there are essentially no syntactic differences in their behavior under argument composition between different classes of predicates in German that have not necessarily been treated alike in the transformational literature.1 Thus, not only classical raising verbs such as schein(en), but also tense auxiliaries and a subclass of modals pattern exactly alike in this characteristic realization of valence requirements of some predicate by some other verb. For instance, all three are eligible
Verbal Complexes
223
governors of impersonal passive constructions: (4) a. daB gearbeitet zu werden scheint. that worked to be seems 'because there seems to be working going on' b. daft gearbeitet worden ist. that worked been is 'because there has been working going on' c. daft gearbeitet werden muB. that worked be must 'because there ought to be working going on' As Kiss further shows, it is in fact possible to correlate the idea of argument composition with the notion of coherent construction, which was introduced into the syntactic description of German by Bech (1955). To this end, it is necessary to allow for a variant of argument composition in which the embedding predicate inherits all the arguments of the lower verb, but at the same time requires there to be a thematic subject. This situation arises in the case of certain EQUi-control verbs such as versuchen ('try'), which allow for a coherent complementation pattern indicative of argument composition as indicated for instance by the "scrambled" argument order in (5a), but at the same time obligatorily take a subject (5b). (5) a. daB ihm der Mann zu helfen versucht. that him-DAT the man-NOM to help tries 'that the man tries to help him.' b. *dafi gearbeitet zu werden versucht. that worked to be tries In such cases, the pattern of shared information is as outlined in (6):2 (6)
[. . . | SUBCAT Q]o ]
2. ARGUMENT COMPOSITION AND BRANCHING STRUCTURE In their original proposal, Hinrichs and Nakazawa assume that a cluster of verbal elements3 related by argument composition is the result of a sequence of binary combinations of an embedding verb with its verbal complement to the exclusion of any inherited ("proper") arguments. Canonical sequences of governors following governed verbs will then arise
224
Andreas Kathol
from left-branching structures, as the following example illustrates: (7) a. daft Peter das Buch finden konnen wird. that Peter the book find can will 'that Peter will be able to find the book' b.
However, the lexical specification of verbs involved in argument composition is actually silent with respect to the way that lexical information is projected onto syntactic structure. This means that alternative branchings are possible—and, in fact, have been proposed in the literature. For instance, Kiss (1992) adopts a right-branching analysis of the sort illustrated in (8): (8)
Apart from the directionality of branching, the analyses in (7) and (8) also differ in another respect, namely, whether the source of raised arguments can itself get raised by a higher predicate. Thus, in (7), only phrasal arguments get successively inherited from finden by konnen and wird. By contrast, in (8) the argument source finden is among the list of arguments that wird inherits from konnen. As a consequence, the highest governing predicate (here, wird) may accumulate an arbitrary number of verbal elements on its SUBCAT list whereas in Hinrichs and Nakazawa's approach,
Verbal Complexes
225
each governing verb will have at most one verbal complement in addition to the list of inherited phrasal arguments. The possibility of allowing sources for raised arguments to be inherited themselves by higher predicates does not automatically lead to a rightbranching structure.4 Rather, the syntactic structure of the verb cluster—if it is to be treated as a syntactic unit at all—crucially depends on the type of syntactic licensing mechanism adopted. While Kiss assumes binary branching generally, a popular alternative has been to assign a flat structure to the German Mittelfeld. On this approach, favored for example by Pollard et al. (1993), Baker (1994), van Noord and Bouma (1995), and Bouma and van Noord (1998), verb clusters simply constitute a subset of sisters within the flat tree making up a clause in a single step:5
(9)
This display of possibilities for the analysis of complex verbal structures in languages such as German raises the immediate question whether there is any empirical ground for a choice among the options. This study is an attempt to provide a positive answer to that question based on considerations of constituency and order variation. The analysis argued for here takes the left-branching analysis of Hinrichs and Nakazawa as its point of departure, contrasting it with the right-branching alternative which is found wanting in serious respects.6 However, closer inspection reveals that Hinrichs and Nakazawa's approach leads to a number of serious shortcomings which can be avoided if the licensing mechanism involved in the construction of verbal complexes is dissociated from that of phrasal arguments. This then also forms the basis for a novel approach toward the linearization of verbal complexes which is able to accommodate certain kinds of discontinuities seen in various German dialects and Dutch.
3. CONSTITUENCY 3.1. The Case for Left Branching One of the crucial arguments adduced by Kiss for his choice of a right-branching structure over a left-branching alternative is that it ex-
226
Andreas Kathol
ploits the division into word vs phrase-level constituents that is already provided by the grammar. That is, the valence characterization of an argument raiser such as wird in (8) makes reference to the fact that the element providing the inherited arguments is a word. In HPSG, the distinction between lexical and phrasal constituents is often made in terms of different values of the feature LEX; accordingly the lexical description of an argument-composing predicate would require a positive specification of this property:
(10) Hence, argument composition involves what Kiss calls lexeme selection. Together with ordinary, phrasal selection, this then sets up a typology of complementation types which is in parallel to the word-phrase distinction already made in the grammar. Contrast this situation now with the kind of constituents involved in left-branching structures. In (8), we need to identify sequences such as finden konnen as syntactic units that can be selected by the embedding verb. As things stand, the only distinguishing property between a "pure" verbal complex like finden konnen and one that has combined with phrasal complements such as Peter das Buch finden konnen lies in their valence—yet the central characteristic of argument raisers like wird is that they are indiscriminate with respect to the length and content of the inherited argument list. Thus in order to force the construction of verbal complexes to the exclusion of phrasal arguments, Hinrichs and Nakazawa need to have a way of distinguishing purely verbal combinations from those containing phrasal constituents and make the selectional properties of argument inheritors sensitive to that distinction. For this purpose, they postulate the feature NPCOMP, which bears a negative value so long as no phrasal constituent has combined with the amalgam of verbal elements:
(11)
Verbal Complexes
227
At first blush this seems like a crude way of achieving the required task, as it does not appear to pick out a natural class of objects or line up with any other distinction made in the grammar. But following a proposal first made by Rentier (1994), a somewhat different conclusion suggests itself in which the combination of verbal elements fits more straightforwardly into the typology of syntactic categories than Kiss suggests. Rentier adopts the characterization in (10), but gives a somewhat different interpretation to the feature LEX. Instead of lining up [LEX ± ] with HPSG's sortal distinction of signs into word and phrase, Rentier also allows verbal combinations such as finden konnen to bear the specification [LEX + ]. Being of sort word is then a sufficient, but no longer necessary, condition of a positive value of the LEX attribute. We can therefore regard elements with such a positive specification as possessing "(quasi-)lexical" status. This means that, even though syntactically the status of (partial) verbal complexes is phrasal, the amalgamation of verbal elements nevertheless behaves as a single element with regard to selecting phrasal arguments.7 Further crosslinguistic support for this conception comes from considering the status of corresponding structures in other head-final languages such as Korean and Japanese. While, in German, verbal complexes clearly constitute combinations of separate lexical elements—this conclusion is inevitable8 given the kind of order variation to be discussed in later sections—a clear separation from morphological amalgams, particularly those involving derivation or compounding, becomes much less clear (cf. for instance causatives (lida et al., 1994:7)): (12) a. Ken-ga Naomi-ni hon-o yom-ase-ta. Ken-NOM Naomi-DAT book-ACC read-CAUS-PAST 'Ken made Naomi read the book.' b. daft Hans Maria das Buch lesen lassen hat. that Hans Maria the book read made has 'that Hans made Maria read the book' From this perspective, then, both German and Japanese verbal complexes consist of combinations of verbal material with an exactly parallel combinatorial structure resulting in a construction that behaves like a single lexical unit ([LEX + ]) with respect to valence properties. The only difference is whether the combination takes place exclusively in the syntax or may also involve other components of the grammar, that is, morphology. By contrast, no comparable parallelism would exist if a right-branching analysis were adopted. This is because, in analogy with (8), one would be required to make the implausible assumption that strings of derivational affixes always combine into some morphological unit before attaching to the main verb. Of course this argument only carries force if one is willing
228
Andreas Kathol
to allow for the possibility of certain linguistic phenomena exhibiting parallel properties even though both may belong to clearly distinct grammatical subsystems, but this seems hardly a contentious issue. Having shown that Kiss' typology-based argument for right-branching structures can easily be countered by an appeal to cross-linguistic parallelism of structure, we now move on to factual empirical evidence for left branching. This involves two constructions in which the subconstituents postulated within the verbal complex actually exhibit properties of a syntactic unit. In the first of these constructions (sequences of) nonfinite verbs occur fronted in prefinite position in V2 clauses. As the examples in (13) show, not only ordinary phrasal constituents such as the direct object in (13a), but verbal constituents as well may be fronted, either alone, as in (13b), or as a sequence, as in (13c):9 (13) a. Das Buch wird Peter finden konnen. the book-ACC will-FIN Peter-NOM find can 'Peter will be able to find the book.' b. Finden wird Peter das Buch konnen. find will-FiN Peter-NOM the book-ACC can c. [Finden konnen] wird Peter das Buch. find can will-FiN Peter-NOM the book-ACC Even though, as we will see in due course, the exact formulation of this pattern presents certain difficulties under standard HPSG assumptions about complementation, the basic explanation is quite straightforward. Any governed subconstituent within a verbal complex can be fronted, that is, any verb or combination of verbs that is selected by a higher predicate (inheriting the former's arguments). By contrast, a right-branching structure not only has to accommodate the frontability of elements that do not form a syntactic unit,10 it also provides syntactic groupings, such as konnen wird, that do not behave as such under any diagnostic.11 Furthermore, it is far from clear how to reconstruct a relevant notion of government under the premise that branching is to the right. One may, for instance, consider contiguous sublists of verbal elements on the SUBCAT list as candidates for preposing. But not only would this solution be ad hoc, more significantly this move would essentially amount to reconstructing a notion of syntactic unit that is already given for free in the left-branching analysis. Another powerful argument against the right-branching analysis comes from order variation within the verbal complex in German. In cases where the verbal cluster contains a contiguous sequence of two or more morphological bare infinitives,12 the governing verb may (as in the case of the future auxiliary werden) or must (as with the perfect auxiliary haben)
Verbal Complexes
229
precede, rather than follow, the sequence of governed verbs. This type of construction is known among German grammarians as Oberfeldumstellung; an alternative label Hinrichs and Nakazawa use is Aux Flip. (14) a. daft Peter das Buch wird finden konnen. that Peter the book will find can 'that Peter will be able to find the book' b. daft Peter das Buch hat finden konnen. that Peter the book has find can 'that Peter has been able to find the book' The exact mechanisms employed by Hinrichs and Nakazawa to account for this construction will be subject to more detailed scrutiny later. At this point, however, it is important to note that their proposal lends itself rather naturally to cover these cases in terms of a simple reordering of the governing verb with respect to the governed subcomplex:13
(15)
By contrast, in a right-branching analysis, Oberfeldumstellung cannot be accounted for in terms of a simple reordering between governing verb and governed subcomplex. In fact, it is not clear if any mechanism could be found that yields an analysis of Oberfeldumstellung which remains faithful to the constituent structure in (8)—to the best of our knowledge, no such proposal has ever been advanced.14 The two phenomena considered here—partial verb cluster fronting and Oberfeldumstellung—thus provide strong evidence for the kind of constituency structure that arises in connection with assigning a binary leftbranching structure to the verb cluster. Whatever the initial plausibility of right-branching analysis, it has been shown that it cannot make up for the complete inability to provide any insight into these constructions. On the other hand, a reinterpretation of the function of the LEX feature allows us
230
Andreas Kathol
to view verbal complexes in German as the syntactic realization of complex predicates which in other languages exhibit more distinctly morphological properties. Despite its success, however, a closer look at partial verb cluster fronting cases will necessitate a reconsideration of Hinrichs and Nakazawa's idea that verbal complexes are built up via the same valence attribute as is involved in phrasal complementation, viz. SUBCAT.15 3.2. The Case for VCOMPL When we consider what parts of a verbal complex can undergo fronting, it turns out that there are severe limits which have long been known in the literature. Thus, while a main verb can be fronted alone, as was shown in (13b), repeated in (16a), this is not the case for a predicate that triggers argument composition. Thus, the example in (16b) is strongly ungrammatical: (16) a. Finden wird Peter das Buch find will-FiN Peter-NOM the book-Ace b. *Konnen wird Peter das Buch can will-FiN Peter-NOM the book-Acc
konnen. can finden. find
On the standard assumption that all types of selectional dependencies arise from regular valence features—that is, either the unitary SUBCAT feature or combination of SUBJ and COMPS features—this intuition turns out to be rather difficult to state succinctly. For instance Nerbonne (1994:139) imposes on his lexical rule for complement extraction the condition that if (the projection of) a modal verb is fronted via SLASH, there must not be any verbal element on that modal verb's SUBCAT list. This condition hence involves negated existential quantification over set elements. In addition, it is not clear if one can find a natural motivation for such a constraint. In contrast to modal verbs, a predicate such as versuchen may take a phrasal complement with zu -infinitival morphology. In such cases, the verbal complement seems to be able to "stay behind" when the governing predicate is fronted—examples such as in (17), while not entirely perfect, nevertheless are markedly better than the one in (16b): (17) ?Versprochen wird er ihr wohl [den Wagen zu waschen] promised will he her probably the car to wash nicht haben. not have 'He will probably not have promised her to wash the car.' If this is correct, Nerbonne's condition against inheriting verbal elements from modals (or tense auxiliaries) seems arbitrary.16
Verbal Complexes
231
While there seems to be a general consensus to assimilate the construction of verbal complexes to complementation in general by regarding the selection to be in terms of SUBCAT/COMPS, the particular assumptions about branching structure nevertheless force Hinrichs and Nakazawa to adopt a separate ID schema dubbed "Verbal Complex ID Schema" (Hinrichs and Nakazawa, 1994b). But from this, it is only a small step toward projecting (partial) verbal complexes by an altogether different valence attribute. This is indeed what Chung (1993, 1998) proposes with the new valence feature GOV for Korean, which has since been adopted for Dutch by Rentier (1994). We will do the same here, albeit using the name VCOMPL for "verbal complement/complex,"17 which is arguably more adequate as it stresses the special involvement in the formation of complex predicates. Given this reformulation, the lexical description of a predicate controlling argument composition is as along the lines given in (18):
(18)
The removal of the verbal complement from the SUBCAT list gives us, in fact, a straightforward way of characterizing frontable constituents as the natural class of elements that bear the specification [VCOMPL < )]. This description covers ordinary phrasal arguments at the same time that it excludes predicates such as konnen or other verbal sequences bearing a nonempty VCOMPL value. In addition, there are at least three more arguments for separating the dependencies that play a role in complex predicate construction from ordinary complementation via SUBCAT. Kiss (1992) presents an attempt to relate extraposability to the incoherent vs coherent construction type in German, which is correlated with the selection of VP complements vs argument composition. From that perspective, it is tempting to identify extraposable verbal constituents in quite general terms by virtue of their valence properties, i.e., to allow for VP complements to be extraposed while the nonphrasal projections selected in argument composition are barred from extraposition. The problem with such a proposal, however, is the existence of VPS headed by lexically intransitive verbs. If a verb of this kind, for instance schlafen ('sleep'), combines with an argument-raising governor, it will bear the same valence properties as a "genuine," phrasal VP, namely, [SUBCAT {synsem}]. Yet, as the example in (19b) illustrates, despite the identity in valence and morphological properties (zw-infinitive), the
232
Andreas Kathol
complement of scheinen may not extrapose: (19) a. weil Lisa versucht [zu schlafen] because Lisa tries to sleep 'because Lisa tries to sleep' b. *weil Lisa scheint [zu schlafen] because Lisa seems to sleep One possibility is to assign a difference in terms of the word-phrase status for the complement of versuchen vs scheinen. This could be accomplished by means of Pollard's (1996) Schema B', which allows the creation of a phrasal verbal projection even when there are no complements to discharge. But then this schema gives rise to a nonbranching structure for which one can arguably provide no theory-independent motivation. On the other hand, once the distinction is made between selection via VCOMPL vs SUBCAT, it becomes straightforward to treat complements differently in each case. Only verbal complements18 selected via SUBCAT are eligible for extraposed occurrence, but never those selected by VCOMPL—regardless of their morphological or valence properties. There is a second aspect in which the formation of verbal complexes via VCOMPL seems advantageous over an approach that does not distinguish between phrasal and quasi-lexical arguments. While separable prefixes in German are mainly prepositional in origin, there are examples of noun + verb combinations such as Autofahren ('drive (a) car'), in which the nominal part arguably has the status of a separable prefix. In some cases, such "incorporated" nouns occur in the plural, as for instance in Kartenspielen ('play cards'). If the object Karten is selected via VCOMPL, then we have the prediction that even though it has the appearance of an accusative object, it should not be eligible to undergo passivization—the latter being a valence alternation only for subcategorized, phrasal arguments selected via SUBCAT. As a result, the whole predicate should count as intransitive, allowing for impersonal passive. As Kroch and Santorini (1991:295) observe, this is precisely what we find, as shown by the number inflection on the passive auxiliary in (20): (20) Damals wurde/* warden haufig Karten gespielt. then was/were often cards played 'At that time, there was frequent playing of cards.' Finally, as has been pointed out for instance by Stiebels and Wunderlich (1994:927), separable prefixes behave differently from regular phrasal arguments in progressive constructions formed with the amalgamated preposition am ('at the') in certain German dialects. As is illustrated in (21), am necessarily precedes the elements of the verb cluster, including
Verbal Complexes
233
separable prefixes such as auf: (21) a. Er ist sein Zimmer am auf he is his room at-the up He is cleaning his room.' b. *Er ist am sein Zimmer auf he is at-the his room up c. *Er ist sein Zimmer auf am he is his room up at-the
raumen. clean-iNF raumen. clean-iNF raumen. clean-iNF
The very same behavior can be seen with noun + verb combinations lending support to the claimed syntactic parallelism of these bare nouns with other elements of the verb cluster. (22) a. Er ist am Auto fahren. he is at-the car drive-iNF 'He is driving a car.' b. *Er ist Auto am fahren. he is at-the car drive-iNF The pattern in (22) is in marked contrast to the behavior of phrasal arguments such as the full NP seinen VW ('his VW). As is shown in (23), it is impossible to place the progressive marker am before the full NP:19 (23) a. ?.£> ist seinen VW am fahren. he is his VW at-the drive-iNF 'He is driving his VW.' b. *Er ist am seinen VW fahren. he is at-the his VW drive-iNF The distinction between arguments internal to the verb cluster and phrasal arguments outside of it is straightforwardly captured in our analysis in terms of the distinction in selectional attributes.20 Having established empirical motivation for the kind of branching structure proposed by Hinrichs and Nakazawa, we can now address the question of what constraints govern the linearization of verbal complexes.
4. LINEARIZATION
It has been a common assumption that despite their close syntactic similarity in other respects, German and Dutch differ in the organization of the elements in the verb cluster. In particular, verb clusters in German are typically head-final, whereas those in Dutch are head-initial. This
234
Andreas Kathol
observation has prompted Rentier (1994) to use his complex-building valence attribute to set up the distinction between orderings in German vs Dutch in terms of the linear precedence (LP) constraints sketched in (24):21 (24) a. German [SYNSEM D] ]
- < [ . . . | VCOMPL < [H >]
b. Dutch [... I VCOMPL < CD >] -< [SYNSEM 0] Thus, while the configuration of clusters in German and Dutch is the same, the two languages differ in that the first places the governor after the governed subcomplex, whereas the opposite order is observed in Dutch. However, this formulation is obviously too simplistic. As the examples of Oberfeldumstellung in (14) demonstrate, there are indeed instances in which the governor precedes, rather than follows, its verbal complement in German.22 Moreover, the treatment of Dutch in terms of (24b) is equally simplistic as it totally ignores the phenomenon, sometimes referred to as inversion (cf. van Noord and Bouma, 1995), in which the order between governor and governee exhibits left-headedness. This is a frequently attested possibility for tense auxiliaries such as the perfectivizer hebben (25), and to a lesser extent also seen with modals (26), such as willen ('want'): (25) a. dat Joop de krant heeft gelezen. that Joop the newspaper has read 'that Joop has read the newspaper' b. dat Joop de krant gelezen heeft. that Joop the newspaper read has (26) a. dat Joop de krant wil lezen. that Joop the newspaper wants read 'that Joop wants to read the newspaper' b. dat Joop de krant lezen wil. that Joop the newspaper read wants It is therefore necessary to make reference to more fine-grained distinctions than the government relationships alone. As a second approximation toward an adequate account of ordering relations within the verb cluster, we adopt a solution which is essentially isomorphic to the one Hinrichs and Nakazawa propose to account for the Oberfeldumstellung cases in (14). Their central idea is to record on nonfinite verbs the possible relative position of any embedding governor. This is
Verbal Complexes
235
achieved via the binary-valued head feature FLIP. Negative specification indicates canonical government to the left, whereas a positive value requires government by a "flipped," i.e., preceding, governor. To avoid unwanted connotations regarding markedness or exceptionality, we replace FLIP with the head attribute G(O)V(ERN)OR whose value directly indicates the required placement of a governor.23 Thus, [GVOR -> ] records on a verb that, in a verb cluster, its governor has to appear to the right, whereas [GVOR <-] encodes the opposite ordering relation. The correlated LP constraints are then straightforward and are given in (27): (27)
a. [SYNSEM DQ [. . . | GVOR
-»]]
- < [ . . . I VCOMPL < E >]
b. [... I VCOMPL < E > ] -< [SYNSEM 03 [... I GVOR «- ]] Apart from being able to make the requisite fine-grained distinctions regarding order, this formulation has the obvious advantage over Rentier's system that German and Dutch can be treated on a par with respect to the involved LP constraints, while differing in the possible values that (classes of) lexical items may bear. For instance, let us assume that nonfinite main verbs in Dutch are underspecified in terms of their value for GVOR (hence as the GVOR value we have dir, which is the supersort of the atomic sorts -> and <-) while German main verbs are obligatorily [GVOR -»]: (28) Classification of nonfinite main verbs a. German [. . . | HEAD I GVOR
-> ]
b. Dutch [. . . | HEAD | GVOR dir]
We then have an immediate account of why inversion with main verbs is never possible in Standard German; cf. (29): (29) a. *dqfi that b. *daB that
Lisa Lisa Lisa Lisa
die Zeitung the newspaper die Zeitung the newspaper
hat gelesen. has read will lesen. wants read
Turning to nonmain verbs in German, we note that their nonfinite forms display different behavior with respect to the positioning of a higher governor. In particular infinitival modals and a few other predicates which themselves take a bare-infinitival complement are underspecified regarding their GVOR value. (30)
236
Andreas Kathol
This then gives rise to an analysis of Oberfeldumstellung that is isomorphic to Hinrichs and Nakazawa's initial proposal. Note in particular that since GVOR is a head feature, the value chosen on the infinitival auxiliary will also be the same borne by the subcomplex headed by a verb such as konnen. The resulting distribution of values is given for the canonical order in (3la) and the Oberfeldumstellung case in (31b):
(31) a.
b.
As Hinrichs and Nakazawa further observe, the account given so far ignores the fact that if both haben and werden occur in a sentence, their properties with respect to Oberfeldumstellung interact. Hence, our previous discussion could be taken to suggest that cluster-final placement is always possible, whereas Oberfeldumstellung is rather restricted, as is evidenced for instance by the ungrammaticality of (29). In particular then, if the verbal complement of werden contains an element such as haben which also displays variation in the linear relations with the complement, the positioning of werden should nevertheless be independent. However, this is not the case, as the examples in (32) show. Only the variant (32a) is grammatical, with its structure shown in (33): (32) a. dafi er die Lieder wird [haben singen konnen]. that he the songs will have sing can 'that he will have been able to sing the songs' b. *daB er die Lieder [haben singen konnen] wird. that he the songs have sing can will
Verbal Complexes
237
Hinrichs and Nakazawa (1994a) propose to account for this fact by assuming that haben takes on the trigger properties of the complement it embeds; hence the infinitival form of haben must satisfy the following description:24 (34)
This means that whenever haben is involved in an Oberfeldumstellung construction, the whole constituent will itself require a higher governor to precede. Otherwise, it will occasion the governor verb to occur finally, as in (35):25 (35)
Hinrichs and Nakazawa's (1994a) account can be shown to correctly capture all the grammatical permutations of verbs and their verbal complements listed in Bech (1955:63). Using his convention of indicating depth
238
Andreas Kathol
of embedding by increasing subscripts, these can be listed as in (36):26
(36)
For comparison, this table also lists two ungrammatical cases in which the highest verb "ignores" the reversal of the governee-governor order occurring in its complement (cf. (32b)). One important aspect of Hinrich and Nakazawa's system, which can be seen rather clearly in Bech's tabular representation (36), is the fact that the sequence of verbs can be divided into at most one subsequence of monotone increasing levels of embedding and exactly one monotone decreasing subsequence of levels of embedding—separated in (36) by means of "|". The first of these subsequences corresponds to what Bech calls Oberfeld ('upper field'), while the second is his Unterfeld ('lower field'), hence the terminology Oberfeldumstellung ('displacement into the upper field'): (37)
In the case of i = n — 1, the Oberfeld is empty; that is, no governor occurs in reversed position. In the next section we will encounter evidence that challenges the idea that verbal complexes can always be correlated with a linear partitioning of the set of elements that form part of the verb cluster. 4.1. Discontinuous Verbal Complexes Bech's theory of order within the verb cluster and its implementation in HPSG terms by Hinrichs and Nakazawa discussed above is to some extent a sanitized version of the facts. For one thing, it ignores the possibility of nominal elements occurring among cluster elements in what are commonly referred to as V-Projection Raising constructions such as in (38):27, 28 (38) daft Lisa dem Jungen wird das Buch geben wollen. that Lisa the boy-DAT will the book give want 'that Lisa will want to give the book to the boy' In this study, however, we focus on another set of apparent counterexamples to the implicit claim that governors may never occur inside the
Verbal Complexes
239
sequence of cluster elements corresponding to the governed verbal complex. There exists substantial evidence that this view is too strict, as such interspersals do indeed occur in a number of dialects.29 Meurers (1994) gives a number of grammatical examples in which the preposed auxiliary's placement is within the Unterfeld. Examples of this construction, referred to by Meurers as Zwischenstellung ('intermediate placement') or Verbal Complex Split, are given in (39) (Meurers, 1994): (39) a. zu dem Zeitpunkt an dem ich mich entscheiden3 hdttel mussen2. at the point at which I me decide had must 'at the point at which I should have made a decision' b. daft er das Examen bestehen3 wird/hat1 konnen2. that he the exam pass will/has can 'that he will be/has been able to pass the exam' c. daft ihm die Entscheidung nicht durchgehen3 hdtte2 durfen2. that him the decision not slip had should 'that the decision should not have passed him by' d. daft es so kommen3 hdttel mussen2. that it so come had must 'that it had to come to this' According to Meurers, the presence of two flipped auxiliaries leads to the following pattern of grammaticality:30 (40) daft er das Buch . . . that he the book a. wird1 haben2 kopieren5 Iassen4 mussen^. will have copy let must b. wirdl kopierens haben2 Iassen4 mussen3. c. ?wjrrf1 kopieren5 Iassen4 haben2 mussen^. d. *wird1 kopieren5 Iassen4 mussen3 haben2. e. kopieren5 wird1 haben2 lassen4 mussen3. f. Ikopieren5 wird1 Iassen4 haben2 mussen3. g. *kopieren5 wird1 lassen4 mussen3 haben2. h. kopieren5 lassen4 wirdl haben2 mussen3. i. *kopieren5 lassen4 wird1 mussen3 haben2. ]. *kopieren5 lassen4 mussen2 wirdl haben2. Furthermore, as we noted earlier, Standard Dutch displays the inversion order with tense auxiliaries, that is, the reversal of the typical governor-governee serialization. As has been pointed out, for instance, by van Noord and Bouma (1995), inversion is also possible with nonfinite forms of the tense auxiliary hebben. This means that in addition to the
240
Andreas Kathol
canonical order in (41a), the examples in (41b,c) are possible as well: (41) a. dat Jan dit boek moetl hebben2 gelezen3. that Jan this book must-FiN have-iNF read-iNF 'that Jan must have read book' b. dat Jan dit boek moetl gelezen3 hebben2. that Jan this book must-FiN read-psp have-iNF c. dat Jan dit boek gelezen3 moet1 hebben2. that Jan this book read-PSP must-FiN have-iNF Under the assumption that the same government and constituency relations are involved in these examples as are in the German cases, the cluster in (41b) is structurally isomorphic to German Oberfeldumstellung, whereas (41c) is a variant of the German Zwischenstellung cases in (39). Thus, a strictly phrase-structure-based view would predict that only (41b) should be possible, while the governor's occurrence internal to the governed cluster in (41c) should be illicit. However, the judgments in these cases are the exact reversal of the predictions of the theory. As van Noord and Bouma (1995) point out, sentences such as (41c) are judged to be grammatical in all dialects of Dutch; those similar to (41b) appear to be rather limited in acceptance and are possible for the most part only in Flemish dialects. A rather similar observation can be made with respect to the placement of separable prefixes. In contrast to most German dialects, separable prefixes in Dutch may "float" to earlier positions away from the main verb they cooccur with. For instance the prefix aan in aanspreken ('address') may occur not only in immediate adjacency with the main verb spreken (42a), but may also be separated from the latter by one or more intervening auxiliaries; cf. also den Besten and Edmondson (1983:193):31 (42) a. dat Jan Marie zou hebben aangesproken. that Jan Marie would have FREF.spoken 'that Jan would have spoken to Marie' b. dat Jan Marie zou aan hebben gesproken. that Jan Marie would PREF have spoken c. dat Jan Marie aan zou hebben gesproken. that Jan Marie PREF would have spoken Since prefix + verb combinations in West Germanic exhibit many of the properties of partial verbal clusters (for instance with respect to stress assignment, nonintervention of adverbs), these cases again arguably involve the discontinuous linearization of the verbal complement of a governor inside the verb cluster (here, hebben)—contrary to the predictions made by the Hinrichs-Nakazawa model of the verbal complex.
Verbal Complexes
241
Various solutions suggest themselves to the problem just encountered. In the tradition of transformational analyses, following Evers (1975), one could propose some manipulation of the constituent structure that achieves the desired linearization possibilities. However, since the ordering alternations are an exclusively local phenomenon, the existing HPSG mechanisms for nonlocal dependencies are arguably ill-suited for a reanalysis-based approach. Therefore a different line will be pursued here, which also has the significant conceptual advantage of allowing for a uniform analysis of government and constituency relations throughout West Germanic. In particular, we propose to analyze the problematic cases at face value as involving a discontinuous (partial) verb complex. As the requisite framework we adopt a variant of HPSG that possesses order domains, developed in Kathol (1995a), which in turn is based on ideas first advanced by Reape (1993). Order domains, appropriate for objects of sort sign, can be understood as totally ordered lists of information bundles which each contain phonological and syntactic-semantic information.32 Simplifying somewhat, at each application of a combinatorial schema, a corresponding order domain is built up involving the domains of the constituents thus far derived. Depending on general principles of domain construction, two basic scenarios need to be distinguished. The first possibility is that a constituent is entered into the resulting domain of the mother as a single phonologically encapsulated informational chunk. Consider for instance the combination of a verb and an NP object in (43): (43)
Here, the domain of the VP consists of two elements, one for the head (sieht) and one constructed from the complement (die Rose). In the domain resulting from the verb-object combination, the internal linear composition of die Rose can no longer be referenced. In the terminology of Kathol (1995a), the object has been domain-inserted into that of the head. As with the construction of local trees in G/HPSG, no assumption about order is mi'de at this level; instead, this will be the task of general
242
Andreas Kathol
LP constraints. As a result, LP constraints such as (27) must be thought to order domain elements, as opposed to arranging the PHON values of daughters in local trees. As the second domain construction possibility, the domain of the mother can be computed as the domain union obtained from the domains of the daughters. Domain union is equivalent to the shuffle operation; that is, given two lists A and B, a shuffling of A and B will contain the same members as A and B combined and preserve the ordering originally holding among members of each of the component lists. What is not guaranteed, however, is that A and B will necessarily be represented as contiguous sublists of the resulting list. The notion of shuffle was first introduced by Reape (1993) under the name sequence union for the analysis of nonfinite complementation structures in West Germanic. However, as will be discussed in more detail in section 5, his conception of constituency is rather different from the one argued for here. The heart of our approach to the linearization of verbal complexes lies in the assumption that the binary governor-governee combinations do not involve the insertion of the governee complex into the domain of the mother node—as is in effect assumed in Hinrichs and Nakazawa's treebased model. Instead, if the combination involves domain union, then as a result, the internal components of the governee will still be "visible" for the purposes of linearization. As a concrete example, consider the domain construction associated with a "canonical" German verb cluster, as in (44): (44)
When finden and konnen are combined, each contributes only one domain element; hence the effect of domain union will be the same as if finden had been inserted into konneris domain. However, when the resulting
Verbal Complexes
243
subcomplex is combined with wird, there are now three placement options: before finden konnen, between finden and konnen, and following finden konnen. In the case of canonical orderings, only the last is grammatical, as required by the specification [GVOR ->] on the subcomplex as inherited from konnen. Yet, that subcomplex can no longer be referred to as a separate element within the order domain. As a result, we need to revise our LP constraints in (27) to reflect this fact by making reference not to the entire governed complex, but only its head. The following LP constraint achieves the desired effect and hence supersedes the earlier formulation in (27a):33
(45) If the governor has to follow the head of the governed verbal complex, it necessarily has to follow all elements of the verbal complex; hence reference to the head in a head-final structure has the same effect as reference to the entire selected subcluster as in (27a.) At this point, we may also want to briefly consider how linearized verbal complexes are embedded into larger clausal structures. Kathol (1995a) proposes a linear model of German clause structure which is rather close in spirit to the traditional "topological fields" model (cf. for instance Drach, 1937; Engel, 1970; Hohle, 1986).34 In that formalization, a topological field constitutes an equivalence class of elements within a clausal domain. While some of these equivalence classes have cardinality restrictions (at most one Vorfeld element, exactly one domain element instantiating linke Satzklammer) others do not have such restrictions. We may then regard the verb cluster simply as one topological field, here encoded as the subsort35 uc on elements of sort dom(ain)-obj(ect).36 An example of a topological organization for a subordinate clause is given in (46):
(46)
As was argued by Kathol (1995b) and Kathol (1995a), the linear organization can also serve as the basis for a positional, albeit non-movement-based account of verb placement variability in German. Thus, a verb-initial clause such as in (47) is distinguished from the verb-final one in (46) solely on the basis of the positional assignment of the finite verb wird to cf
244
Andreas Kathol
instead of vc: (47)
Even though in the analysis adopted here, both fronted verbs and members of the verb cluster are part of the same clausal domain, the frontal occurrence does not interact with the linear constraints operative within the verb cluster. This becomes immediately apparent if one compares the relative order possibilities as in (48): (48) a. Kann er das Buck finden? can he the book find 'Can he find the book?' b. *daB er das Buck kann finden. that he the book can find In other words, a verb in frontal position is outside the scope of any constraints on order that apply to verb cluster elements. For that reason, we need to assume that the LP constraint in (45 a) only applies to elements that are topologically marked as part of the verb cluster. The requisite refinement given in (49) yields the final version of the verb cluster LP constraints. It also lists the head-initial mirror version of that LP constraint, and it is to those structures that we turn next:37
(49)
What are the valid serializations that comply with constraint (49b)? Considering first the case of Dutch, where it is possible—in fact the unmarked option—to build verbal complexes from sequences of headinitial combinations, we notice that such situations result in the precise mirror image of the German case (44). That is, if the governor must precede the head of a head-initial structure, then this will have the same linear effect as if the governor had been linearized with respect to the entire governed complex. In (50), an example is given in which arrows link
Verbal Complexes
245
each governor and the head of the governed subcomplex:
Once we have mixed precedence requirements, the linearization-based approach is clearly more flexible than one based on deriving order from tree encodings directly. Thus consider what happens when konnen's GVOR value is instantiated as <- . Any governor will have to precede it, but there is prima facie no requirement that precedence has to be immediate and therefore the lowest governed verb, finden, may intervene between the two. This is possible so long as finden's linear relation to its governor is in accordance with its own GVOR value. As a German main verb, it requires the governor to follow—as a result, both precedence requirements are satisfied in the order domain in (51):
(51)
However, this is not the only solution to the linearization constraints among the elements of the pairs finden-konnen and wird-konnen. A second possibility that is consistent with the LP requirements involves placement of the highest governor wird immediately before the head of the governed complex, konnen. The latter in turn only has to follow the dependent verb finden, but not immediately. As a result, we obtain a situation in which the governed subcomplex is linearized in a discontinuous fashion:
(52)
246
Andreas Kathol
In fact, Meurer's examples of more complicated cases in (40) can immediately be accounted for. In each of the acceptable examples in (40), the required orders between each of the pairs kopieren-lassen, lassen-mussen, haben-mussen, and wird-haben is indeed given. In each of the ungrammatical cases, one of the required orders does not hold, namely, haben illicitly follows mussen. Now what about those dialects in which linear intrusion of a governor into a verb cluster is only marginally possible or outright bad? This is where we can exploit the shift from Hinrichs and Nakazawa's binaryvalued FLIP feature to the attribute GVOR. While we have so far only encountered -> and <- as subsorts of dir, nowhere has it been stipulated that those are the only possibilities. In fact let us assume that in addition to these values there are (at least) two more possibilities, for adjacent precedence to a higher governor to the right or left, respectively. For notational perspicuity, these values will be given here as <-» and ^ .38
(53)
Values of GVOR requiring adjacent linear relations are referenced by the correlated immediate LP constraints, notated as " -<-< " in (54):39 (54) a. [...|HEAD a [GVOR «-»]] -<x [v [VCOMPL ([HEAD Q]]>]] b. [v [VCOMPL ([HEAD CD])]] -<-< [. . .|HEAD Q] [GVOR <-*]] If in the less tolerant dialects, the linear constraints on main verbs given in (27a) are instead as in (55a), the ungrammaticality of (55b) is then straightforwardly accounted for. The main verb bestehen requires its governor to follow immediately, yet the intrusion of wird/hat makes it impossible to comply with this constraint. (55) a. [. . . | GVOR «-> ] b. *daB er das Examen bestehen3 \vird/hatl konnen2. that he the exam pass will/has can 'that he will be/has been able to pass the the exam' Furthermore, this allows us to correctly characterize the difference in discontinuity effects depending on the lexical class of governees. Thus,
Verbal Complexes
247
many German dialects are significantly less tolerant toward discontinuous prefix + verb combinations than in the case of purely verbal governees, as seen in the badness of (56b):40 (56) a. daft that 'that b. *daB that
Hans Maria wird ansprechen konnen. Hans Maria will PREF.speak can Hans will be able to talk to Maria' Hans Maria an wird sprechen konnen. Hans Maria PREF will speak can
Consequently, the specification of separable prefixes as [GVOR ^> ] in most German dialects will ensure that their linear distribution is more restricted than that of regular main verbs which for the most part appear to allow for discontinuities via [GVOR -> ]. The upshot of the present discussion is therefore that much, if not all,41 of the possible orderings in West Germanic verb clusters may be reducible to lexical variation. At the same time, however, the constituency and government relations, as well as the formal means (in particular the (immediate) linear precedence constraints) can be taken to be identical across all West Germanic dialects. As another example of how the behavior of lexical classes can be captured in the present approach, consider the ordering possibilities of argument raisers in Dutch. As we observed earlier, a tense auxiliary such as hebben may display inversion with a governed predicate even if the first is nonfinite. For this reason, both hebben gelezen and gelezen hebben are possible in the examples in (41), repeated below: (41) a. dat Jan dit boek moetl hebben2 gelezen^. that Jan this book must-FiN have-iNF read-iNF 'that Jan must have read the book' b. dat Jan dit boek moet1 gelezen^ hebben2. that Jan this book must-FiN read-PSP have-iNF c. dat Jan dit boek gelezen3 moetl hebben2. that Jan this book read-PSP must-FiN have-iNF However, if a nonfinite tense auxiliary is itself governed by a modal such as moeten, no inversion is possible, as is shown in (57): (57) a. dat Jan het boek moetl hebben2 gelezen. that Jan the book must have-iNF read-PSP 'that Jan must have read the book' b. *dat Jan het boek hebben2 moet1 gelezen. that Jan the book have-iNF must read-PSP
248
Andreas Kathol
This fact is correctly captured if nonfinite tense auxiliaries in Standard Dutch are required to comply with the partial lexical description in (58): (58)
Turning to nonfinite modals now, as van Noord and Bouma (1995) observe, their linearization possibilities are more restricted than those of tense auxiliaries. An infinitival modal such as moeten disallows inversion with a governed predicate such as lezen. As a result in the examples in (59), only the governor-governee order moeten lezen is grammatical, whereas the reverse order results in unacceptability: (59) a. dat Jan dit boek zou moeten2 lezen 3. that Jan this book would must-iNF read-iNF 'that Jan would have have to read the book' b. *dat Jan dit boek zou lezen 3 moeten 2. that Jan this book would read-iNF must-iNF c. *dat Jan dit boek lezen 3 zou moeten 2. that Jan this book read-iNF would must-iNF This suggests that nonfinite modals are required in their lexical specification to comply with the following description:42
(60) To some extent, nonfinite modals in Dutch are similar to German tense auxiliaries in Oberfeldumstellung constellations. Just as infinitival haben requires governing werden to precede it, the former precedes its infinitival subcomplex; an infinitival model like moeten also links up the ordering possibilities of both governing and governed predicates resulting in an ascending sequence of governing verbs.
5. A RADICALLY LINEARIZATION-BASED THEORY? As was mentioned earlier, the analysis of verb clusters adopted here in terms of a more flexible relation between constituency and linearization rests on the proposal for order domains made in Reape (1993) and elsewhere, whose purpose was to account for the linearization of verb clusters and their dependent phrasal arguments. However, in contrast to the present approach, Reape neither links the argument structures of the
Verbal Complexes
249
elements of a verb cluster by argument composition, nor does he assume constituent status for the elements of the verb cluster. Thus, instead of forming a complex predicate, his account relies entirely on domain construction to derive the grammatical surface strings. This means that in an example such as (61), the direct object es is taken to bear a direct, syntactically manifest relation with its head lesen, as opposed to being mediated through the argument raisers versprochen and hat: (61) daB es ihm jemand zu lesen versprochen hat that it-ACC him-DAT someone-NOM to read promised has 'that someone promised him to read it' It is then, in order to allow for the heads and their respective arguments to be separated linearly, that the notion of order domain is needed. Considering the constituents involved in such an example, we can then compare Reape's "syntax tree" in (62a) with the one advocated here in (62b):
(62) a.
250
Andreas Kathol
Reape's proposal involves successive domain formation sketched in the "domain tree" in (63):
The initial appeal of this view of nonfinite complementation is that it allows an analysis of the German constructions of interest which, notwithstanding the differences in linearization, is remarkably similar to the constituency commonly proposed for the equivalent English sentences in nontransformational approaches—with the exception of Reape's analysis of German raising as S-embedding, as will be seen shortly. In effect this represents a nonderivational implementation of Ever's (1975) idea that coherent structures in German involve the syntactic representation of underlying head-argument relations which in later stages in the derivation may become "reanalyzed" into different arboreal configurations. In Reape's proposal this second reanalysis part is supplanted by a purely linear liberation of the head-argument relations. It is straightforward to bring out the parallelism between English and German if we give the syntax tree in (62) an English order. We get the very same nonfinite VP constituents (VP 2 ,VP 3 ) as in the corresponding English sentence. It is only in the analysis of noninverted finite sentences in terms of an entirely flat structure that Reape's analysis of the German example has to part company with its English counterpart, assuming a standard NP-VP analysis in the latter case. However, considering the standard HPSG analysis for English inverted clauses, the syntax trees become again iso-
Verbal Complexes
251
morphic for both languages:
(64)
Another point worth mentioning in which Reape's assumptions differ from the ones espoused here pertains to the control of domain formation. Instead of general principles, his domain construction is controlled lexically via the attribute UNIONED. Thus, [UNIONED + ] corresponds to domain union, whereas [UNIONED - ] triggers domain insertion. This attribute plays a major role in characterizing different classes of predicates that take nonfmite complements. Besides VP, the other argument category which is eligible for domain union is s. This is the basis of Reape's analysis of raising verbs whose only argument—similar to the standard GB analysis in terms of CP—is a nonfinite sentence. As a result, we have the following typology of nonfinite complementation in German consisting of three types of verbs embedding nonfinite complements: (65) a. Raising, e.g., scheinen
b. Control, coherent, e.g., versuchen
c. Control, incoherent, e.g., bezweifeln
252
Andreas Kathol
The three classes are characterized in terms of, first, the level of projection (s vs VP) and, second, whether or not the argument's DOM value is domain-unioned. Nonfinite clauses which are not domain-unioned constitute a gap in the German paradigm in Reape's typology, as Raisingto-Object type constructions43 are analyzed in terms of an NP and a VP complement (Reape, 1993:136), quite parallel to the treatment of Raisingto-Object constructions in standard HPSG. Despite its promise of a cross-linguistically uniform typology of nonfinite complementation, however, there are a number of reasons to believe that this line of analysis is fundamentally flawed and should be abandoned in favor of an argument composition approach for German. The shortcomings can be characterized in terms of the different kinds of arguments to which Reape's typology prevents selectional access to.
5.1. Subjects in Nonfinite Complements As Reape himself notes (Reape, 1993:136), if a raising verb subcategorizes for a whole sentence, then a subject contained in that sentence will no longer be directly accessible to the embedding verb via valence features. However, whether the postulated clause contains a subject, and if so, how it is specified, determines the agreement marking on the governing verb. If there is no subject, as in (66a,b), the raising verb invariably shows third singular inflection; otherwise, it agrees with the subject in number and person (cf. (66c)): (66) a. Heute scheint gearbeitet zu werden. today seems worked to be 'There seems to be work going on today.' b. Den Mddchen scheint/*scheinen schlecht zu werden. the girls-DAT seem-3RD.SG/seem-3RD.PL ill to become 'The girls seem to be getting ill to their stomachs.' c. Du scheinst/*scheint mal wieder nichts zu verstehen. You seem-2ND.SG/seem-3RD.SG yet again nothing to understand 'You don't seem to understand anything again.' It is standardly assumed in HPSG accounts of agreement (cf. Pollard and Sag, 1994; Kathol, to appear; Kiss, 1992; and others) that subject-verb agreement can be captured in terms of a dependency between a verb's morphology and its valence properties. Other kinds of agreement, such as
Verbal Complexes
253
NP-internal concord in case, number, gender, and to some extent declension type, can be described in equally local terms. However, Reape is forced to abandon this well-established paradigm of explanation in lieu of an approach that needs to take other kinds of information into account, such as syntactic constituency. It is far from clear that anything can be gained by this move, especially since a syntax-based account of agreement has to take into account, for instance, the possibility of intermediate sentential adjuncts in the determination of the subject daughter of a clause. A somewhat more severe criticism pertains to the classification of auxiliaries such as the tense auxiliaries haben and sein. While they are characterized as vp-embedding in Reape's example, they behave in essentially the very same ways as raising verbs, including the fact that they do not require subjects: (67) a. Gestern ist gearbeitet warden. yesterday is worked been 'There were people working here yesterday.' b. Mich hat gefroren. me-ACC has frozen 'I was cold.' For this reason, raising verbs and tense auxiliaries should be characterized in terms of the same complementation class. If they are both specified as vp-embedding, then raising verbs will in effect instantiate class (65b), leaving no representatives of verbs that domain-union their nonfinite complement clauses. On the other hand, if tense auxiliaries are taken to be s-embedding, then—unless we assume lexical ambiguity according to complement type—another problem looms, as was first noticed by Sailer (1993:20-21). In German, certain modal verbs such as wollen allow their complements to occur with infinitival perfect, i.e., with a perfect auxiliary; consider for instance (68): (68) weil er (schon langst) gekommen sein wollte. because he already come-PSP be-iNF wanted-FiN 'because he claimed to have come already' In such contexts, the modal takes on a meaning similar to English 'claim,' which requires the subject to be accessible to establish the obligatory link between the utterer of the claim and the subject of the state of affairs thus claimed to hold. However, that subject is unavailable to the modal if the tense auxiliary is s-embedding. This means that the grammatical sentence
254
Andreas Kathol
in (68) cannot be derived, as demonstrated in (69):
(69)
It follows, therefore, that there is little support for fully sentential complements whose constituents are liberated into the matrix clause. 5.2. Direct Objects in Nonfinite Complements Besides subjects, it can be shown that direct objects too should in principle be accessible to the embedding predicate. Evidence for this comes from a phenomenon known as "distant" or "long" passive (cf. Hohle, 1978; Kiss, 1992:222-228). In (70), the NP der Wagen is the direct object of the embedded verb zu reparieren, yet its nominative case marks it as the subject of the passivized predicate wurde uersucht: (70) IDer Wagen wurde zu reparieren uersucht. the car-NOM was to repair tried 'Someone tried to repair the car.' There is significant variation in the acceptability of such constructions, both across different speakers and for different predicates. A possible line of explanation for this may lie in the fact that, unlike more typical instances of passive, in cases like (70) the passivized constituent der Wagen does not bear a direct thematic role within the predicate undergoing the passivization, zu reparieren versuchen.*4 What is important for our purposes here is that if all predicates of the uersuchen-class invariably embed VPS, as suggested by Reape, the direct object of the embedded verb (den Wagen in (70)) would never be "visible" to the passivization of versuchen. To see this, consider that if we are to account for passives in terms of a lexical rule, it can only apply to single lexical items, which means that the rule has to operate either on the valence specification of reparieren or versuchen. The infinitival, active morphology on the first clearly makes this possibility implausible, especially in the absence of a modal interpretation that accompanies combina-
Verbal Complexes
255
tions of sein and zw-infinitive in "modal passive" constructions: (71) Der Wagen ist zu reparieren. the car-NOM is to repair 'The car can be/has to be repaired.' On the other hand, if control verbs such as versuchen are taken to only take VP complements, arguments within that VP become inaccessible to valence changing operations on argument structures, rendering the nominative on the NP der Wagen inexplicable. That distant passives exist at all makes a good case for why the arguments of embedded predicate should—at least in principle—be able to become arguments of the governing verb. Reape's typology on the other hand, very clearly predicts, wrongly, that distant passives should not be grammatical under any circumstance. In contrast, as Kiss first pointed out, argument composition does yield a straightforward account. Suppose for the moment that we are to treat passives by means of a lexical rule along the lines in Pollard and Sag (1987:215):
(72)
Among the arguments that versuchen will attract from its complement zu reparieren will be the accusative NP. But then, if versuchen with its expanded valence specification "is fed" to a lexical rule such as (72), the accusative complement will undergo the case alternation and be realized as the nominative der Wagen. 5.3. Indirect Objects in Nonfinite Complements Once the case is made for cluster-internal arguments being accessible to the embedding predicate, the question arises whether other kinds of complements can also be shown to participate in valence changes that are not strictly localized within the hypothesized VP. It is argued here that there does indeed exist indirect, yet highly suggestive evidence in favor of argument composition involving indirect (dative) objects.
256
Andreas Kathol
Consider the phenomenon of "recipient passives" or "dative passives," i.e., a type of dative-nominative alternation extensively discussed in Kathol (1994) and Webelhuth and Ackerman (1992): (73) a. Ich schenke dem Jungen ein Buch. I present the boy-DAT a book-Ace 'I present the boy with a book.' b. Der Junge bekommt/kriegt/erhalt (von mir) the boy-NOM receives by me ein Buch geschenkt. a book-Ace presented 'The boy is presented with a book by me.' Such sentences involve a case alternation that affects the dative argument, unlike the case of the usual passive, where it is the direct (accusative) object that undergoes an ACC/NOM alternation. As has been argued in Kathol (1994) (cf. also Webelhuth and Ackerman, 1992), a treatment of passive in terms of lexical rules manipulating the participle's argument structure similar to the one in (72) will necessitate diacritics (or "invisible affixes") to differentiate the participles that can occur in either the regular or the dative passive along the following lines: (74) a. werden
b. bekommen
This is so because if the respective passive auxiliary (i.e., werden in the ACC/NOM alternation and bekommen, kriegen, or erhalten in the DAT/NOM alternation) only selects the complement according to its participial morphology (psp) it will be impossible to rule out the wrong combination of auxiliary and participle, as in (74): (75) a. *Der Junge wird ein Buch gegeben. the boy-NOM is a book-Ace given b. *Das Buch bekommt dem Jungen geschenkt. the book-NOM receives the boy-DAT presented In other words, without diacritics like pas-a and pas-d, one cannot tell which of the two argument alternations has taken place. The alternative
Verbal Complexes
257
solution suggested in Kathol (1994) is to make the passive auxiliaries the "locus" of the case alternation and to extend the notion of argument composition, allowing for changes on the list of arguments attracted from the governed participle. This move then obviates the need for diacritics and lexical rules (for a somewhat similar proposal, cf. Heinz and Matiasek, 1994).45
(76)
While it is the case that those who otherwise advocate argument composition (cf., Kiss, 1992) have proposed to treat passive in terms of lexical rules, it is only by argument composition that the alternative analysis is possible in the first place. In contrast, if we were to follow Reape's lead and only allow VP (or s) complements, an auxiliary-centered analysis would not even be an option and therefore a solution using diacritics on the participle would be inevitable. Thus, under the premise that a diacritic-free approach is to be preferred, we have an argument that not only accusative objects, but dative objects too, should—at least in principle—be accessible to governing verbs. But if both accusative and dative objects are in principle attractable, we can feel justified in assuming that one of the possible argument structures for verbs indeed involves attraction of all of the embedded verb's arguments. Another piece of evidence suggesting that this conclusion is on the right track comes from considerations of what motivates the choice of auxiliary in either of the two types of case alternation. As Webelhuth and Ackerman (1992) point out, there is an interesting correlation between the valence found with these verbs in their occurrences as passive auxiliaries and as main verbs. For instance, bekommen as a main verb is a monotransitive verb whose subject bears the thematic role of a recipient or beneficiary: (77) Der Junge bekommt ein Buch. the boy receives a book 'The book receives a book.'
258
Andreas Kathol
(78)
Now, if we compare this argument structure with the one for bekommen as a passive auxiliary in dative passive constructions such as (73), it turns out that the fully instantiated SUBCAT argument structure in the passive case is identical with the one in (78). The only difference resides in whether the accusative argument is a direct argument or is linked to an argument from an embedded verbal complement. By contrast, if we were instead to treat bekommen essentially as a copula taking a predicative (dative-)passivized VP as its complement, the associated argument structure is rather different from that of main verb bekommen. The former no longer qualifies as a transitive verb, which becomes apparent in the formulation that makes a distinction between SUBCAT and VCOMPL:
(79)
Thus, from the perspective of argument composition, the fact that bekommen and semantically similar verbs seem to have been appropriated as ingredients for the dative passive constructions is eminently plausible under an argument composition approach, as the necessary adaptations leave the fundamental (suBCAT-encoded) properties of bekommen as a transitive verb with a recipient-beneficiary subject unaffected. This also illuminates another aspect of dative passives, pointed out by Webelhuth and Ackerman (1992), namely, that they do not permit impersonal construals, in contrast to the ordinary ACC/NOM variety of passive; cf. (80b): (80) a. *weil because b. *weil because
einen Blumenstraufi a flower bouquet-Ace einen Blumenstrau.fi a flower bouquet-ACC
bekommt. receives geschenkt bekommt. given receives
This restriction is precisely what one would expect if the argument structure of the passive auxiliary is required to match that of the main verb as much as possible, as the latter in turn does not permit a missing subject either; cf. (80a).
Verbal Complexes
259
As Webelhuth and Ackerman point out, by the same token, the fact that impersonal instances of the regular ACC/NOM passive exist can be accounted for along very similar lines.46 They note that, in general, the auxiliaries involved in the two varieties of the ACC/NOM passive in German, werden and sein, do not impose any thematic requirement on their subject; in fact they do not even require a subject. For that reason, one can find them in lexically impersonal constructions such as the following, with an argument structure along the lines in (82): (81) a. well dem Jungen schlecht wird. because the boy-DAT sick becomes 'because the boy is getting ill to his stomach' b. weil dem Jungen schlecht ist. because the boy-DAT sick is 'because the boy is ill to his stomach'
(82)
It is this lack of thematic requirements on the subject that makes werden and sein perfect candidates to occur in constructions in which passive can only effect the demotion of a subject, but in which there is no direct object to be promoted to subject. As a result, the impersonal passive constructions in (83) are correctly predicted to be grammatical: (83) a. weil getanzt wurde. because danced was 'because there was dancing' b. weil zu tanzen ist. because to dance is 'because people must dance' As in the nonpassive case, the lack of subject selection on the part of the passive auxiliaries is reflected directly in the argument structure of the impersonal versions of werden /sein employed in (83), given schematically in (84):47 (84)
260
Andreas Kathol
Thus, if we follow Webelhuth and Ackerman in assuming that much of the organization of the lexicon emerges from structures motivated by and related to already existing structures, it becomes clear that only a view espousing argument composition can account for the observed nonrandomness between lexical elements in auxiliary and main verb function. 5.4. Conclusion If the arguments presented in this section hold up to further scrutiny, they strongly indicate that whether verb clusters are entirely accounted for via domain construction or also involve the valence properties in the form of (extended) argument composition is not just an aesthetic issue. Rather, the choice has important empirical ramifications having to do with locality and the motivation of lexical structures. What does not seem to be involved, however, is the question of weak generative capacity. That is, one could imagine ways to reconcile Reape's approach with the problems pointed out here by means of diacritics etc., but as this discussion has shown, the price appears quite high in comparison to allowing for the interactions of valence requirements of the sort proposed by argument composition.
6. CONCLUDING REMARKS Since the early 1990s there has been a widespread perception that the two types of analysis of the verb cluster in German—the argument composition approach and the one using order domains—bear a complementary relation to each other. This is not all too surprising considering the original conceptions of the respective authors. However, on closer inspection, this complementarity turns out in fact to be illusory. There is certainly nothing inherent in the analyses that prevents a (partial) convergence. Thus, even though the present study argues, pace Reape, for the formation of a complex predicate in a language like German, it nevertheless represents a conceptual rapprochement toward the basic intuition behind order domains. Specifically, order domains afford a view of syntax in which constituency and linear structure are not as tightly linked as in phrase-structure-based theories and hence allow for a natural implementation of the discontinuity phenomena observed in German and Dutch. As with many other areas of linguistics, progress is commonly not achieved by trying to make the observed patterns fit the Procrustean bed of a single theoretical concept but instead by trying to find a way in which
Verbal Complexes
261
seemingly disparate insights can converge. While there is obviously much more to be said about constituency and linearization of verbal complexes in West Germanic, the hope is that the current proposal will serve as a new basis for further explorations into this fascinating, albeit highly complex, research area.
ACKNOWLEDGMENTS I would like to acknowledge financial support for this article from the Dutch Organization for Scientific Research. I am grateful to Gosse Bouma, Jack Hoeksema, Gertjan van Noord, Bob Kasper, Stefan Muller, John Nerbonne, and Carl Pollard for stimulating discussions and suggestions. Native speaker judgments on the Dutch sentences from Ferdinand de Haan, Mark-Jan Nederhof, Gertjan van Noord, Rob Koeling, and Mettina Veenstra are greatly appreciated. I also would like to thank two anonymous reviewers for helpful comments. The usual disclaimers apply.
NOTES 1For instance it has been standard practice to treat raising predicates as CP-selecting, whereas tense auxiliaries have often been considered to take mere verbal projections (v° or VP) as their complements. 2 Additional complications arise with predicates such as versprechen ('promise') which feature an optional (dative) argument of their own. This does not affect the general point to be made here, namely, that argument composition plays a crucial role in the analysis of clause-final verb clusters. 3 In this paper, verb cluster will be used as a descriptive term to refer to the accumulation of verbal material in German and Dutch clauses. By contrast, the term verbal complex more specifically indicates that the verb cluster (or at least parts thereof) is also given constituent status in the syntactic analysis. 4 On the other hand, a left-branching structure is inevitable as soon as the recursive inheritance of argument raisers is disallowed. 5 However, the adoption of a flat analysis for the phrasal elements within the Mittelfeld does not automatically entail that the lack of internal grouping has to extend to the verb cluster. Both Hinrichs and Nakazawa (1994b) and Rentier (1994) assume a flat analysis together with a contoured (nonfinite) verb cluster. 6 We will, however, not engage here in a discussion of analyses in terms of flat structures, in particular Bouma and van Noord (1998), which is quite similar to the current study in empirical coverage.
262
Andreas Kathol
7
A rather similar idea is pursued by Haider (1993), who analyzes verbal complexes as recursive x° structures (komplexe Projectionsbasis ('complex projection base')) with merged argument structure. See also Bierwisch's (1990) more morphologically based account which is even more similar to Hinrichs and Nakazawa's proposal in adopting a notion of functional composition. 8 Thus, we disagree with Bierwisch (1990), who blurs the syntax-morphology distinction by treating verbal elements in German verb clusters as "pseudo-affixes." A necessary connection between morphological and syntactic wordhood is also denied by Zwicky (1990:206), who states that "being word-like syntactically does not entail being word-like morphologically." It is also possible to front phrasal arguments together with the (sequence of) nonfinite verbs, as in (i): (i)
[Das Buch finden] wird Peter konnen. the book find will-FiN Peter-NOM can 'Peter will be able to find the book.'
This phenomenon has been discussed in the literature under the heading Partial VP Fronting or Remnant Topicalization. While such constructions lie outside the scope of the current study there have been numerous proposals in the HPSG literature; see, for instance, Pollard (1996), Nerbonne (1994), and Hinrichs and Nakazawa (1994b). For more recent proposals that are congenial to, albeit formally more elegant than, the solution in Kathol (1995a), see Meurers (1996) and Muller (1996). 10 While Kiss correctly points out that there appears to be evidence that constituency is not a necessary condition for frontabiliry, we also cannot take fronted occurrence to preclude constituency. All things being equal, an account that can assimilate (13b,c) to the well-attested fronting of constituents is clearly to be preferred over a theory that has nothing at all to say about such examples. In this connection it should also be noted that Meurers's (1996) reconstruction of verb cluster fronting in Kiss' right branching approach relies on the introduction of a structure with a nonlexical ([LEX -]) left daughter, hence one that emulates the left-branching analysis and thus in effect betrays the initial concept of building verb clusters via lexeme selection. 11 However, lack of frontability may be explained due to the independent constraint against fronting finite verbs. As Hinrichs and Nakazawa (1994a:15) argue, this is only a sufficient condition; i.e., there are examples of the governing verb preceding only one infinitive. Also, this formulation requires the concept of morphological infinitive which may depart from the properties of the nonfinite form in terms of its status government. See Kathol (1995a) and the discussion in note 24 later for a proposal for relating the two notions in a systematic way. 13 Here, the feature LEX is used to indicate (quasi-)lexical status of the sort discussed above.
Verbal Complexes
263
14
Moreover, it has been taken for granted so far that, in both the canonical as well as the Oberfeldumstellung order, the governing verb bears the same grammatical relation (viz. head-argument) to the embedded verbal complex. However, there are proposals, most notably Meurers (1994), which question this assumption. Specifically, Meurers takes Oberfeldumstellung to involve head-marker structures. 15 In much of HPSG, it has become customary, following an idea first advanced by Borsley (1989), to distinguish between valence attributes for subjects (SUBJ) and complements (COMPS). None of what is said here will have any bearing on that distinction; hence we will assume the somewhat simpler version of a single subcategorization list (SUBCAT). 16 An alternative solution is proposed in Hinrichs and Nakazawa (1994b:15), where valence features have lists of signs as their value such that their version of the complement extraction rule may only remove phrasal, as opposed to word-level, objects of sort sign from the valence list into SLASH. Such a move raises important questions regarding the locality of syntactic selection, which constitutes the original motivation in Pollard and Sag (1994) for distinguishing between the sorts sign and synsem. In the absence of other compelling evidence for conflating this distinction, it seems preferable to seek solutions that maintain the locality of syntactic selection argued for by Pollard and Sag (1994). This name is slightly reminiscient of the VCOMP grammatical function proposed by Bresnan et al.'s (1982) LFG-based account of Dutch verb clusters. In terms of constituent structure, that analysis also separates the verbal element from the phrasal ones, the latter belonging to a "verbal spine." However, while the VCOMP values contain all the grammatical functions of clausal structures (SUBJ, OBJ, etc.), it is only in the semantic representation that syntactic arguments are correlated with their respective predicates under the argument composition approach. 18 This type of selection will in addition also require verbal projections to be phrasal in their valence properties; exceptions to this generalization involve verbal arguments in the Third Construction. See Kathol (1995a) and Hinrichs and Nakazawa (1998) for discussion. 19 Note that the deciding factor cannot be lexicality alone. In the absence of unary syntactic expansions or lexical underspecification regarding their wordphrasal status, bare plurals will be grouped with single words; yet they exhibit the same pattern in progressive constructions as uncontroversially phrasal NPS: (i) a. ?Er ist Autos am waschen. he is cars at-the wash-iNF 'He is washing cars.' b. *Er ist am Autos waschen. he is at-the cars wash-iNF 20
However, as an anonymous reviewer points out, there are a number of aspects which the current treatment of separable prefixes as on a par with nonfinite verbs does not yet address. For instance, while fronting of a particle together with a finite verb is expectedly ungrammatical in most cases, as demonstrated in (i), verb-
264
Andreas Kathol
particle combinations with a (semitransparent) resultative interpretation, and thus involving secondary predication, yield significantly better results; cf. (ii): (i) a. Er hat ihr das Angebot ab geschlagen. he has her the offer PRT struck 'He refused her offer.' b. *[Das Angebot ab] hat er ihr geschlagen. the offer PRT has he her struck (ii) a. Er hat ihr den Kopf ab geschlagen. he has her the head off struck 'He beheaded her.' b. ?[Den Kopf ab] hat er ihr geschlagen. the head off has he her struck 21 For a formal characterization of LP constraints see, for instance, Kasper et al. (1995). 22 One possibility which will not be discussed here is that, even in these cases, we are dealing with the regular governor-before-governee order. Apart from the many unresolved technical issues brought about by such a treatment, it is not clear what descriptive or explanatory gains such an approach would yield over the assumption that the reversal resides solely in the ordering relations. 23 Furthermore, we do not require values of GVOR to be binary. This will provide us with a way to deal with adjacency effects, to be discussed in section 4.1. 24 Here, the selected verbal complement's VFORM value is simply given as past participle in spite of the morphological realization as bare infinitive. In Kathol (1995a:224-226), it is therefore argued that a more adequate account has to distinguish instead between the morphological realization in terms of VFORM and the governed Status, given here as STATUS. In the vast majority of cases, the two values are identical; yet for Ersatzinfinitiv forms, the two differ as is shown in (i):
(0
[ [ STATUS psp 11 [. . . | H E A D [VFORM m / J j
This allows us to state the generalization that typical instances of Oberfeldumstellung all involve the contiguous sequence of two or more forms specified as [VFORM inf]. On the formulation given in (34) alone (also the one proposed by Hinrichs and Nakazawa, 1994b), the correlation between Ersatzinfinitiv and Oberfeldumstellung would be wholly accidental. 25 However, as was pointed out by an anonymous reviewer, for many speakers there also seem to be acceptable cases in which infinitival haben selects a genuine past participle but nevertheless is preceded by a governing tense auxiliary, as in (i): (i) dafi er das Lied wird gesungen haben. that he the song will sung-PSP have-iNF 'that he will have sung the song' This indicates that more refined distinctions are needed and that the structure sharing in GVOR values as shown in (34) should be restricted to Ersatzinfinitiv cases.
Verbal Complexes
265
This can be accomplished straightforwardly by exploiting the difference between STATUS and VFORM proposed in note 24. 26 The fourth pattern in the fourth column (V1V2V3|V5V4) is only possible in those dialects for which modals such as kdnnen may undergo Oberfeldumstellung. With Hinrichs and Nakazawa (1994a:16,n.4) we will assume that only the tense auxiliaries werden and haben are uncontroversial when occurring in an Oberfeldumstellung construction. 27 For discussion on how to account for V-projection raising cases, see Hinrichs and Nakazawa (1994a:27-33) and Kathol (1995a:252-259). A somewhat different proposal can be found in Hinrichs and Nakazawa (1994b). 28 There are other irregularities that can be seen with the verb cluster which will not be dealt with here. For example, as has been observed by Reis (1979:14), if zu-infinitival haben occurs cluster-initially in an Oberfeldumstellung configuration, the expression of the zu -infinitival morphology shifts from the head to the rightmost element; cf. (ia) vs (ib): (i) a. ohne zuuor Fritz [[sehen gekonnt} zu haben]. without earlier Fritz see-iNF could-PSP to have-zu-iNF 'without having been able to see Fritz before' b. ohne zuvor Fritz [haben [sehen zu konnen]]. without earlier Fritz have-iNF see-iNF to can-zu-iNF 'without having been able to see Fritz before' 29
Such constructions have often been attributed to Southern German dialects such as Franconian (cf. Kroch and Santorini, 1991:304), yet, even though I am not a native speaker of such a dialect myself, I tend to find such examples fairly good. 30 Note that the pattern in (40d) is not generally excluded, but only in those cases in which haben selects an Ersatzinfinitiv and hence must undergo Oberfeldumstellung (with respect to at least one governed verb); cf. the following example from Bech (1955:64): (i) daB man ihn hier wird1 Iiegen5 bleiben4 lassen3 konnen 2 that one him here will lie remain let can 'that one will be able to let him keep lying here' 31 Again, there is a certain amount of dialectal variation involved here. For many speakers, the structure in (42b) is at best marginal. There is a similarity between this example and the one in (41b) in that both involve the interruption of a governor-governee sequence (moet hebben and zou hebben). However, the ungrammaticality of (42) seems significantly less severe (or even nonexistent) for many speakers than that of (41b) (Gosse Bounia, personal communication). On the other hand, as an anonymous referee points out, judgments are opposite for Flemish speakers, who find (41b) impeccable, but nevertheless rate (42b) worse than (42a,c). 32 In Reape's proposal the list elements are taken to be signs. In Kathol's theory, on the other hand, domains are populated with domain-objects, that is, information bundles whose information content is significantly reduced as compared to signs. The distinction will not be significant for present purposes.
266
Andreas Kathol
33
In order to prevent the LP constraint in (45) from "accidentally" matching against the head values of an element which is not the head of the governed subcomplex, we need to assume that, in general, domain objects are distinct in their HEAD values. See Kathol (1995a:229) for discussion of this point. The topological indices used in this study are hence (in part) inspired by the terms used in the traditional literature: vf, Vorfeld ('initial field'); cf, Comp/Finite, linke Satzklammer ('left sentence bracket'); mf, Mittelfeld ('middle field'); vc, verb cluster, rechte Satzklammer ('right sentence bracket'); and nf, Nachfeld ('final field'). 35 In Kathol (1995a), the topological indices are not encoded via subsorts on domain elements, but rather as values of the attribute TOPO. The difference is insignificant for the purposes of this study. 36 Note that in the analysis pursued here, the homogeneity in linear assignment is correlated with the fact that all verb cluster elements belong to a single syntactic constituent. In contrast, elements in mf and nf may not be exhaustively contained in a single syntactic constituent. 37 Because of examples such as anfangen ('begin'), which are argument raisers and at the same time contain a separable prefix (an), the LP constraints in (49) will ultimately have to be refined to accommodate the situation in which a VCOMPL list consists of more than one element. We will leave this problem for further research. 38 This particular organization was suggested to me by Gosse Bouma and Gertjan van Noord. 39 For a proposal giving formal content to the notion of immediate precedence in HPSG, see Kathol (1995a:133-134). 40 Similarly, as an anonymous reviewer points out, floated prefixes in Dutch seem impossible if the base verb is followed by a higher governor in the inversion order: (i) *dat Jan Marie aan zou gesproken hebben. that Jan Marie PREF would spoken have This can be accounted for by requiring prefixes such as aan to be specified as [GVOR e-* ] just in case the nonfinite base verb (gesproken) is [GVOR -> ]. 41 However, we currently have little insight to offer as to the nature of the unacceptability of the sequences in (41b) moet gelezen hebben and (42b) zou aan hebben gesproken. One possible line to pursue may be based on the observation made in Abraham (1994:42) that elements bearing a nuclear accent generally seem to have a strong tendency to occur at the left or right edge of the verb cluster. Both in (41b) and in (42b), this Fokusrandprinzip ('principle of peripheral focus') is violated. However, this still does not address the (diverging) distinctions in acceptability reported for different varieties of Dutch for such examples. 42 This requirement is too strong to accommodate examples such as (ib) from Johnson (1988:118-119), in which a higher finite governor follows, rather than precedes, a cluster headed by a nonfinite modal. However, examples of this type are uniformly rejected by all of my informants. (i) a. dat hij het [heeft [kunnen zien]]. that he it has can-iNF see-iNF 'that he has been able to see it'
Verbal Complexes
267
b. dat hij het [[kunnen zien] heeft] that he it can-iNF see-iNF has 'that he has been able to see it' 43
These are typically referred to among German grammarians as "Acl" (for accusativum cum infinitivo 'accusative with infinitive'). 44 For discussion on this point, see Kiss (1992:226). 45 Here, we make the simplifying assumption that DAT-NOM passives should only be possible if the verb in addition takes an accusative object. As Webelhuth and Ackerman (1992) observe, the acceptability of accusative-less instances as in (i) exhibits a great deal of speaker- and lexeme-dependent variation. (i) a. Ich helfe dem Jungen. I help the boy-DAT 'I help the boy.' b. DerJunge bekommt geholfen. the boy-NOM receives helped 'The boy is helped.' 46
For accounts of impersonal passives in HPSG, see for instance Pollard (1994) and Kathol (1994). The former proposes a uniform account for both personal and impersonal variants of passive. 47 The issue of finding a uniform characterization of passive auxiliaries occurring in personal and impersonal passives will have to remain unresolved here.
REFERENCES Abraham, W. (1994). Kaynes Asymmetriehypothese und die Abfolge im V-Komplex. In J.-W. Zwart (ed.), Studies in Minimalism and Kayne's Antisymmetry Hypothesis, Groninger Arbeiten zur Germanistischen Linguistik, Vol. 37, 19-46. Groningen. Baker, K. (1994). An integrated account of "modal flip" and partial verb phrases fronting in German. In Proceedings of the 30th Regional Meeting, Chicago Linguistic Society. Bech, G. (1955). Studien uber das deutsche Verbum Infinitum. Danske Historiskfilologiske Meddelelser 35, 2. Bierwisch, M. (1990). Verb cluster formation as a morphological process. In G. Booij and J. van Marie (eds.), Yearbook of Morphology, 173-199. Foris Publications, Dordrecht. Borsley, R. (1989). An HPSG approach to Welsh. Journal of Linguistics 25, 333-354. Bouma, G. and G. van Noord (1998). Word order constraints on verb clusters in German and Dutch. In E. Hinrichs, A. Kathol, and T. Nakazawa (eds.), Complex Predicates in Nonderivational Syntax. Syntax and Semantics, Vol. 30, 43-72. Academic Press, San Diego.
268
Andreas Kathol
Bresnan, J., R. Kaplan, S. Peters, and A. Zaenen (1982). Cross-serial dependencies in Dutch. Linguistic Inquiry 13, 613-635. Chung, C. (1993). Korean auxiliary verb constructions without vp-nodes. In S. Kuno, J. Whitman, Y.-S. Kang, I.-H. Lee, J. Maling, and Y. J. Kim (eds.), Harvard Studies in Korean Linguistics V, 274-286. Hanshin, Seoul. (Proceedings of the 1993 Workshop on Korean Linguistics.) Chung, C. (1998). Argument composition and long-distance scrambling in Korean: An extension of the complex predicate analysis. In E. Hinrichs, A. Kathol, and T. Nakazawa (eds.), Complex Predicates in Nonderivational Syntax. Syntax and Semantics, Vol. 30, 159-220. Academic Press, San Diego. den Besten, H., and J. Edmondson (1983). The verbal complex in continental West Germanic. In W. Abraham (ed.), On the Formal Syntax of the Westgermania. Papers from the Third Groningen Grammar Talks, 155-216. Benjamins, Amsterdam/Philadelphia. Drach, E. (1937). Grundgedanken der deutschen Satzlehre. Diesterweg, Frankfurt. Engel, U. (1970). Regeln zur Wortstellung. In U. Engel (ed.), Forschungsberichte des Instituts fur deutsche Sprache Mannheim, Vol. 5, 1-148. Gunter Narr, Tubingen. Evers, A. (1975). The Transformational Cycle in Dutch and German. Dissertation, University of Utrecht. Haider, H. (1993). Deutsche Syntax—Generativ. Vorstudien zur Theorie einer projektiven Grammatik. Gunter Narr, Tubingen. Heinz, W., and J. Matiasek (1994). Argument structure and case assignment in German. In J. Nerbonne, K. Netter, and C. Pollard (eds.), German in HeadDriven Phrase Structure Grammar, 199-236. CSLI Publications, Stanford, CA. Hinrichs, E., and T. Nakazawa (1990). Subcategorization and VP structure in German. In S. Hughes and J. Salmons (eds.), Proceedings of the Third Symposium on Germanic Linguistics. Benjamins, Amsterdam. Hinrichs, E., and T. Nakazawa (1994a). Linearizing finite AUX in German verbal complexes. In J. Nerbonne, K. Netter, and C. Pollard (eds.), German in Head-Driven Phrase Structure Grammar, 11-38. CSLI Publications, Stanford, CA. Hinrichs, E., and T. Nakazawa (1994b). Partial VP and split NP topicalization in German: An HPSG analysis. In E. Hinrichs, D. Meurers, and T. Nakazawa (eds.), Partial-VP and Sptit-NP Topicalization in German—An HPSG Analysis and Its Implementation. Arbeitspapiere des SFB 340, Vol. 58. Hohle, T. (1978). Lexikalistische Syntax: Die Aktiv-Passiv-Relation und andere Infinitkonstruktionen im Deutschen. Niemeyer, Tubingen. Hohle, T. (1986). Der Begriff "Mittelfeld," Anmerkungen uber die Theorie der topologischen Felder. In Akten des 7. Internationalen Germanisten-Kongresses, Gottingen 1985, Vol. 3, 329-340. Tubingen. lida, M., C. Manning, P. O'Neill, and I. Sag (1994). The Lexical Integrity of Japanese Causatives. Presented at the LSA 1994 Annual Meeting, Boston. (Unpublished manuscript, Stanford University.) Johnson, M. (1988). Attribute-Value Logic and the Theory of Grammar. CSLI Lecture Notes, No. 16. CSLI Publications, Stanford, CA.
Verbal Complexes
269
Kasper, R. T., A. Kathol, and C. Pollard (1995). A Relational Interpretation of Linear Precedence Constraints. Unpublished manuscript, Ohio State University. Kathol, A. (1994). Passives without lexical rules. In J. Nerbonne, K. Netter, and C. Pollard (eds.), German in Head-Driven Phrase Structure Grammar, 237-272. CSLI Publications, Stanford, CA. Kathol, A. (1995a). Linearization-Based German Syntax. Dissertation, Ohio State University. Kathol, A. (1995b). Verb-'movement' in German and topological fields. In A. Dainora, R. Hemphill, B. Luka, B. Need, and S. Pargman (eds.), Papers from the 31st Regional Meeting, Chicago Linguistic Society, 231-245. Chicago Linguistic Society, Chicago. Kathol, A. (to appear). Agreement and the syntax-morphology interface in HPSG. In R. Levine and G. Green (eds.), Readings in HPSG. Cambridge University Press, Cambridge, UK. Kiss, T. (1992). Infinite Komplementation: Neue Studien zum deutschen Verbum Infinitum. Dissertation, Bergische Universitat-Gesamthochschule Wuppertal. Kroch, A. S., and B. Santorini (1991). The derived constituent structure of the West Germanic verb-raising construction. In R. Freidin (ed.), Principles and Parameters in Comparative Grammar, 269-338. MIT Press, Cambridge, MA. Meurers, W. D. (1994). A Modified View of the German Verbal Complex. Presentation given at the 1994 HPSG workshop in Heidelberg. Meurers, W. D. (1996). German Partial-VP Fronting Revisited—Back to Basics. Unpublished manuscript, University of Tubingen. (Paper presented at the Third International Conference on HPSG, Marseille, France.) Miiller, S. (1996). Yet another paper about partial verb phrase fronting in German. In Proceedings of COLING-96, Copenhagen, 800-805. Nerbonne, J. (1994). Partial verb phrases and spurious ambiguities. In J. Nerbonne, K. Netter, and C. J. Pollard (eds.), German in Head-Driven Phrase Structure Grammar, 109-150. CSLI Publications, Stanford, CA. Pollard, C. J. (1994). Toward a unified account of passive in German. In J. Nerbonne, K. Netter, and C. Pollard (eds.), German in Head-Driven Phrase Structure Grammar, 273-296. CSLI Publications, Stanford, CA. Pollard, C. J. (1996). On head non-movement. In H. Bunt and A. van Horck (eds.), Discontinuous Constituency, 279-305. de Gruyter, Berlin/New York. Pollard, C. J., D. Levine, and R. T. Kasper (1993). Studies in Constituent Ordering: Toward a Theory of Linearization in Head-Driven Phrase Structure Grammar. Grant proposal to the National Science Foundation, Ohio State University. Pollard, C. J., and I. A. Sag (1987). Information-Based Syntax and Semantics, Vol. 1. CSLI Lecture Notes Series No. 13. CSLI Publications, Stanford, CA. Pollard, C. J., and I. A. Sag (1994). Head-Driven Phrase Structure Grammar. CSLI Publications, Stanford, CA, and University of Chicago Press, Chicago. Reape, M. (1993). A Formal Theory of Word Order: A Case Study in West Germanic. Dissertation, University of Edinburgh.
270
Andreas Kathol
Reis, M. (1979). Ansatze zu einer realistischen Grammatik. In K. Grubmiiller, E. Hellgart, H. Jellissen, and M. Reis (eds.), 1-21. Niemyer, Tubingen. Rentier, G. M. (1994). A lexicalist approach to Dutch cross dependencies. In K. Beals, J. Denton, E. Knippen, L. Melnar, H. Suzuki, and E. Zeinfeld (eds.), Papers from the 30th Regional Meeting of the Chicago Linguistic Society, 376-390. Chicago Linguistic Society, Chicago. Sailer, M. (1993). Zuruckin die Felder. 3 Ansatze zur Beschreibung infiniter Konstruktionen in Deutschen. Unpublished manuscript, University of Tubingen. Stiebels, B., and D. Wunderlich (1994). Morphology feeds syntax: The case of particle verbs. Linguistics 32, 913-968. van Noord, G., and G. Bouma (1995). Dutch Verb Clustering without Verb Clusters. Unpublished manuscript, University of Groningen. Webelhuth, G., and F. Ackerman (1992). Form and Function: The German Passive. Unpublished manuscript, University of North Carolina and University of California, San Diego. Zwicky, A. (1990). Syntactic words and morphological words, simple and composite. In G. Booij and J. van Marie (eds.), Yearbook of Morphology, Vol. 3, 201-206. Foris Publications, Dordrecht.
A DEDUCTIVE ACCOUNT OF FRENCH OBJECT CLITICS ESTHER KRAAK Utrecht Institute of Linguistics OTS Utrecht University Utrecht, The Netherlands
1. INTRODUCTION
Over the past few years, there has been a development in the field of computational linguistics which has led to the emergence of a so-called "grammar logic." In this framework, insights from two logical traditions have been combined: Linear Logic (Girard, 1987) has provided the notion of a resource-conscious logic, where each assumption can and must be used only once, while from the categorial line of work initiated by Lambek (1958) has come the idea of structuring the assumptions in various ways. A detailed account of the grammar logic program can be found in Moortgat (1997). The purpose of the grammar logic is to provide us with a set of constants (connectives) of grammatical reasoning which are thought to underlie our linguistic abilities. Using these constants, analyses of a wide range of phenomena from a variety of languages can be construed in an entirely deductive way. In this article we present a complex test case for the framework, in the form of the intricate data on object clitics in French. The structure of the article is as follows: In section 2 we introduce the base logic on which our grammar fragment will be built. Section 3 then gives a short overview of the empirical facts which the fragment intends to cover, and in section 4 the base logic is shown to be adequate for analyzing Syntax and Semantics, Volume 30 Complex Predicates in Nonderivational Syntax
271
Copyright © 1998 by Academic Press All rights of reproduction in any form reserved. 0092-4563/98 $25.00
272
Esther Kraak
the most elementary of these constructions. In sections 5 and 6 the grammar is further refined, to deal with the more complex data concerning order requirements for multiple clitics and interactions between clitic pronouns and complex verbal constructions. Finally, in section 7 the approach presented here is compared with other analyses of the same phenomena.
2. CATEGORIAL LOGICS This section starts with an introduction to the fundamentals of categorial logic, which constitutes the basis of the grammar fragment that will be developed. In section 2.2, two key features of this basic formalism are presented that will play an important role in the analysis of French object clitic behavior: the use of unary control operators and of multiple modes of linguistic composition. To conclude, section 2.3 shows the working of the base logic with an example. 2.1. The Base Logic Consider the language ^"of category formulas, obtained by closing a set of atomic types s/ (the basic categories or basic types) under binary connectives /, •, \:
In words, a category A*B represents the composition of an element of type A with an element of type B. The slash can be regarded as the leftward- or rightward-looking "dual" of the product operator •: categories A/B and B\A express incompleteness with respect to composition. Elements of those categories need to be composed with an element of type B, on the right- or left-hand side, respectively, to yield an expression of type A. Because modal logic is ideally suited for reasoning about structures, in our case structured configurations of pieces of linguistic information, we interpret these category formulas in a Kripke style modal frame (W, R3y. Here, W represents the set of linguistic resources or signs, i.e., units of linguistic form and meaning which belong to a certain syntactic category and represent the words and phrases of a language. The ternary accessibility relation R3 in this frame represents linguistic composition, R3zxy meaning that the signs x and y can be composed to yield z. Adding a valuation v to the frame turns it into a model. This valuation takes
French Object Clitics
273
atomic categories to subsets of W (namely, those subsets containing all words and phrases of that category) and satisfies the following in case of compound formulas:
Intuitively, the relation R3 holds between two parts (x and v) and a whole (z) composed of those parts. In case of a slash category, one of the parts is unsaturated: in order to form a whole z of type C, an unsaturated part x of type C/B (resp., y of type A\C) must be composed with another part y to its right (resp., x to its left) which belongs to its argument category B (resp., A). To characterize the derivability properties of the categoryforming operators • and /, \, we will make use of the concept of residuatlon. Let j/ = (A, and 38 = be partially ordered sets. Then a pair of functions (/, g), with /: A -> B and g: B -» A is called residuated if the following holds:
For our purpose, the sets j/ and £% are the same: A and B represent the set of category formulas ^ and in both cases the ordering is the derivability relation ->. If we assume /= — *B and g = —/B, with some fixed B e ^ and take x and y in the definition of residuation to be A and C (e &~~), respectively, the above biconditional gives us
Analogously, if we assume f = A* — and g = A\- , with some fixed A e ^ and take x and y to be B and C (e &*), respectively, we get:
Combining these results yields the following laws of residuation, defining the derivability properties of the product and its left- and right-residual operators, the slashes:
The essential soundness and completeness result states that A -> B iff v(A) c u(B), for all valuations on all frames. These residuation laws, together with the inherent characteristics of the derivability relation, reflexivity (identity) and transitivity, constitute the
274
Esther Kraak
grammatical base logic. It is known as NL for nonassodatiue Lambek calculus, since it does not include the structural postulate of associativity, nor any other restriction. From this pure logic of residuation a whole landscape of categorial logics can be obtained, by relaxing its sensitivity for linguistically relevant structural parameters such as immediate dominance (constituent sensitivity) and linear precedence (order sensitivity). One way this can be done is by adding the structural postulates of Associativity and Commutativity to NL. Whereas in the base logic the interpretation of the accessibility relation R3 is completely free, adding these structural postulates introduces constraints on its interpretation. Notice that the more restrictions we impose on R3, the less discriminating it becomes. Below are the structural postulates and frame conditions that allow the linguistic resources to be restructured and reordered (permuted), respectively (for all u, x, y, z e W): Associativity A*(B*C) **(A • £)• C 3t.R3zxt & R3tyu **lu.R3uxy & R3zvu Commutativity A • B *+B*A R3zxy **R3zyx
Adding Associativity to NL gives us the Lambek system L; adding Commutativity to NL yields NLP (P for Permutation). If we adopt both postulates we get the system LP, which is also known as the Lambek-van Benthem calculus. Throughout this paper we will be using a system of natural deduction1 to present derivations of linguistic constructions; a successful derivation is called a proof. The things we want to prove are statements of the form F \- A, pronounced 'T derives A," where F is a structured configuration of category formulas and A is a single category formula. For each of the logical connectives /, •, \ there are two rules: an Introduction rule that introduces the connective to the right of the turnstile, and an Elimination rule for the removal of the connective.2 In the natural deduction rules below, the category formulas are annotated with semantic (lambda) terms, following the Curry-Howard correspondence between natural deduction proofs and terms of the A-calculus (note that Elimination corresponds to functional application, and Introduction to functional abstraction):
French Object Clitics
275
For instance, a rule such as I/ should be read as follows: if a structure F composed with a category formula B to its right-hand side derives a category formula A, then that structure F alone derives a complex category formula A/B. As we remember from the residuation laws, something is of category A / B if combining it with something of category B immediately to its right yields something of category A. This is expressed by the E/ rule, a version of the well-known rule of Modus Ponens in classical logic. It states that if some structure F derives a category A/B, and if some other structure A derives a category B, then the composition of these two structures F and A derives a category A. It is important to notice the parallellism between the logical product operator • to the right of h- and the structural product operator ° to the left. The structural connective builds a compound configuration of grammatical signs from more elementary configurations, whereas the logical operator expresses the type-logical properties of these configurations. As a simple example, we take the string of words "Mary loves Paul" and see if it derives an s, the category of a sentence. The top of the derivation (corresponding in a sense to the lexical entries) shows both "Mary" and "Paul" to be of category NP, and the verb "loves" of category (NP\S)/NP, meaning that "loves" yields a sentence if it is first combined with an NP (the object) to its right, and then with another NP (the subject) to its left. For ease of presentation, we use structural labels in the antecedent, and type formulas in the succedent:
Because the type of the verb indicates that it needs to be combined with its object first, we start by applying the E/ rule to "loves" and "Paul." It composes the words to form the string "loves Paul," which is of category NP\S—compare this category to the subcategorization list of a transitive verb, from which the object has been removed. The newly constructed phrase is not fully saturated yet, as it is still looking for a subject to its left. This is taken care of by the E\ rule, composing "Mary" and "loves Paul," which results in the string "Mary loves Paul" of the desired category s. The square brackets reflect the bracketing in the category of the verb, in that the verb and its object have been combined first, before the subject was added (the outermost brackets are dropped to improve readability). In a system where Associativity is assumed, the brackets can be omitted because all possible bracketings of a string can be generated, which are all
276
Esther Kraak
equivalent. We, however, assume a nonassociative base system, sensitive to constituent structure, for which the use of brackets is essential. As we already mentioned, structural rules of Associativity (rebracketing) and Comrnutativity (permutation) can be added to the nonassociative system NL, which will destroy the system's potential to discriminate between different structurings (dominance) and orderings (linear precedence) of the linguistic material. In natural deduction format, these rules assume the following form, where the double horizontal line indicates premise and conclusion can be interchanged, and F{A} means there is a substructure A somewhere in T:
By adding these structural rules to our system, we would not only be able to derive strings such as Mary loves Paul, but also Loves Mary Paul or Paul Mary loves as being of category s. In some cases this kind of freedom of constituent structure and/or word order can be desirable. For instance, we would like both Yesterday it was raining and It was raining yesterday to be derivable (with the same type assignment to "yesterday"}, which is not the case in (N)L. But at the same time we want to avoid derivability of such ungrammatical strings as Was yesterday it raining or Loves Mary Paul that can be derived in (N)LP. Simply assuming the structural rule of Commutativity causes our system to lose its order sensitivity completely, leaving us with no means to distinguish between grammatical and ungrammatical word order. The same holds, mutatis mutandis, for Associativity and structure sensitivity. We will see in the sections to come that it is nevertheless possible to combine different levels of structure- and order-sensitivity in one framework, by adopting a multimodal architecture which allows more fine-structuring of the notion of linguistic composition. Crucially, it will give us explicit control over the structural management of linguistic resources, instead of having the structural possibilities hard-wired into the system. 2.2. Structural Control In the previous section we have seen, in a nutshell, that although each of the Lambek systems in the structural landscape has its merits, no single one of them is capable of handling all the facts of linguistic analysis on its own. Their shortcomings can be remedied in two different ways, both providing control over the structural aspects of resource management: by
French Object Clitics
277
extending the framework with unary control operators that license or constrain structural operations on the resources (cf. Hepple, 1990; Kurtonina and Moortgat, 1996; Moortgat, 1997; Morrill, 1994; Morrill and Gavarro, 1992; Morrill et al., 1990; Nishida, 1991; Stabler, 1996; Versmissen, 1996) and by introducing a multiplicity of composition modes which all have different resource-management properties hard-wired into them (cf. Moortgat and Oehrle, 1994, 1996). 2.2.1. Unary Control Operators The central concept of residuation can be generalized to the case of ft-ary operators. The language of binary operators is extended by Kurtonina and Moortgat (1996) with a pair of unary residual operators O (diamond) and D* (box), for which the definition of residuation (fie < y iff x < gy) takes the following form:
We can see how this instantiates the general definition if we assume that O = / and D = g, and substitute A for x and B for y. By putting together the unary and binary category-forming operators we obtain a "mixed" language ^ defined inductively from a set of atomic categories j/ as
Category formulas of this language can be interpreted in terms of a mixed Kripke frame {W, R2, R2\ where the set of possible worlds W represents the set of linguistic resources, and linguistic composition is modeled by binary and ternary accessibility relations R2 and R3. The frame is turned into a model by adding a valuation v for the new unary operators (compare with that for the binary operators):
By analogy with the ternary relation .R3, we can think of binary R2 as a relation composing a whole, z, but this time out of a single part, x. It is
278
Esther Kraak
important to note that the valuation for D; is not the same as that for the box (necessity operator) in modal logic; the downarrow on the box is indicative of this difference. The accessibility relation is visualized in the diagram below, where the first and second graphs represent the relation R2 interpreting the unary category-forming operators O and D^, and the third graph represents the relation R^L interpreting the modal logic necessity operator D, given here for comparison. The diagram clearly shows the operators O and nl to be duals, as the composition relation R2 moves in opposite directions: whereas OA is the whole, composed by R2 from the part A, D^4 is the part which R2 composes to form the whole A:
The unary operators will serve as "features" that control the structural aspects (order and constituency) of linguistic composition. They can either be employed to constrain the use of permutation and rebracketing in the less discriminating logics or to license restricted use of these structural operations in the more discriminating ones. In this they show great similarity to the ! and ? operators of Linear Logic. The control features can be seen as consisting of a "question" (n 1 ) and an "answer" (O) part that cancel each other out, as expressed by the basic law Oni4 -> A (derivable in one residuation step from n^A -> \31A. What this law means, intuitively, is that a feature must be explicitly checked before it can be consumed in a grammatical deduction. In this sense the law bears a strong resemblance to the principles governing the checking of formal features that have been proposed by Stabler (1996). Rules for introducing and eliminating the diamond and box operator in natural deduction are given below.3 Analogous to the product operator, the diamond has a structural counterpart on the left-hand side of the turnstile, written here in the form of angled brackets { ). The antecedent (F> stands for a structure F which has this structural ( > connective as its main (outermost) operator.
French Object Clitics
279
It is easy to see that the rules for the box correspond to the residuation laws at beginning of section 2.2.1. The IO rule is not hard to understand either: if a structure F derives a category A, then a diamond can be introduced as main operator on both F (the structural diamond) and A (the logical diamond). The elimination rule for the diamond is somewhat more complex, but it is in fact a rule of substitution (like the elimination rule for the product; cf. the Appendix): the structural operator { ) can be replaced by the logical operator O. Kurtonina and Moortgat (1996) develop a theory of systematic communication between the systems NL, L, NLP, and LP on the basis of the diamond and box operators licensing and/or constraining the use of structural operations of Associativity and Commutativity. Through a series of embedding theorems, they show that it is possible to entirely recover the discriminatory potential of a weaker logic in a system which is less sensitive to the structural management of resources, and, in the opposite direction, that the flexibility of a stronger logic can be reintroduced in a system with a more fine-grained notion of structure-sensitivity. Like the binary connectives /, •, \ the unary operators O and D; are defined in the pure logic of residuation: there are no restrictions on the binary accessibility relation R2 interpreting the diamond and box.4 This base logic could be extended with postulates imposing constraints on R2, such as reflexivity and transitivity axioms (just as adding associativity and commutativity postulates would restrict R3). However, a logic based solely on residuation is appropriate for the unary operators used in the analysis to be presented: it will be shown in section 6.1 that a logic containing a reflexivity axiom would produce undesirable derivability results. Applications for unary control operators can be found in a variety of linguistic phenomena, for example in the realm of the locality domains that Hepple (1990) and Morrill (1994) have tried to account for with their domain modalities and bracket operators. Versmissen (1996) uses unary operators to give a categorial account of word order domains, and in the next section we will see a concrete example of the use of the diamond and box operators in an analysis of Dutch verb raising, following Moortgat and Oehrle (1996).
280
Esther Kraak
2.2.2. Multiple Forms of Linguistic Composition The second direction in which Lambek systems can be extended is by moving from mixed (W, R2, R3} frames, where connectives are interpreted in terms of a single form of linguistic composition, to multimodal frames (W, R2, R?), with i, j ranging over sets of indices J^ f. Each relation Ri j models a different way of composing the linguistic resources, and therefore the i e J^and j e^are referred to as the resource management modes. At the end of section 2.1 we already touched upon this idea of multimodality, which we will now present more formally. Multimodal systems contain a variety of (unary and binary) logical connectives <>• and •;. For each of these logical connectives there is a matching structure-building operator < >;- or 0;. As in the unimodal setting, each multimodal operator is related to its corresponding residual(s) by laws of residuation now taking the form
The indices i, j range over sets of modes in which the linguistic resources (words, phrases) can be combined, such as commutative mode, associative mode, etcetera. With each comes a mode-specific rule defining the behavior of the operator in that mode. For example, the commutative product •co is accompanied by the postulate A »co B -> B »co A, stating that if two signs are put together in commutative mode, reversal of their order is allowed. The multimodal language & can be inductively defined in a straightforward manner, and a valuation for the category formulas in the multimodal language can be given in terms of accessibility relations R2 and R^, with /,;' variables over the diverse modes of linguistic composition. The natural deduction rules for the multimodal connectives are modified, to ensure that the resource management modes are respected when introducing or eliminating any of the operators. They can be found in the Appendix. Addition of frame conditions linking the different .R, and Rj will enable the various logics in a multimodal system to interact with each other. These linking principles give the system a much greater control over resource management, making it more powerful than the sum of its parts. Linking of the different families can be realized in two ways: through Inclusion and Interaction principles. Inclusion principles order the different resource management modes. A certain mode i is ordered before another mode k by the following frame condition, with the corresponding Inclusion axiom and natural deduction
French Object Clitics
281
rule (for all x, y, z e W}\
Interaction principles mix the distinct modes. Two kinds of interaction can be distinguished: interactions between a binary Rj and a ternary Rif and interactions between two different Rt that are both ternary. Instances of the first type of interaction are the strong and weak distributivity principles K, Kl, K2 that distribute unary O over binary •. In a multimodal system, these principles can be restricted to specific mode interactions by defining the frame conditions for specific modes i,j. In axiomatic form, they will look like this:
Communication between two binary families can be established through the weak distributivity principles of Mixed Commutativity and Mixed Associativity, that can only apply in situations where the accessibility relations are in different modes (/ ¥= k). They are given here as postulates:
Whereas in the classical systems (NL, L, NLP, and LP) the structural rules of Associativity and Commutativity are either always available (hard-wired into the system) or never, in the multimodal setting these systems are put together in such a way that "substmctural travel" between them is possible. This means that elements of a logic with less discriminative power, say LP, can be imported into a more sensitive logic such as NL, through the mixed rules that define interactions between the various modes of linguistic composition. As this traveling is not allowed in general, but is restricted to situations where specific modes i and k are in construction, collapse of the more discriminating into the less discriminating logic is avoided. In other words, the interaction principles provide us with the structural flexibility needed to handle certain linguistic phenomena, without creating the problems of overgeneration that would ensue from adopting the unimodal rules of Associativity and Commutativity.
282
Esther Kraak
2.3. An Example: Dutch Verb Raising We conclude this section with an example taken from Moortgat and Oehrle (1996), showing how the multimodal framework enables us to deal with constructions that cannot be handled in a unimodal setting. We have chosen this example because the analysis it presents shows a strong resemblance to the one we will give for French object clitics in sections 4-6. Consider the following Dutch (subordinate clause) verb raising construction: (Jan weet dat Piet) Marie wil (Jan knows that Piet) Marie wants NP
plagen. to tease
VP/INF NP\INF
'Jan knows that Piet wants to tease Marie.' Given the type-assignments above, the grammatical string of words Marie wil plagen does not derive a VP in an order-sensitive system such as NL: the auxiliary wil is in between the infinitive plagen and its argument Marie, thus blocking application of the \E rule. On the other hand the string wil Marie plagen, which is ungrammatical as an embedded verb phrase, does derive a VP. In a multimodal system these problems of under- and overgeneration can be solved. In Moortgat and Oehrle (1996), two binary modes are assumed: normal phrasal composition mode ^, used for combining for instance a verb with its noun phrase arguments; and head adjunction mode »0, for putting together two elements that are both heads. These modes interact via the principles of Mixed Commutativity (MC) and Mixed Associativity (MA) below, which define the head adjunction product to take scope over the phrasal composition product: after application of these rules the »0 operator ends up governing the configuration with the •l operator.
The words of the string above are assigned to the categories NP, VP/OINF, and NEWEST, respectively: the infinitive combines with its NP argument in phrasal composition mode, whereas the auxiliary and the infinitive are linked together using the stronger of the two modes. The MC rule now enables derivation of Marie wil plagen as a VP, because the NP and the auxiliary can swap positions:
French Object Clitics
283
However, we are still left with the derivability of the ungrammatical string, and this is where the unary operators come into play. Moortgat and Oehrle introduce two unary operators, C^ and O0, whose control behavior is restricted by the following inclusion and distributivity postulates:
It is important to note that these rules allow O1 to be distributed only over •1, and O0 only over «0. The linguistic intuition behind these rules is best understood if we regard O1 as checking a phrasal head, and OQ as checking a lexical head. Distribution principles then regulate the flow of head feature information through the structure.5 By equipping the categories of verbs in the lexicon with a marker for lexical headedness DO , which needs to be counterbalanced by a corresponding diamond O0 (remember that OD^ derives A), derivation of the ungrammatical wil Marie plagen is blocked:
Reading from the bottom up,6 the goal of the derivation is to determine whether the string wil Marie plagen derives a properly headed verb phrase, indicated by the box operator on the VP. The rule I lifts this box over the turnstile where it assumes the form of its residual, the structural diamond operator < }\. This phrasal composition diamond needs to be distributed over a head adjunction configuration, but the mismatch of modes blocks the use of rule K. Application of the inclusion principle causes the diamond to switch from phrasal composition to head adjunction mode. Now it can be strongly distributed over the product, thus identifying both substructures of the head adjunction as lexical heads. The elimination rule splits up the structure, leaving us with the lexical entry axiom (in one more step) for wil on the left and Marie plagen on the right. The diamond then must continue its nondeterministic search for the lexical head in the substructure Marie plagen. However, the operator is now in mode 0, for
284
Esther Kraak
which distribution over the phrasal composition product connecting Marie and plagen is not allowed. The inclusion principle being one-way only, switching the diamond back to mode 1 is not an option, and derivation fails at this point. To conclude, we present the derivation of the grammatical construction. It shows that Marie can be composed with the infinitive, by applying the Mixed Commutativity rule, but only after (again reading from the bottom upward) the headedness of the entire phrase has been checked via distribution of the diamond. In other words, as long as headedness has not been checked, phrasal composition cannot occur within a head adjunction configuration, for this would cause the distributivity problem we have just seen in the ungrammatical case.
If the semantically interesting part of this derivation (here just the elimination steps for the slashes, as only these correspond to an operationfunctional application-in the semantics) is annotated with semantic terms, we see the verb wil take correct scope over the application of plagen to Marie:
The basic grammar logic that has been presented in this section constitutes the starting point for the grammar fragment to be developed in sections
French Object Clitics
285
4-6. The following section gives a brief overview of the data which the fragment aims to cover.
3. DATA
In this section we list the facts about the behavior of French object clitics for which we want our grammar fragment to account. They concern the relations object clitics have with their full NP counterparts, with the verbs they lean on and with other clitics. Our analysis of these data, based on the multimodal framework as we have just seen it, will then be presented in sections 4-6. • Clitics are pronouns in their weak form. French pronouns can be classified into weak and strong forms. The first group, also known as clitics, includes the subject pronouns je, tu, il, elle, ce, on, nous, uous, Us, elles; the direct object pronouns me, te, le, la, se, nous, uous, les; the indirect object pronouns me, te, se, lui, nous, vous, leur; locative y and genitive or quantitative en; while the latter consists of the corresponding strong forms moi, toi, lui, elle, nous, vous, ewe, elles, and cela. Not all clitics have a corresponding strong pronoun, such as, for example, v, which replaces a combination of the preposition a and a normal NP. (1) a. Je le lui donnerai. ('I will give it to him.') b. Eux n'auraient pas fait cela. ('They would not have done that.') • Object clitics are in complementary distribution with strong pronouns and normal NPS. Weak object pronouns occur in positions that cannot be occupied by a normal object NP; conversely, object clitics cannot occupy NP positions: (2) a. b. c. d.
Marie *Marie Marie *Marie
les connait. ('Marie knows them.') mes amis connait. ('Marie knows my friends.') connait mes amis. connait les.
Apart from constructions like (3a), where the NP is added to a sentence (but separated from it by a pause) to emphasize the object in that sentence, a clitic pronoun can never occur simultaneously with its NP counterpart or with its strong pronoun variant: (3) a. Marie les connait, mes amis. ('Marie knows them, my friends.') b. *Marie le voit Jean. ('Marie sees him Jean.')
286
Esther Kraak
• Clitics cannot occur at all in the absence of a host. Unlike strong pronouns and full NFS, clitics need an adjacent, phonologically strong word to support them. Without such a host the sentence is ungrammatical: (4) a. b. c. d.
Qui as-tu vul ('Who did you see?') *Les. Ewe. ('Them. Them.') Ou as-tu etel ('Where have you been?') *Y. A la maison. ('There. At home.')
• Object clitics occur preverbally. Apart from their occurrence in positive imperatives, object clitics can only occupy preverbal positions, the verb serving as their host: (5) a. b. c. d.
Jean lew donne des cadeaux. ('Jean gives them presents.') *Jean donne lew des cadeaux. Donne-les-lui. ('Give them to him.') *Les lui donne.
• A clitic's verbal host cannot be a past participle. Although every clitic or combination of clitics needs an adjoining verb to lean on, not all verb forms are possible hosts for clitics. In French, verbs in their finite, infinitival, or present participle form (including the gerundive construction with en) can fulfill this role, whereas past participles never do: (6) a. Je lui parlerai demain. ('I will speak to him tomorrow.') b. Marie veut la lui montrer. ('Marie wants to show it to her.') c. Jean regarde Paul en Vecoutant. ('Jean looks at Paul while listening to him.') d. *Elle a les manges. ('She has eaten them.') e. Elle les a manges. • Nothing can intervene between object clitic and verb. Apart from another object clitic, no lexical material can intervene between the object clitic and its host verb: (7) a. b. c. d.
Jean ne Va pas vu. ('Jean has not seen him.') *Jean le n'a pas vu. Marie le lui dira. ('Marie will say it to him.') *Marie le lui souvent a dit. ('Marie has often said it to him.')
• Clitics cannot be conjoined. Unlike full NFS and strong pronouns, clitics cannot be conjoined, nor can they have wide scope over a conjunction of verbs: (8) a. *Jean le et la voit. ('Jean sees him and her.') b. *Jean le voit et entend. ('Jean sees and hears him.')
287
French Object Clitics
• Clitics can occur on a host of which they are not a complement. Some verbs allow a clitic to be attached to them, even though it is an argument of the embedded verb. This is the case with verbs expressing causation, such as laisser and faire, or perception, like voir, entendre, and others, when the understood subject of the embedded verb is a dative NP: (9) a. Marie laisse manger unepoire a Jean. ('Marie lets Jean eat a pear.') b. Marie la lui laisse manger. ('Marie lets him eat it.') In sentence (9b) the clitics are attached to the finite verb, although they are the (understood) subject and the object of the infinitive lower in the structure. However, if the embedded verb has an accusative NP as its subject, its object cannot be attached to the matrix verb: (10) a. b. c. d.
Marie laisse Jean manger une poire. ('Marie lets Jean eat a pear.') * Marie la laisse Jean manger. Marie le laisse manger une poire. *Marie le la laisse manger.
Most verbs do not allow the so-called clitic climbing of example (9b), leaving the clitic(s) attached to the infinitive: (11) a. Jean veut les manger. ('Jean wants to eat them.') b. Jean les veut manger. • Clitics appear in a fixed order. An important property of clitics is that their relative order is not subject to any variation. It is generally assumed that French pronominal clitics are realized according to a kind of "template," describing the order in which clitics (and negative particle ne) must occur: Position:
I
II
NOM
ne
III me te nous vous se
IV
V
VI
VII
SACC
3DAT
y
en
Here NOM represents the subject clitics; 3ACC the third person accusative clitics le, la, les; and BOAT the third person dative clitics lui and leur. Usually the following assumptions accompany this template: clitics must occur in the indicated order, there can be only one clitic per position, normally there are at most two object clitics, and in a cluster of clitics the positions III and V cannot be filled simultaneously. The following examples reflect these restrictions on order and co-occurrence of the clitic
288
Esther Kraak
pronouns: (12) a. b. c. d. e. f.
Je ne le lui donneraipas. ('I will not give it to him.') *Je ne lui le donnerai pas. Jean me Va dit. ('Jean has said it to me.') *Jean le m'a dit. *Elle te lui presentera. ('She will introduce you to him.') Je I'y ai vu. ('I saw him there.')
The phenomena described here will have to be accounted for in any adequate theory of clitic pronouns. We will see in the following sections that the data from examples (2)-(5) can be taken care of by assigning clitics to the proper category. Interaction between different modes, controlled by corresponding unary operators, will be responsible for correct treatment of the facts in examples (6)-(12).
4. SIMPLE CLITIC CONSTRUCTIONS If we want our grammar fragment to cover the data presented above, it is essential that we assign the clitics to the right category. In this section we will build the appropriate type step by step, motivating our choice as we go along: the use of a higher order type, of hypothetical reasoning, of features, resource management modes, and unary control operators will all be reviewed. 4.1. Assigning a Category: Stepwise Refinement In accordance with the way pronouns are standardly treated, we assign the object clitics to a higher order category. To capture the fact that (in nonimperative sentences, to which we restrict ourselves here) they are situated immediately before their host verb, we assign them to a functor category looking rightward for a verbal argument. From this verbal argument an NP must be missing, namely the one that has been cliticized. This incompleteness will, through hypothetical reasoning, block co-occurrence of a clitic with its corresponding full NP form, thus ensuring the complementary distribution of the pair (clitic, NP) mentioned in section 3 (examples (2) and (3)). As the NP that has been cliticized does not necessarily come from a position adjacent to the verb, we do not use the normal slash operator but the extraction1 operator T • An item of category A^B will yield something of category A if it is combined with an item of category B somewhere.
289
French Object Clitics
(Note that this makes the slash operator a special instance of the extraction operator.) An informal natural deduction rule for the introduction of this operator can be found in the Appendix. If we assume the verbal argument of a clitic to be of category VP, the provisional category to which we assign the clitics is then of the form VP/(VPtNP)
Because we need to distinguish between finite, infinitival, and past participle verb phrases, which will be represented in this article simply by VPy (abbreviating NP\S), VP,-, and vppp, respectively, the clitic's category must be specified for a certain type of host verb as well. In the clitic types we will use the general category VP^ for verb phrases, where X is a variable over {/, i, pp}. To make certain that the grammatical function of a clitic is the same as that of the NP it replaces, we add a subscript for case to the missing NP in the clitic's category. This means that clitics take as their argument a verb phrase which is incomplete for an NP y , with Y ranging over {a, d} for accusative and dative NFS. In the derivations that follow, the variables X and Y will be directly instantiated to their proper values.8 The phonological bond between clitics and their hosts is a strong one: we saw in example (7) that nothing can intervene between the two. Following Moortgat and Oehrle's analysis of Dutch verb raising (section 2.3), we assume two modes for putting together linguistic material: "normal" phrasal composition mode •c and a stronger mode, equivalent to their head adjunction, which we will call clitic attachment •ca. Likewise, we assume two modes for the unary operators, Op for phrasal heads and O, for lexical heads (corresponding to Moortgat and Oehrle's Ol and O0, respectively), and the following rules of inclusion and interaction (we will not use the rule MC):
In light of the evidence from example (7), we assume the main operator of the clitic category to be in strong clitic attachment mode, whereas the missing hypothetical NP in the argument subcategory is combined with the verb in standard phrasal composition mode. The verb, being the head of the sentence, is equipped in the lexicon with an/ operator identifying it as the lexical head. Because clitics form a strong cluster with the verb, they
290
Esther Kraak
too have D/ as the main (outermost) operator in their lexical category. The general form of the category to which clitics are assigned is
The lexical entry that we are assuming for a transitive verb like voit ('sees') is the following:
Now that the foundation of our categorial treatment of object clitics has been laid, we can give a simple example to illustrate the working of the system. Consider the sentence Jean la voit, 'Jean sees her.' Derivation of this sentence is as follows (we use VP^ as abbreviation for NP^\ C S):
The goal of this derivation is to find out whether the string of words Jean la voit is a properly headed sentence (category D^ s). Via the rules for distribution of the diamond and a switch from Op to O,, the clitic and the verb both end up (reading from the bottom upward) with a O, diamond, which is the correct one considering their category. If we start from the top, the verb voit first combines with a hypothetical accusative NP, the one that has been cliticized. This hypothesis is then withdrawn through application of the I|c rule, resulting in a VP category missing an accusative NP somewhere. By taking this complex category as its argument, the clitic binds the argument position "occupied" by the hypothetical NP, and in this way fulfills the role of direct object to the verb. Now it also becomes clear how co-occurrence of the two forms is blocked: if the verb is combined with a real NP first, it is no longer possible to construe the proper argument for the clitic, as there is no hypothesis to be withdrawn.
French Object Clitics
291
Looking at the derivation, the impression might arise that it is necessary to present the system with a string in which the correct operators and brackets are already present. It has been shown by Moot (1996), however, that this is not the case. He uses labeled proof nets to show that the configuration of the antecedent is "projected" automatically in the unfolding of the lexical type assignments. What we see is that (analogous to the Projection Principle, for instance) structural information originating from the lexical assignments is projected through the derivation.
4.2. Coordination As it stands, the multimodal framework can handle the aspects of object clitic behavior that were demonstrated in section 3 with examples (2)-(5): object clitics do not occupy normal object NP positions and are in complementary distribution with full object NFS; they occur directly in front of their verbal host and can never appear without it. An interesting consequence of the way in which the system works is that the right predictions about the conjoinability properties of object clitic pronouns follow automatically from it. Consider example (8), repeated here as (13): (13) a. *Jean le et la uoit. ('Jean sees him and her.') b. *Jean le uoit et entend. ('Jean sees and hears him.') From example (13a) we see that it is not possible to conjoin two object clitics; example (13b) shows that an object clitic cannot have scope over a coordination of two verbs. To reflect the fact that the conjunction et can be used to combine all kinds of categories, as long as they are identical,9 we assign it to the category schema below, where X is a variable ranging over the entire set of categories. The fact that the conjunctive et constitutes the head of a conjunction is indicated by the box control operator in lexical head mode /:
As with the variables in the category for clitics, the variable X will be instantiated, in each specific derivation, in accordance with the category of the elements to be conjoined. In example (13b), for instance, X would be substituted by the lexical type of a transitive verb. To see how exactly the ungrammatical cases mentioned above are ruled out, let us try to derive
292
Esther Kraak
the first sentence, (13a):
Derivation fails at the point where the diamond in lexical head mode needs to be distributed over the phrasal composition (mode c) product, which is not allowed. And since the < )/ diamond cannot be turned back into a < )p diamond (because of the unidirectionality of the inclusion principle I) there is no way in which the derivation can proceed. The same happens with the second sentence, where the clitic le takes wide scope over a coordination of two verbs:
The basic idea behind the system, which causes these constructions to be ruled out as ungrammatical, is that the prosodically strongest mode of linguistic composition (clitic attachment mode) does not allow any material using the weaker mode of phrasal composition to occur within its domain, as long as headedness has not been checked. This is precisely the reason why things go wrong in the attempted derivations above: in both cases a phrasal composition structure is contained within a clitic attachment configuration.
5. CLITIC ORDER
Now that we are able to deal with sentences containing one clitic pronoun, the grammar must be refined in such a way that, in cases where
French Object Clitics
293
there are multiple clitics, it allows only derivation of those sentences in which the clitic order is correct, i.e, corresponds to the order template we have seen following example (11). In its present form any combination and order of object clitics can be derived, which is clearly an undesirable situation. Since the unary operators already identify the clitic-verb cluster as the head of a sentence, they are ideally suited for conveying information about the order within that cluster. To this effect, we inscribe the diamond and box operator with a feature representing a position in the order template. We let this feature range over the integers 0 to 8, corresponding to the position numbers in the following order schema:
The order -< defined for the position features is as follows: 0 •< 1 -< 2 •< 3 x 4 x 5 x 6 x 7 x 8 and O x l x 2 x 3 ' x 6 x 7 x 8. What this means is that the first and second person accusative clitics me, te, nous, vous, with order feature 3', can never precede (nor follow) any of the other object clitics.10 They can, however, follow a nominative clitic and precede en and y and, evidently, the verb. The feature-inscribed unary operators will have the form <3>j and H^ for the logical operators, and < >* for the structural counterpart of the diamond, where j ranges over the modes as before, and x ranges over the set of position features.11 Lexical entries for the third person accusative clitic la and the transitive verb uoit, for instance, now assume the form
Rules of introduction and elimination for the feature-inscribed unary operators remain the same, as do the inclusion rule and the weak distributivity rules Kl and K2. An important role in enforcing the correct clitic order will be played by the strong distribution rule K, as this rule distributes the diamonds within the clitic-verb cluster. We refine it as follows:
Together with its side condition, rule K states that, whenever a diamond in lexical head mode is distributed over a product in clitic attachment mode,
294
Esther Kraak
the left conjunct must get a diamond whose position feature strictly precedes that of the right conjunct. This will ensure that the left-to-right order of the clitics must always correspond to the order prescribed by the template, and that no two clitics with the same order feature can co-occur. Two derivations will illustrate the working of this rule. As a first example, consider the partial derivation of the grammatical sentence Jean la lui donne ('Jean gives it to her'):
The aim is to find out whether this particular string of words (composed as shown) derives something of category s which is properly headed and of which the clitic-verb cluster is well ordered. This last requirement is expressed by an order feature 0 in the box operator on the succedent category s. It can be regarded as a kind of start symbol for the clitic order, expressing the fact that any grammatical combination of clitics is still allowed to occur. After the box operator has changed into its residual diamond ( >° (via rule I Dp), has been distributed over the phrasal composition product to the right conjunct la lui donne (rule K2), and its mode has been switched from phrasal to lexical (rule I), the ( >/ diamond is distributed over the clitic attachment cluster via two applications of rule K. The first one distributes a diamond with order feature 4 to the left conjunct la, and a diamond with order feature 5 to the right conjunct lui donne, respecting the condition on K. In the second step this diamond with order feature 5 must be distributed over the clitic attachment product again, this time yielding a diamond with order feature 5 for the left conjunct lui, and one with order feature 8 for the right conjunct donne. We can tell that the order in the clitic-verb group is grammatical from the
French Object Clitics
295
top of the derivation: it shows the correct lexical entries for the object clitics and the verb. This is not the case with the ungrammatical—as a result of non-well-ordering—sentence *Jean lui la donne:
This derivation proceeds like the previous one, up to the point where rule K must be applied. The left conjunct of the outermost clitic attachment is lui, which gets a diamond with order feature 5. Because of the condition on rule K, the right conjunct la donne must get a diamond whose position feature follows 5, so for instance 6. The second application of K then distributes this diamond with order feature 6 to la and one with feature 8 to the verb donne.12 However, this distribution of diamonds does not result in a successful derivation: the lexical entry for la does not have E/ as its main operator, which is the outcome of the applications of rule K, but H/ instead. We could also have started the above derivation from the top, beginning with the correct lexical entry for la, but then the problem would arise at the point where rule K must be applied for the second time:
It is not possible for rule K to be used without violating the condition on its application, which says that the order feature on the first conjunct must
296
Esther Kraak
precede the order feature on the second. Since the left conjunct lui has an outermost diamond operator with order feature 5, and the diamond on the right conjunct la donne is inscribed with order feature 4, it is clear that the condition cannot be met and that failure of derivation must result. So we can see how the correct distribution of diamonds to clitics and verb is not deterministically chosen, but is in fact enforced by the interaction between rules and lexical entries: any wrong choice of diamond (with an order feature that does not correspond to the order feature in the category of the word) will bring about a failure of derivation. Only in those cases where the order within the clitic-verb group matches the order in the schema at the beginning of this section does our system produce a successful derivation, which is exactly what we aimed to achieve.
6. CLITICS AND VERBAL COMPLEXES Having shown the working of our system for simple clitic sentences, we will now consider cases in which the verb phrase consists of a finite verb followed by a past participle or an infinitive. 6.1. Modal and Perception Verbs Our first case concerns sentences that contain a clitic and a verbal complex consisting of a modal verb (for instance uouloir, devoir, aimer) and an infinitive. The following types are assigned to the infinitives:
Modal verbs never serve as hosts to clitics, for they do not allow clitic extraction out of the infinitival VP domain. This is evident from the following sentences: (14) a. Marie veut la manger. ('Marie wants to eat it.') b. *Marie la veut manger. In order to account for these data, the category for the modal verb must be such that the domainhood of its argument VP is respected. To this effect, the argument subcategory of the modal verb is equipped with a
French Object Clitics
297
complex of unary operators:
In words, what this combination of O and [Hip does is as follows: first, it should be noted that the diamond introduced here is of a different type than the unary operators used to convey information about headedness and clitic order. Crucially, the distribution rules do not apply to it, but otherwise it obeys the by now familiar rules of introduction and elimination. The function of this diamond is to impose a locality constraint13 on the infinitival VP argument of veut. It does so via the rule IO, that can only be applied if the entire antecedent structure is contained in unary brackets { >. If this is not the case, the diamonds on either side of the turnstile cannot be removed (reading from bottom to top) and derivation fails. After the domainhood of the argument VP has been successfully verified, GDp —just as in the other derivations we have seen—checks it for correct headedness and clitic order, through the introduction rule for the box and the distribution rule K. When every head in the infinitival verb phrase has been appropriately assigned a (structural) diamond operator with the correct order feature, all that remains is to determine whether the antecedent derives VP(, the category of an infinitival VP. In the following derivation, showing the grammaticality of the sentence Marie veut la manger, the clitic la will be instantiated as taking an infinitival argument (the rightmost part of the derivation is straightforward, as the reader can check):
We see that the structural diamond operator < > defining la manger as a closed-off domain is a necessary condition for derivational success. It is not conveniently put wherever it happens to be needed, but is entirely projected from the lexicon. Derivation of the ungrammatical *Marie la veut manger, where the direct object la has been cliticized outside the infinitival
298
Esther Kraak
domain of the embedded verb, is blocked as the hypothetical NP bound by the clitic cannot penetrate the diamond brackets around manger in order to be combined with the infinitive:
It is interesting to mention at this point that a logic with the modal reflexivity postulate A -» OA would enable us to derive the ungrammatical sentence above. By means of this postulate, the domain brackets ( > could simply be introduced around the structure [manger oc x] halfway up the deduction. Then it would become possible to apply rule IO, leading to a successful derivation. But, as we have seen, the reflexivity postulate is not included in the base logic for our control operators, so this problem of overgeneration does not occur in the present framework. Another group of verbs that do not allow arguments of the infinitival VP embedded under them to be cliticized comprises laisser and perception verbs such as uoir, entendre, ecouter, when these take an accusative NP argument which functions as the understood subject of the infinitive. Consider the sentences of example (10), repeated here with the verb uoit: (15) a. b. c. d.
Marie uoit Jean manger une poire. ('Marie sees Jean eat a pear.') *Marie la voit Jean manger. Marie le voit manger une poire. *Marie le la uoit manger.
The direct object une poire of the embedded verb manger cannot be cliticized onto the finite verb, as is shown by the ungrammaticality of sentences (15b) and (15d). However, cliticization of manger's subject Jean is allowed in sentence (15c). Note that the corresponding clitic le has accusative case, not nominative. In line with proposals made by Dowty
French Object Clitics
299
(1985), we therefore take the understood subject of the infinitive to be an accusative argument of the matrix verb. These considerations lead us to assign this group of verbs to the following category:
The argument VP constitutes a closed domain as in the case of the modal verbs, thus not allowing extraction of any vp-internal object. Since the understood subject of the embedded verb is not vp-internal but belongs to the matrix verb, it can freely be cliticized onto that verb, which is shown in the following derivation:
An attempt to derive sentences (15b, d) will result in failure for exactly the same reason as the derivation of *Marie la veut manger earlier in this section did: at a certain point we are left with the subgoal (manger) oc x \OMp VP,-. As the label x of the hypothetical accusative NP is unable to enter manger's domain, the sequent fails to meet the criterion for applicability of rule IO, demanding that the entire antecedent be contained in unary brackets. Thus the types for voit and veut correctly imply the ungrammaticality of sentences like (14b) and (15b, d) above. With the category for the perception verbs and laisser we can also derive instances of so-called "downstairs" cliticization where the clitics binding arguments of the infinitive are not extracted out of the VP,
300
Esther Kraak
domain: (16) a. Marie laisse Jean la manger. ('Marie lets Jean eat it.') b. Marie le laisse la manger. Because the clitic la is strongly attached to the infinitive, it remains within that verb's domain. This means that the entire VP argument of the matrix verb is contained within unary domain brackets, and therefore that the problem we encountered while attempting to derive *Marie la veut manger does not arise. Derivation of example (16b) for instance proceeds exactly like that of sentence (15c), if we replace (manger oc une poire) by (la oca manger). After successful checking of the antecedent's domain status and removal of the domain brackets, we are left to prove the goal [la oca manger] h- GD^ VP,-, an exercise we will again leave to the reader. 6.2. Clitic Climbing Finally, we consider cases of so-called "clitic climbing," in which clitics attach themselves to the matrix verb although they are complements of some verb embedded under it. The first major group of verbs exhibiting this phenomenon consists of the tense auxiliaries avoir and etre, which serve as hosts for clitics belonging to the past participle. Clitic climbing is obligatory, as can be seen from sentences in which clitics appear on the past participle: they are always ungrammatical (cf. (6d, e)). (17) a. Marie Va mange. ('Marie ate it.') b. *Marie a le mange. We assign the past participles of verbs like dormir, manger, and donner to the types given below:
Cliticization onto the tense auxiliary is in clitic attachment mode as before. We do not assume the auxiliary to form an equally strong cluster with the participle, since elements like pas and peut-etre can intervene between the two. However, to combine tense auxiliary and past participle in phrasal composition mode would not reflect the intuition that these two verbs together do constitute a stronger type of complex, for clitics are not allowed to occur in between. They must climb over the entire verbal complex instead, as we saw in example (17b). This intuition can be
French Object Clitics
301
formalized by introducing a third mode of linguistic composition, which we will call clause union mode. The clause union product and its residuals, the slashes, will be subscripted «, and the following interaction rules are added to the system:
In words, the distribution rule Ku for the clause union mode states that a clause union configuration consists of two verbs—two lexical heads. For both conjuncts receive a diamond in mode /, indicating their lexical head status, and with order feature 8, which the reader will remember is the position of the verb in the order schema for the clitic-verb cluster. We will see below how this forces clitics to climb over the entire verbal complex formed by the clause union product. The Mixed Associativity postulate defines clause union mode to take scope over phrasal composition mode. The tense auxiliaries est and a which trigger clitic climbing are assigned to the following category, where the slash operator in mode u reflects the clause union between auxiliary and participle:
With this category we can derive a sentence like Marie Va mange. Derivation fails for ungrammatical cases such as *Marie a le mange where the clitic is attached to the participle, because both conjuncts of the clause union product must get a diamond with order feature 8, for the verb position. Consequently, the clitic will not receive the diamond it needs: because of the condition on rule K the only possible order feature that can be distributed to le is 8, which does not match the order feature 4 in the clitic's lexical box operator. In addition, according to that same rule the second conjunct mange should be given an order feature following 8, which is impossible as 8 is the last position in the schema.
302
Esther Kraak
The other group of verbs showing obligatory clitic climbing contains faire, and the perception verbs and laisser insofar as these take a dative complement which functions as the understood subject of the embedded infinitival verb. For instance (see also (9a, b): (18) a. Marie fait manger une poire a Jean. ('Marie makes Jean eat a pear.') b. Marie la fait manger a Jean. c. Marie la lui fait manger. d. *Marie fait la manger a Jean. The clitic la corresponding to the (vp-internal) direct object of manger must attach itself to the front of the finite-infinitival verb cluster as in sentences (18b) and (18c); cliticization onto the infinitive is not allowed, as can be seen from the ungrammaticality of (18d). The fact that clitic climbing is obligatory in these cases is indicative of a clause union relation between the two verbs. We therefore assign this group of clitic climbing triggers to the following category:
As with the versions of the perception verbs and laisser that host cliticization of their vp,-external accusative NP argument, here the dative NP "subject" of the embedded verb cliticizes onto the main verb, of which it is a complement. But unlike the former verbs, clitic climbing verbs such as faire allow the vprinternal NP to be extracted from its domain and cliticized onto the triggering main verb, as in examples (18b, c). The category assumed here for the clitic climbing verbs produces the desired derivational results for these constructions. That the clause union mode does indeed force a clitic to climb, instead of allowing it to occupy a downstairs position, becomes clear when we try to derive (18d), *Marie fait la manger a Jean. At a certain point in the derivation, the diamond in lexical head mode must be distributed over the structure [fait ou [la oca manger]]. Following rule Ku, both conjuncts fait and la manger receive a diamond with order feature 8. But now we can see that we are in the same situation as before when we attempted to derive the ungrammatical sentence (lib), * Marie a le mange: the clitic cannot get a diamond carrying the order feature 4 which it needs in accordance with its lexical category, as such a distribution would violate the condition on rule K. So, with this last refinement of the framework, all the facts about the behavior of French object clitics presented in section 3 can now be accounted for by our grammar fragment.
303
French Object Clitics
7. COMPARISON WITH OTHER ANALYSES In this section, the analysis of preceding pages is compared with and Head-Driven Phrase Structure HPSG analysis of French clitics in
French object clitics presented in the other analyses in Categorial Grammar Grammar. In particular, we look at the Miller and Sag (1995).14
7.1. Clitics as Pronominal Affixes Following the argumentation of Miller (1991), who applies the criteria for distinguishing affixation from postlexical cliticization (given by (Zwicky and Pullum, 1983)) to the case of clitics in French, Miller and Sag, (1997) consider clitics to be pronominal affixes instead of independent syntactic words. They propose a lexical rule of pronominal affixation that turns a word into a cliticized word by removing one or more complements from the word's COMPS list, thereby reducing its valence. These complements must be of type afftixal-synsem). Since complements are identified with members of the ARG-ST list, a cliticized word will have one or more elements of type aff in ARG-ST. Morphological realization of the pronominal affixes is then triggered by the presence of these aff-type elements. Thus the rule guarantees complementary distribution of clitics and their full NP counterparts. Unlike Miller and Sag, we see no reason to draw a firm distinction between syntax and morphology. Rather, we like to think of the relation between the two as a sliding scale. This is not to claim that the distinction does not matter. Instead of embracing a strictly modular view, however, we take it that syntax and morphology are intricately intertwined. In our analysis this is expressed by the coexistence of different modes of linguistic composition, reflecting varying strength of prosodic attachment. The strongest mode, connecting clitics with their host, occupies a position nearer the "morphological side" of the scale, whereas the weakest mode, combining verbs and their arguments, is situated on the "syntactic side." Since we consider clitics as having their own lexical entries, this variation in compositional strength can be entirely projected from the lexicon. It is also worth noting that the complementary distribution of a clitic and its corresponding full NP, for which Miller and Sag need the abovementioned lexical rule, follows in our account from the interaction between the clitic's higher order type and the general mechanism of hypothetical reasoning triggered by that type. As the clitic binds the hypothetical position of the noun phrase it replaces, it will automatically be identified with that missing NP, thus blocking their simultaneous occur-
304
Esther Kraak
rence. It is important to see that in our view it is not the verb which selects for a "clitic" argument, as is the case in Miller and Sag's lexical rule. Instead, it is the clitic which serves as the functor, selecting a verb that is unsaturated for the NP complement whose role is fulfilled by the clitic. Combining the clitic with its verb argument (binding the hypothetical argument position) yields a verb whose valence is reduced with that same NP complement. Since we work in a resource-conscious logic where each assumption can and must be used only once, there is always exactly one clitic that binds an argument position. Additional constraints enforcing this uniqueness, as needed by Miller and Sag in order to prevent affixal haplology, are therefore superfluous.
7.2. Clitic Order
Miller and Sag treat the restrictions on order and co-occurrence of clitics within the clitic cluster by characterizing it as an ordered sequence of slots. In addition, they assume a substantial number of cluster conditions, stating for each of the slots whether and how it is to be filled, depending on the presence of certain elements on the ARG-ST list of the cliticized word. This is in line with their view of cliticization as an entirely morphological phenomenon. As we mentioned earlier, our account of cliticization does not embrace this modular view. Instead, the make-up of the clitic cluster results from the interaction of lexical type-assignments for the clitics with a language-specific order schema and generally valid deductive principles. In this sense, our position is more in line with work by Kathol (1995), who analyzes word order in terms of so-called topological fields. Sentences are regarded as consisting of an ordered sequence of fields or domains, in each of which only certain types of constituents occur. Every word is explicitly marked in the lexicon with a feature for the specific field(s) where its occurrence is allowed. This corresponds to our representation of clitics as lexical items marked for a specific position within the clitic cluster15 whereas in Miller and Sag's view it is the ARG-ST list of the clitic host which constitutes the locus of information relevant to clitic ordering. Another (categorial) treatment of cliticization which makes use of order features is that of Morrill and Gavarro (1992), who propose an account of Catalan clitics similar to the analysis put forward in the present article. They treat clitic order with a feature for cliticization class. This feature, in the form of an integer, is connected with both subtypes s in the higher order type which Morrill and Gavarro assume for the clitics. The idea is
305
French Object Clitics
that the value of this feature on the argument category of the clitic is higher than that of its result category. Thus, clitics that appear near the front of the clitic cluster will have lower values in their argument categories than those appearing closer to the verb. The verbs themselves, following the clitic cluster, have the highest value. The main difference with the analysis presented in this article is that, in the multimodal grammar logic, checking of these features is effected through deductive principles, whereas in Morrill and Gavarro's account it is done by means of unification.16
7.3. Clitic Climbing: Argument Composition The analysis of clitic climbing proposed by Miller and Sag17 is described by them as a case of argument composition: a clitic climbing trigger combines with an unsaturated verb and then with the arguments selected by that verb. In the lexical entry of such a trigger, the COMPS requirements of the verb argument are added to the trigger's own COMPS list, and their occurrences are identified via reentrancy. As Miller and Sag point out, the notion of argument composition is related to the concept of division type-shift in Categorial Grammar, which can be schematized as A/B ->
(A/C)/(B/C)
The analogy becomes clear if we take A/B to be the initial type of the clitic climbing trigger, B/C the type of its (unsaturated) verb argument, and C the complement of that verb argument. The higher order type for the trigger verb, on the right-hand side of the arrow, requires to be combined first with the unsaturated verb B/C, and after that with the verb's complement C. Accounts of clitic climbing based on a rule of division have been proposed in the field of Categorial Grammar by Casadio (1993) for Italian and Nishida (1991) for Spanish. However, these categorial analyses adopt a version of division valid only in systems which include the structural rule of Permutation, leading to overgeneration in the form of ungrammatical word order. A drawback of all analyses of clitic climbing founded on division is that, compared to the multimodal approach advocated in this article, they constitute a derivational detour. Why this is so can best be illustrated with an example. Consider the sentence Jean a vu Marie ('Jean saw Marie'), where the tense auxiliary a is a clitic climbing trigger. Deriving division in a proper (but unimodal) natural deduction setting, we arrive at the
306
Esther Kraak
following analysis for this sentence:
We see the division category for a being derived via hypothetical reasoning18 (the initial type and the higher order type are boxed for the sake of clarity). Then the newly obtained category is combined with unsaturated vu (its first argument, VPpp/NPa) and after that with the participle's complement Marie (its second argument, Npa). The immediate succession of an Introduction and an Elimination step acting on the same connective, in the middle of the derivation, indicates the possibility of simplification. Here, the difference between analyses based on rules (such as division) and deductive approaches like ours becomes clear: the former lack a notion of normalization of proofs, in the sense that the proofs they produce are not necessarily minimal derivations without any detours. In the deductive account, which does possess this normalization property, the steps needed above for the derivation of the division type can be pruned away, resulting in the following derivation:
French Object Clitics
307
As was shown in section 6.2, this analysis yields the desired results for the clitic climbing data. Among other things, it accounts in a principled way for the fact that, in French, cliticization on a past participle is not allowed. The analyses based on division mentioned above, on the other hand, need to stipulate application of this rule to be blocked specifically for verbs in past participial form. There is also another, more serious problem with the division-based approaches to clitic climbing, evidence of which is given in the first derivation of this section. In the process of constructing the higher order type, the structural rule of Associativity is needed to enable application of the I/ rule withdrawing the hypothetical NPa. Since the structural rules in their unimodal version are either always available or never, the consequence is that full Associativity must hold, with the consequent loss of constituent sensitivity. The multimodal analysis presented in the second derivation assumes restricted (mixed) Associativity only as an interaction principle, relating modes that are in themselves nonassociative, and therefore does not cause the system to collapse into constituent-insensitivity.
8. CONCLUSION It has been shown in this article that a deductive approach to cliticization is fruitful in that it throws new light on the "fine-structure" of grammatical resource management. Comparing the proposed multimodal approach with the resource management assumptions underlying competing analyses (both categorial and HPSG-style), two points have been made.19 First, from the point of view of derivational economy, argument composition approaches to clitic climbing are suboptimal: they represent detours that in the present deductive setting are eliminated as a result of proof-normalization. Second, the argument composition analyses discussed here violate linear order sensitivity (Casadio, 1993; Nishida, 1991) and/or constituent structure sensitivity (Miller and Sag, 1995; Monachesi, 1995). This means that, both within the unimodal categorial and the HPSG-based analyses, extra non-logical principles have to be imposed to constrain grammatical resource management. Such non-logical principles can be avoided as a result of the more delicate multimodal resource management regime advocated here. The framework used here for the analysis of French object clitics needs to be extended in a number of ways. I have already touched upon the idea
308
Esther Kraak
of adding a feature logic component to handle those features that were simply represented by indices in the above. The fragment can also be made to cover the rest of the French clitics: subject clitics, reflexives, en and y. Moreover, preliminary investigations into the realm of Italian object clitics strongly suggest that clitic systems from other Romance languages can be accounted for as well. Cross-linguistic variation can be captured by choosing different packages of resource management rules for the different languages. These three topics constitute my aims for future research.
9. APPENDIX: THE FRAMEWORK Logical Rules
French Object Clitics
309
Resource Management Rules
NOTES The natural deduction format is used here for ease of presentation, but it does not constitute a proper basis for proof search. However, equivalence of the natural deduction, axiomatic, and Gentzen sequent presentations has been proven (cf. Moortgat and Oehrle, 1996, and Moortgat, 1997). The latter has the property of decidability: it will tell us whether there is a derivation for a given string or not. Every possible step is examined, guaranteeing a complete search producing all possible derivations if there exist any, and a negative answer otherwise. 2 The product rules will not be needed here, as we will not make use of lexical categories containing the product operator. For the sake of completeness, however, they are included in the Appendix. The category formulas in these rules are not annotated with semantical terms, as the unary control operators are considered to be semantically inert. 4 In this respect they differ from the box operator proposed by Morrill et al. (1990), which requires a reflexive and transitive interpretation. 5 In this sense, they provide the deductive basis for the Head Feature Principle in GPSG and HPSG. However, the distribution principles are more flexible than the HFP in that they allow multiple headedness in head adjunction mode. This difference is essential for Moortgat and Oehrle's analysis of Dutch verb raising and, as we will see shortly, also for an adequate treatment of French objects clitics. Throughout this article we will use the term bottom up in the following sense: the bottom of a derivation is its conclusion, the goal to be proven, whereas the top is constituted by the premises—the lexical entry axioms. 7 Moortgat (1997) shows that this operator can be defined in the present def framework as follows: A\B =A/OE\B, where the unary operators give access to
310
Esther Kraak
modally controlled versions of the rules for Associativity and Commutativity, which look like this (the condition holds for each of the rules):
with some A, of the form (A) When the desired configuration has been reached, the diamond and box operators cancel each other out, such that OD^B can be used as a normal item of category B. 8 A topic for further research in this respect would be to see whether it is possible to combine the multimodal framework with a feature logic along the lines of Dorre and Manandhar (1995). 9 We restrict ourselves here to coordination of like types. 10 Note that this schema differs from the template following example (11) on two points. First, whereas the template does not allow combination of l,2DATwith 3 DAT clitics, our schema does. This leaves room for sentences such as Jean me lui semble fidele ('Jean seems to me to be faithful to her'), considered grammatical by a number of speakers. Second, in the schema we have separated 1,2ACC from SDAT clitics, blocking derivability of such sentences as *// me lui presente ('He presents me to her'). Because SACC clitics can co-occur with 3DAT clitics, they are positioned in the same branch of our order schema. As a consequence, the 1,2ACC clitics are separated from the 3ACC clitics, whereas the template does not prohibit simultaneous occupation of positions III and IV. As far as we know, however, there seems to be no evidence refuting our analysis. 1] It is also possible to express other features in the unary operators, such as case or verb form. We will not pursue this line of thought here, but see Versmissen (1996) for an implementation. 12 In sentences without clitics, the verb gets a diamond with the correct order feature 8 via a special rule (called 0; see the Appendix) which takes a diamond with order feature 0 to a diamond with order feature 8. The idea behind this rule is that, trivially, the empty clitic sequence is well ordered. 13 Comparable to the function of the "bracket" operators of Morrill (1994), who assumes a less restrictive logic, however. 14 Miller and Sag (1995) was the representative HPSG analysis at the time when contributions to the present volume were written. The reader should notice that in the revised analysis of Miller and Sag (1997)-summarized in Abeille et al. (this volume)-the use of lexical rules has been abandoned. 15 For a more general categorial implementation of the concept of word order domains, see Versmissen (1996). 16 There are other differences as well, concerning the unary operators Morrill and Gavarro introduce for the treatment of medial cliticization and clauselocality. The logic of these operators is less restrictive than the pure logic of residuation for our diamond and box operators. As was pointed out in section 6.1
French Object Clitics
311
(cf. also the end of section 2.2.1), such a less restrictive logic can lead to overgeneration. 17 And proposed by Monachesi (1995) for clitic climbing in Italian. 18 Note that lexical assignment of the trigger verb to the higher order type is not an option, as this would result in multiple type-assignments to enable selection of verbal arguments with different complement structures. ^EDITORS' NOTE: The editors of this volume have pointed out to the author that her criticism of HPSG-style analyses is based on some purported "translation" of these analyses into categorial grammar-something that the authors whom Kraak cites have never proposed. To the editors of this volume it, therefore, seems far from clear how HPSG analyses that employ argument-composition techniques are to be fit into this same scheme of evaluation as analyses stated in the framework of categorial grammar. After all, Head-Driven Phrase Structure Grammar constitutes a framework with radically different logical foundations and with radically different assumptions about the underlying grammar formalism and the grammatical constraints imposed on individual grammars.
REFERENCES Abeille, A., D. Godard, and I. A. Sag (1998). Two kinds of composition in French complex predicates In Syntax and Semantics (E. Hinrichs, A. Kathol, and T. Nakazawa, eds.), Vol. 30, 1-41, Academic Press, San Diego. Casadio, C. (1993). A Categorial Approach to Cliticization and Agreement in Italian. CLUEB, Bologna. Dorre, J., and S. Manandhar (1995). On constraint-based Lambek Calculi. In P. Blackburn and M. de Rijke (eds.), Specifying Syntactic Structures. CSLI Publications, Stanford, CA. Title appeared in 1997. Dowty, D. (1985). On recent analyses of the semantics of control. Linguistics and Philosophy 8, 291-331. Girard, J.-Y. (1987). Linear Logic. Theoretical Computer Science 50, 1-102. Hepple, M. (1990). The Grammar and Processing of Order and Dependency: A Categorial Approach. Ph.D. dissertation, Centre for Cognitive Science, University of Edinburgh. Kathol, A. (1995). Linearization-Based German Syntax. Ph.D. dissertation, Ohio State University. Kurtonina, N., and M. Moortgat (1995). Structural control. In P. Blackburn and M. de Rijke (eds.), Specifying Syntactic Structures. CSLI Publications, Stanford, CA. Title appeared in 1997. Lambek, J. (1958). The mathematics of sentence structure. American Mathematical Monthly 65, 154-170. Miller, P. H. (1991). Clitics and Constituents in Phrase Structure Grammar. Ph.D. dissertation, Universiteit Utrecht.
312
Esther Kraak
Miller, P. H., and I. A. Sag (1995). Une analyse lexicaliste des affixes pronominaux en Frangais. Revue Quebecoise de Linguistique 24(1), 135-171. Miller, P. H., and I. A. Sag (1997). French Clitic Movement without Clitics or Movement. Natural Language and Linguistic Theory 15, 573-639. Monachesi, P. (1995). A Grammar of Italian Clitics. Ph.D. dissertation, ITK/TILDIL Dissertation Series 1995-3, Tilburg University. Moortgat, M. (1996). Multimodal linguistic inference. Journal of Logic, Language and Information, 5(3-4), 349-385. (Also in R. Kempson (ed.), Bulletin of the IGPL 3(2,3), 371-401, special issue on Deduction and Language.) Moortgat, M. (1997). Categorial type logics. In J. van Benthem and A. ter Meulen (eds.), Handbook of Logic and Language. Elsevier, Amsterdam. Moortgat, M., and G. Morrill (1991). Heads and Phrases. Type Calculus for Dependency and Constituent Structure. Unpublished manuscript, OTS, Utrecht. Moortgat, M., and R. Oehrle (1994). Adjacency, dependency and order. In Proceedings 9th Amsterdam Colloquium, 447-466. Moortgat, M., and R. Oehrle (1996). Structural abstractions. In V. M. Abrusci and C. Casadio (eds.), Proofs and Linguistic Categories. Applications of Logic to the Analysis and Implementation of Natural Language. Proceedings of the 1996 Roma Workshop, CLUEB, Bologna. Moot, R. (1996). Proof Nets and Labeling for Categorial Grammar Logics. Masters thesis, Universiteit Utrecht. Morrill, G. (1994). Type Logical Grammar. Kluwer, Dordrecht. Morrill, G., and A. Gavarro (1992). Catalan clitics. In A. Lecomte (ed.), Word Order in Categorial Grammar, 211-232, Editions Adosa, Clermont-Ferrand. Morrill, G., N. Leslie, M. Hepple, and G. Barry (1990). Categorial deductions and structural operations. In G. Barry and G. Morrill (eds.), Studies in Categorial Grammar. Edinburgh Working Papers in Cognitive Science, Vol. 5, 1-21, Centre for Cognitive Science, Edinburgh. Nishida, C. (1991). A non-transformational analysis of clitic climbing in Spanish. In Proceedings of WCCFL 9, 395-409. Stabler, E. P. (1996). Acquiring and Parsing Languages with Movement. Forthcoming. Versmissen, J. A. G. (1996). Grammatical Composition: Modes, Models, Modalities. Logical and Linguistic Aspects of Multimodal Categorial Grammars. Ph.D. dissertation, OTS Utrecht. Zwicky, A., and G. Pullum (1983). Cliticization vs. inflection: English n't. Language 59.3, 501-513.
ITALIAN RESTRUCTURING VERBS: A LEXICAL ANALYSIS PAOLA MONACHESI Utrecht Institute of Linguistics OTS Utrecht University Utrecht, The Netherlands
1. INTRODUCTION In Italian, there are some verbs which act as a class with respect to certain phenomena such as clitic climbing, long NP-movement, tough constructions, and auxiliary selection. They are usually called restructuring verbs after Rizzi (1982), who proposed a restructuring rule in order to account for their properties. In the first part of this article, I propose an analysis of these verbs in terms of argument composition. It is based on the idea that the arguments of the complement verb can be raised to become arguments of the restructuring verb. The analysis will be cast within the framework of Head-Driven Phrase Structure Grammar (HPSG) (Pollard and Sag, 1987, 1994) and will capture intuitions similar to those underlying Rizzi's Restructuring Rule (Rizzi, 1982) and the Clause Reduction analysis of Aissen and Perlmutter (1983). A comparison will be made with an alternative account based on the use of nonlocal features and the Nonlocal Feature Principle, which is the mechanism used within HPSG to account for Unbounded Dependency Constructions. However, I show that the argument composition analysis is indeed the most appropriate to provide a uniform account of the properties of the restructuring verbs. Syntax and Semantics, Volume 30 Complex Predicates in Nonderiuational Syntax
313
Copyright © 1998 by Academic Press All rights of reproduction in any form reserved. 0092-4563/98 $25.00
314
Paola Monachesi
The second part of the article addresses the issue of dialectal variation. I consider restructuring verbs in Salentino, a southern Italian dialect, and I focus on one of their properties: clitic climbing. In particular, I consider the behavior of the verb ulire 'want' in the variety of Salentino spoken in the provinces of Brindisi and Taranto. This verb is rather peculiar in that it allows clitic climbing out of a finite clause. I show that the argument composition analysis proposed for Standard Italian can account for this property and I suggest that within HPSG the parameterization of linguistic variation occurs in the lexicon.
2. BASIC FACTS
Rizzi (1982) identifies three classes of restructuring verbs in Italian: (1) Restructuring verbs • modal verbs (e.g., potere 'can,' dovere 'must,' volere 'want') • aspectual verbs (e.g., cominciare 'to begin,' finire 'to finish,' continuare 'to continue') • motion verbs (e.g., venire 'to come,' andare 'to go,' tornare 'to come back') As already mentioned, restructuring verbs exhibit certain properties, among which is clitic climbing. This is one of the most obvious diagnostic revealing complex predicate formation: (2) Anna lo vuole compare. Anna cl.(acc) wants to buy 'Anna wants to buy it.' In this construction, a clitic which originates as dependent of a complement verb can climb and attach to the trigger verb. Therefore, it seems that the two verbs act as a unit with respect to clitic placement and that what we have is a complex verb. In Standard Italian, clitic climbing is optional with these verbs; the following sentence where the clitic stays within the lower clause is also grammatical:1 (3) Anna vuole comprarlo. Anna wants to buy cl.(acc) 'Anna wants to buy it.'
Italian Restructuring Verbs
315
In Italian, there is a rather peculiar clitic, namely, the third person plural pronoun low. There are certain differences between low and the other clitics; however, they pattern alike with respect to climbing in the presence of restructuring verbs. In fact, these verbs also trigger the climbing of low 'to them,' as can be seen in the following example: (4) Non sapevano cos a potesse low capitare. not know what could to them happen 'They didn't know what could happen to them.' It should be noticed that low does not appear before the finite verb, as is the case for the monosyllabic clitic in (2). It is well known that morphosyntactic features determine the direction of attachment for monosyllabic clitics; they are proclitic if the verb is finite and enclitic if the verb is nonfinite or imperative. On the other hand, the placement of low is insensitive to the finiteness of the verb. In fact, this clitic usually follows the verb.2 As in the case of monosyllabic clitics, climbing of low is optional with these verbs. This is shown in the following example, where the clitic stays within the lower clause: (5) Non sapevano cosa potesse capitare low. not know what could happen to them 'They didn't know what could happen to them.' Restructuring verbs have also the property of triggering long NP-movement. If the clitic si, which acts as a passivizing element, is attached to the restructuring verb, the direct object of the verb in the embedded clause acts as subject of the restructuring verb: (6) Queste case si vogliono vendere a caw prezzo. these houses SI want to sell at high price 'These houses are wanted to be sold at a high price.' In this example, queste case originates as direct object of vendere and becomes the subject of si vogliono. Another property which is usually associated with this class of verbs is that they trigger auxiliary change. In fact, restructuring verbs which have avere as auxiliary can change it into essere if the embedded verb requires essere as auxiliary: (7) a. Rocco ha voluto partire. Rocco has wanted to leave 'Rocco has wanted to leave.' b. Rocco e voluto partire. Rocco is wanted to leave 'Rocco has wanted to leave.'
316
Paola Monachesi
Furthermore, in the presence of restructuring verbs, the Italian tough construction which is strictly bounded, appears to be unbounded: (8) Questa canzone e facile da cominciare a cantare. This song is easy to begin to sing In this case, the object of the lower verb questa canzone acts as superficial subject of the higher verb. 3. THE SYNTACTIC STRUCTURE OF RESTRUCTURING VERBS Before showing how an approach in terms of argument composition can account for the different properties of restructuring verbs, I will address the controversial issue of their syntactic structure. Rizzi (1982) proposed a rule of Restructuring to account for clitic climbing and for the other properties exhibited by restructuring verbs. He argued that sentences to which Restructuring has been applied have a different structure from those ones which have not undergone this rule. Restructuring turns a bisentential structure into a simple sentence. Therefore, if no clitic climbing occurs, Rizzi associates a bisentential structure to the sentence Anna vuole comprarlo given in example (3): (9) Anna vuole [s comprarlo]. While for a case of clitic climbing like Anna lo vuole comprare exemplified in (2), there will be a simple structure where the two verbs are reanalyzed into a verbal complex: (10)
Anna [v lo vuole comprare].
Rizzi considers clitic climbing as a sign that Restructuring has been triggered and he suggests that while there is motivation to let the embedded verb and its complements form a constituent if no clitic climbing occurs, this is not the case if clitic climbing is triggered. Rizzi gives four tests, on the basis of which he argues that the embedded verb and the following material do not form a constituent if clitic climbing is triggered. They are summarized here: • Pied-piping applies to an infinitive with following material, but not in a sentence where clitic climbing has occurred: (11) a. Questi argomenti, a parlarti dei quali verrb these topics, to talk cl.(dat) of which will come al piu presto, mi sembrano interessanti. very soon, cl.(dat) seem interesting 'These topics, to talk to you about which I will come as soon as possible, seem very interesting to me.'
Italian Restructuring Verbs
317
b. *Questi argomenti, a parlare del quali ti these topics to talk of which cl.(dat) verrb al piu presto... will come very soon... • Clefting can apply to an infinitive with following material, but not in a sentence where clitic climbing has occurred: (12) a. E proprio a riportargli i soldi che sto andando. is just to bring back cl.(dat) the money that am going 'It is really to bring him back the money that I am about to.' b. *E proprio a riportare i soldi che gli sto andando. is just to bring back the money that cl.(dat) am going • Right Node Raising (RNR) can apply to an infinitive with following material, but not in a sentence where clitic climbing has occurred:3 (13) a. Mario vorrebbe, ma a mio parere non potra mai, Mario would like but to my opinion not can never pagargli il debito. pay cl.(dat) the debt 'Mario would like, but in my opinion will never be able, to pay his debt to him.' b. * Mario gli vorrebbe, ma a mio parere non potra mai Mario cl.(dat) would like but to my opinion not can never pagare il debito. pay the debt • Complex NP shift can postpose an infinitive with following material, but not in a sentence where clitic climbing has occurred: (14) a. Fra qualche giomo verrb a Firenze ad esporti in few day will come to Florence to submit cl.(dat) la mia idea. the my idea 'In a few days, I will come to Florence to submit my idea to you.' b. *Fra qualche giomo ti verrb a Firenze ad esporre in few day cl.(dat) will come to Florence to submit la mia idea. the my idea The constituency tests proposed by Rizzi are based on the idea that constituents can move. The ungrammatical examples where clitic climbing has occurred show that the embedded verb and its complements do not form a constituent since they cannot move. Given an example such as (11),
318
Paola Monachesi
Rizzi considers the following structures as input to wh-mouement: (15) a. Questi argomenti [s verro a parlarti dei quali] alpiu presto. b. Questi argomenti [s ti verro a parlare dei quali] al piu presto. In the former case, the string a parlarti dei quali is a constituent and can be preposed, while the latter case shows an instance of clitic climbing. The string a parlare dei quali is not a constituent and it is for this reason that the ungrammatical (lib) is ruled out. If the embedded verb and its complements do not form a constituent, there are two possible configurations which are available in the case of clitic climbing: the main verb and the embedded verb can merge together in a sort of compound structure or there is a flat structure where the infinitival verb and its complements are sisters of the restructuring verb. Both structures are exemplified below, where the abbreviation CPL stands for complements:
As Rizzi points out, the choice between the two structures is quite difficult since there is not much clear evidence in favor of one or the other. However, it should be noticed that lexical material such as adverbs can intervene between the two verbs, arguing against a verbal compound: (17) Anna lo vuole immediatamente comprare. Anna cl.(acc) wants to immediately buy 'Anna wants to buy it immediately.' Therefore, I will assume that restructuring verbs are associated with the following structures:
Italian Restructuring Verbs
319
The flat structure will be associated with the clitic climbing configuration, while the hierarchical structure will represent those cases where the clitic remains attached to the lower verb.4 It should be noticed that Moore (1991) (a.o.) proposes instead that in Spanish, the embedded verb and the following material do form a VP constituent in cases of clitic climbing. He addresses the issue of the constituency tests and he says that not all of Rizzi's tests are possible in Spanish, but the clefting test is:5 (19) a. Lo que quiero es escribirles una carta. cl. that want is write cl. a letter 'What I want is to write them a letter.' b. *Lo que les quiero es escribir una carta. cl. that cl. want is write a letter The test produces the same result as in Italian, showing that the infinitival and the following material do not form a VP if clitic climbing has occurred. However, he argues against the clefting test, by saying that VPS may not cleft, since clefting may not strand an auxiliary: (20)
*Lo que ha fue llegado tardi. what he has was arrived late
This is also the case in Italian, but following Napoli (1981) it can be argued that this fact constitutes evidence for a flat structure also in the case of auxiliary verbs.6 He also shows that, in Spanish, right node raising is possible with upstairs cliticization. Furthermore, he discusses the coordination test, which is not mentioned by Rizzi; in Spanish it is possible to have a coordination of two VPS with a restructuring verb which has undergone clitic climbing: (21) Juan le quiere mandar un regalo y comprar un coche. Juan cl.(acc) wants send a gift and buy a car 'Juan wants to send her/him a gift and buy her/him a car.' This is also the case in Italian and it constitutes a potential problem for a flat structure: (22) Rocco le vuole spedire un regalo e scrivere una lettera. Rocco cl.(acc) wants to send a gift and write a letter 'Rocco wants to send a gift to her and write a letter to her.' Abeille and Godard (personal communication) suggest analyzing similar cases which occur with French auxiliaries as instances of Non Constituent Coordination. Their suggestion could be adopted also in the case of restructuring verbs; in fact, cases of Non Constituent Coordination are
320
Paola Monachesi
well attested in Italian: (23) Rocco scrivera una lettera a Martina e una cartolina a Vito. Rocco will write a letter to Martina and a postcard to Vito 'Rocco will write a letter to Martina and a postcard to Vito.' Therefore, if example (22) is assimilated to them, it will not constitute a serious problem for an analysis in terms of flat structure.
4. THE ANALYSIS
The data I have presented in section 2 show that restructuring verbs act as a class with respect to a variety of phenomena. In Monachesi (1993a) and in Monachesi (1995), I have suggested that an analysis in terms of argument composition can provide a uniform account of the different properties of these verbs. The idea behind this approach is that the arguments of a verbal complement are raised to become arguments of the restructuring verb, thus creating a verbal complex. The mechanism crucially relies on the notion of structure sharing which is widely used within HPSG. In this case, there will be structure sharing between the complements of the embedded verb and those of the main verb. The mechanism of argument composition was first introduced in HPSG by Hinrichs and Nakazawa (1990) for the treatment of the German verb cluster, and it is currently employed in the analysis of Dutch (Rentier, 1994; van Noord and Bouma, 1994), French (Miller and Sag, 1993; Abeille and Godard, 1994; Miller and Sag, 1995; Abeille et al, 1998), and Polish (Przepiorkowski and Kupsc, 1997). Its effect is similar to that of division in Categorial Grammar (Moortgat, 1988), which has also been used to account for Spanish clitic climbing (Nishida, 1991). In the following sections, I show how the argument composition analysis can account for the relevant properties of Italian restructuring verbs. However, it could be argued that certain properties of these verbs could be better treated as instances of long distance movement operations. Within HPSG, it would be possible to make use of nonlocal features and the Nonlocal Feature Principle, which is the mechanism used to deal with Unbounded Dependencies Constructions. Clitic climbing can be thus analyzed as a case of unbounded dependency. The nonlocal features can be employed to encode the information that a clitic, which originates as argument of the embedded verb, can appear attached to the higher verb. In section 5, I discuss the consequences of such an analysis and show the problems it faces. In particular, specific locality constraints need to be
Italian Restructuring Verbs
321
postulated and certain generalizations are missed if the analysis is extended to account for the various phenomena which are triggered by restructuring verbs. 4.1. Restructuring Verbs and Argument Composition In section 3, I have suggested that two different structures should be associated with restructuring verbs. In fact, restructuring verbs can subcategorize for a VP, but they can also inherit the complements of their infinitival verbal argument. It is the latter configuration that will be necessary to allow clitic climbing. Within HPSG, a lexical rule can be used to establish a relation between these two sets of words:7 (24)
Argument Composition Lexical Rule (ACLR)
The lexical rule relates words, namely restructuring verbs which subcategorize for a saturated VP, with other ones which have the same properties, except that they do not subcategorize for a VP, but for a verbal argument and the complements of the latter. The partition of SYNSEM in phrase synsem (p-ss) and word synsem (w-ss) together with the appropriate feature declaration represents a way to distinguish between words and phrases, alternatively the attribute LEX + / - could be used. In this rule, the notation © stands for the append relation, while the tag GO indicates structure sharing between the elements in the COMPS list of the infinitival and that of the restructuring verb. The condition CUTS { } ensures that argument composition occurs only if the embedded verb does not constitute a cliticized verb form. Complements will be ordered according to Linear Precedence constraints; in particular, the following constraints are relevant in this context: • A lexical head precedes all its complements. • Verbal complements are ordered according to the obliqueness hierarchy. • A lexical verbal complement precedes a phrasal complement.
322
Paola Monachesi
The lexical rule presented above should relate only verbs which belong to the restructuring class. In this work, I will assume that the set of the relevant verbs is specified lexically since there is a big variation among speakers as to which verbs belong to the restructuring class.8 As noticed in Rizzi (1982), for every Italian speaker there is a welldefined class of verbs which allows clitic climbing, and opposed to it there is another class of verbs which clearly does not trigger it. In between the two classes there is a group of verbs for which judgments vary from speaker to speaker, as in the case of the conative verbs (e.g. cercare, tentare, provare 'to try') and sometimes there are isolated verbs which according to some speakers trigger clitic climbing, such as the verb sembrare 'seem.' It seems that membership in the class of trigger verbs is rather idiosyncratic, and this is even more evident if dialects are taken into consideration, as will become clear in section 6.3.1. Attempts to derive the class of trigger verbs semantically, though interesting, are not very convincing since they are not sufficiently developed to capture the idiosyncratic variation which is typical of restructuring verbs. For example, Burzio (1986) suggests that trigger verbs are characterized as being semantically impoverished. He presents the following minimal pair: (25) a. Lo voglio leggere. cl.(acc) want read 'I want to read it.' b. *Lo desidero leggere. cl.(acc) wish read 'I wish to read it.' According to him, volere acts as a trigger verb because it is semantically weak, while this is not the case for desiderare? However, he does not spell out this idea in more detail, so it is difficult to see what its scientific content is. The concept of semantic impoverishment is reconsidered by Emonds (1996), but also in his case no attempt is made to make this idea more concrete. A more formalized attempt to derive the class of trigger verbs semantically is Napoli (1981). She claims that the rule of Restructuring proposed by Rizzi should apply whenever the structural conditions are met, after which a rule of semantic interpretation applies. If the latter rule is blocked for some reason, the sentence does not receive an interpretation and it is therefore judged as not acceptable. According to her, it is through the rule of semantic interpretation that it is possible to determine which verbs act as a trigger and which do not. While I believe that there might be some semantic characterization that defines the set of trigger verbs, I think that the work of Napoli is not supported by enough empirical evidence. How-
Italian Restructuring Verbs
323
ever, her hypothesis is interesting and it would be relevant to devote some research to formalize it into a more up-to-date framework and to test it against a broader set of data.10 In the following sections, I consider the different properties which are triggered by restructuring verbs: clitic climbing, climbing of loro, long NP-movement, tough constructions, and auxiliary selection, and I show how the argument composition lexical rule presented above can provide a uniform account of them. 4.2. Restructuring Verbs and Clitic Climbing The analysis of clitic climbing is based on the assumption that Italian clitics behave as lexically attached inflectional affixes, as argued at length in Monachesi (1995). Therefore clitics will not be considered lexical items which are located in a specific position by the rules of syntax, but featural information which is provided in the lexicon and used in morphology and phonology for the realization of the cliticized verb form.11 However, it should be noticed that Italian clitics are in complementary distribution with full complements, as exemplified below: (26)
*Valentina lo mangia il gelato. Valentina cl.(acc) eats the ice-cream 'Valentina eats the ice-cream.'
Therefore it is necessary to have a mechanism which can provide the necessary featural information and which can ensure the complementary distribution between clitics and full phrases. The following lexical rule, proposed in Abeille et al. (1998) for French and adopted in Monachesi (1995) for Italian, can serve this purpose. It will license cliticized verb forms:12 (27) Complement Cliticization Lexical Rule (CCLR)
The effect of this rule is that the complements of the verb are removed from the COMPS list one at a time and are added as members of the CUTS set.13 If an argument is realized as a clitic it cannot be realized also as a
324
Paola Monachesi
full phrase, accounting in this way for their complementary distribution. The CLTS feature will encode the agreement information relative to the clitic and will act as interface to morphology. Clitic ordering will be dealt with in morphology; in particular the rigid order of Italian clitics can be accounted for by assuming that clitics belong to different position classes. Such a notion can be expressed by means of template morphology (Simpson and Withgott, 1986). The following template should be adopted for Italian: (28) Template for the Italian clitic cluster Position:
I mi ti gli le vi
II ci
III si (ref)
IV lo la li le (ace)
V si (imp)
VI ne
Italian clitics will be thus assigned to different position classes and they will appear in the order prescribed by the template. The lexical rule presented above plays a key role in the analysis of cases of cliticization such as the one exemplified below, where the clitic has not climbed: (29) Anna vuole comprarlo. Anna wants to buy cl.(acc) 'Anna wants to buy it.' The Complement Cliticization Lexical Rule applies to words which are licensed by the lexical entry for comprare, the relevant part of which is shown below:
The effect of the rule is that the direct object will be removed from the COMPS list and it will be added as member of the CLTS set:
The necessary featural information will be available for spelling out the cliticized verb form in phonology. It should be noticed that, in those cases where the clitic has not climbed, the restructuring verb subcategorizes for
Italian Restructuring Verbs
325
a VP. Sentence (29) will receive the following representation:
On the other hand, cases of clitic climbing are accounted for by means of an interaction of the Complement Cliticization Lexical Rule and the Argument Composition Lexical Rule (presented in the previous section), which is repeated here:14 (24) Argument Composition Lexical Rule (ACLR)
Given a sentence such as (33), which exemplifies a case of clitic climbing, (33) Anna lo vuole comprare. Anna cl.(acc) wants to buy 'Anna wants to buy it.' the rule will apply to words which are described by a lexical entry, the relevant part of which is represented as:
The lexical rule will establish a relation with other words, with similar properties, but with crucially different subcategorization requirements. In fact, the complements of the lower verb will be inherited by the restructur-
326
Paola Monachesi
ing verb, creating thus a "clause union" effect. A partial description of vuole after the application of ACLR is:
The CCLR can then apply to license cliticized verbs. Its effect is that the relevant complement will be removed from the COMPS list and added as value of the CLTS set:
The following flat structure will be associated with sentence (33):15
Given a sequence of restructuring verbs, the ACLR would predict that clitics can attach to any of them. This is indeed the case: (38) a. Anna lo vuole poter comprare. Anna cl.(acc) wants to can buy b. Anna vuole poterlo comprare. Anna wants to can cl.(acc) buy c. Anna vuole poter comprarlo. Anna wants to can buy cl.(acc) 'Anna wants to be able to buy it.' In (38a), the ACLR is triggered twice, applying to the two restructuring verbs. Therefore, the finite verb vuole will inherit the complement of the lowest verb which will be realized as a clitic. The result will be a com-
Italian Restructuring Verbs
327
pletely flat structure:
As for (38b), the ACLR will be triggered once; the middle verb subcategorizes for the complement of the embedded verb which is realized as a clitic. The middle verb and the lower one form a constituent, namely a VP which is the complement of the higher one:
The last example is similar to the one presented in (29). In particular, the Complement Cliticization Lexical Rule applies to the lower verb in order to license the cliticized verb comprarlo. Italian clitics tend to cluster together; therefore if there are two clitics which originate as complements of the same verb, they must climb together:16 (41) a. *Valentina lo voleva spedirgli. Valentina cl.(acc) wanted to send cl.(dat) 'Valentina wanted to send it to him.' b. Valentina glielo voleva spedire. Valentina cl.(dat) cl.(acc) wanted to send 'Valentina wanted to send it to him.'
328
Paola Monachesi
Sentence (4la) is ungrammatical because the clitics are split. The condition that the curs set of the embedded verb must be empty, as stated in the output of the lexical rule in (24), enforces that the two clitics must climb together. In fact, the main verb can inherit the complements of the embedded verb only if the verbal complement does not constitute a cliticized verb form. This implies that the curs set of the lower verb must be empty, while this is not the case in (4la) where it contains the information related to the dative clitic. Therefore, the restructuring verb will not be able to inherit the subcategorization requirements of the verbal complement and so it cannot license the presence of the object clitic lo. Clitic climbing is triggered only by certain verbs; therefore a sentence like the following is ungrammatical because promettere does not belong to the class of restructuring verbs: (42)
*Anna lo promette di comprare. Anna cl.(acc) promises to buy 'Anna promises to buy it.'
An advantage of the argument composition approach is that it is lexically constrained; therefore the mechanism is triggered only by restructuring verbs. In the case of the sentence above, the main verb is not lexically marked as a restructuring verb; therefore it can only subcategorize for a saturated VP and not for the arguments of the verbal complement. The presence of the clitic attached to the main verb is thus not licensed, which is the desired result. 4.3. The Bisyllabic Clitic loro Restructuring verbs allow not only for the climbing of monosyllabic clitics, but also for the climbing of the bisyllabic clitic loro 'to them.' Loro, which is the third person plural dative pronoun, has very peculiar properties. On the one hand, it behaves like the other Italian clitics since it occurs in a different position from that of the related complement, it cannot be coordinated, and it cannot be modified, but on the other hand it differs from them in several respects, showing properties of lexical words.17 As already mentioned, the morphosyntactic features of the verb determine the direction of attachment for monosyllabic clitics. On the other hand, the placement of loro is insensitive to the finiteness of the verb; in fact, it always follows: (43) a. Dava loro notizie delle due figlie sposate. gave to them news of the two daughters married '(She) gave them news about the two married daughters.'
329
Italian Restructuring Verbs
b. Una valanga di letters di ammiratrici che gli chiedono an avalanche of letters of fans that cl.(dat) ask di dare loro un figlio. to give to them a child 'An avalanche of letters from fans that ask him to give them a child.' While no material can intervene between monosyllabic clitics and the verb, certain adverbs like spesso 'always' and mat 'never' can separate loro from the verb: (44) // mercato non dam mai loro le somme the market will not give never to them the amounts necessarie per effettuare il pagamento. necessary to make the payment 'The market will never give them the amount necessary to make the payment.' When an accusative and a dative monosyllabic clitic combine in a cluster, they will follow the order dative-accusative. On the other hand, when loro combines with an accusative clitic, the order will be accusative-dative: (45) Martina deve spedirla loro. Martina must send cl.(acc) to them 'Martina must send it to them.' In the previous section, I have shown that, in case of clitic climbing, monosyllabic clitics cannot be split (cf. (41)); however loro does not need to cluster with the other clitics in case of climbing. In addition, while Italian monosyllabic clitic are unstressed, loro bears word primary stress. The following table summarizes the differences between monosyllabic clitics and loro:18
(46) Monosyllabic clitics Position Adjacency
Proclitic/enclitic to the verb Cannot be separated from the verb Ordering Dative-accusative Clitic climbing Cannot be split when they climb Phonology Do not bear stress
Loro Follows the verb Certain adverbs can intervene Accusative-dative Can be split Bears stress
330
Paola Monachesi
In Monachesi (1995), I have argued that it is possible to account for the similarities and the differences between monosyllabic clitics and low, if they are assigned a different status. In fact, the former show affix like behavior while the latter exhibits word like behavior. Therefore, low can be considered an element of category XP[ + CLitic].19 If low is considered a saturated XP, it is possible to account for the fact that it cannot have complements. However, low should be distinguished from other XPS, in that it has different properties; therefore the feature [ + CLitic] will be used to this purpose.20 If low is an XP, its position will be determined by Linear Precedence constraints. The following constraint, which is independently motivated, accounts for the fact that low follows the verb:21 (47) Linear Precedence Constraint 1 (LPI) • A lexical head precedes all its complements. However, this LP constraint would not be sufficient to account for a sentence such as the following, where low precedes the direct object: (48) Una valanga di lettere di ammimtrici che gli chiedono an avalanche of letters of fans that cl.(dat) ask di dare low un figlio. to give to them a child 'An avalanche of letters from fans that ask him to give them a child.' An additional LP constraint can be proposed in order to account for the fact that low precedes all the phrasal complements: (49) Linear Precedence Constraint 2 (LP2) • A complement marked [ + CL] precedes all the other phrasal complements. Given these LP constraints, it is possible to account for the order of low in sentence (48). I refer to Monachesi (1995) for an analysis of the other properties of low discussed in this section. In what follows, I will be mainly concerned with the fact that low can undergo clitic climbing. As already mentioned, low can undergo optional clitic climbing with restructuring verbs: (50) a. Non sapevano cosa potesse low capitare. not know what could to them happen b. Non sapevano cosa potesse capitare low. not know what could happen to them 'They didn't know what could happen to them.'
Italian Restructuring Verbs
331
I will show that the interaction of the Argument Composition Lexical Rule with the LP constraints proposed above can account for the climbing of low. As already shown in the previous section, the output of the ACLR licenses restructuring verbs which subcategorize for a verbal complement and for the arguments of the verbal complement. I repeat the lexical rule here, for convenience:
The output of the lexical rule will thus license a word with the following COMPS list:
Therefore, in example (50) if argument composition has been triggered, the main verb will inherit the complements of capitare. Given the XP status of low, it will be ordered by the LP constraints; in particular, LPI will order low after the main verb. Since capitare is a v, and does not represent a phrasal complement, LP2 will not apply; therefore two orders will be allowed which are both grammatical, one as in (50a), exemplified below:
and another one where low follows the verbal complement as in (50b).22 However, it should be noted that a normal XP is not allowed to climb; it is not possible to have an XP (in this case a Giovanni) in the position
332
Paola Monachesi
occupied by loro: (53)
*Rocco non sapeua cosa potesse a Giovanni capitare. Rocco not know what can to Giovanni happen 'Rocco didn't know what could happen to Giovanni.'
In order to rule out this sentence it is necessary to revise in the following way the LP constraint (introduced in section 4.1) which says that a lexical verbal complement precedes a phrasal complement: •
A lexical verbal complement precedes a phrasal complement which is marked [ — CL].
In this way, sentence (53) will be correctly ruled out since a Giovanni is marked [ - CL]. In fact, in this case there is a verbal complement (capitare) which follows an XP [ - CL] and this is contrary to what is stated in the LP constraint. On the other hand, a sentence such as (50a) where the verbal complement follows the XP loro will be allowed. This is because loro is marked [ + CL], therefore it escapes the constraint mentioned above. As already mentioned, loro does not have to cluster with the other clitics. It is possible to have a monosyllabic clitic attached to the restructuring verb while loro follows the infinitival verb: (54) La pub fomire loro una dichiarazione del ministero. cl.(acc) can provide to them a declaration from the Ministry 'A declaration of the ministry can provide them with it.' This is crucially different from the case where two monosyllabic clitics are present; in fact in that situation clitics cannot be split. This different distribution provides further support for the word status of loro. Since loro is not an affix (unlike monosyllabic clitics), but an XP, there will be no cliticized verb form fomire loro in the lexicon. If the infinitival verb has not combined with a clitic, the curs set will be empty, argument composition will be thus possible, and the restructuring verb will inherit the complements of the infinitival verb. The following COMPS list will be associated with it:
The Complement Cliticization Lexical Rule will license the cliticized verb form la pub where the clitic realizes the object NP. On the other hand, since loro is an XP, it will be ordered according to the Linear Precedence constraints. In this particular case, it complies with LPl, which states that a
Italian Restructuring Verbs
333
complement should follow the head, accounting in this way for the fact that it does not cluster with the monosyllabic clitics. The following representation shows this:
Climbing of loro is possible only with restructuring verbs; therefore the following sentence is ungrammatical because decidere does not belong to this class: (57)
*Rocco decide loro di spedire la lettera. Rocco decides to them to send the letter 'Rocco decides to send the letter to them.'
Since the argument composition mechanism is lexically constrained, only restructuring verbs can inherit the complements of the lower verb. This is not the case of decidere; therefore the presence of loro in that position is not allowed.
4.4. Restructuring Verbs and Long Np-movement Restructuring verbs act as a class also with respect to another phenomenon, namely, long NP-movement. If the clitic si, which acts as a passivizing element, is attached to the restructuring verb, the direct object of the verb in the embedded clause is promoted to subject of the restructuring verb: (58)
Queste case si uogliono uendere a caro prezzo. these houses SI want to sell at high price 'These houses are wanted to be sold at a high price.'
In this example, queste case originates as direct object of vendere and is promoted to subject of si uogliono.
334
Paola Monachesi
The parallelism between this type of construction and the cases of clitic climbing discussed in section 4.2 is straightforward. I have already mentioned that only restructuring verbs can trigger clitic climbing, as shown in (42), repeated here: (42)
*Anna lo promette di comprare. Anna cl.(acc) promises to buy 'Anna promises to buy it.'
The same holds for long NP-movement; the object cannot be promoted to subject of the main verb if it does not belong to the restructuring class: (59)
*Le nuove case popolari si sono promesse di costruire. the new houses council SI are promised to build 'The new council houses are promised to be built.'
Furthermore, if more than one restructuring verb is present, the clitic can appear on the highest verb, as in (38a), repeated here: (38a) Anna lo vuole poter comprare. Anna cl.(acc) wants to can buy 'Anna wants to be able to buy it.' A similar situation can be found in the long NP-movement configuration if more than one restructuring verb is present: (60) Queste case si uogliono poter vendere a caro prezzo. these houses SI want can sell at high price 'These houses are wanted to be able to be sold at a high price.' In this case, queste case, which is the object of vendere, acts as subject of the main verb. Recall that when two clitics are subcategorized by the same verb, if they climb, they cannot be separated, as can be seen in (41a), repeated here: (4la)
*Valentina lo voleva spedirgli. Valentina cl.(acc) wanted to send cl.(dat) 'Valentina wanted to send it to him.'
Similarly, if long NP-movement interacts with clitic climbing, it is not possible to have a sentence where the clitic has not climbed: (61)
*Queste case si uogliono uendergli a caro prezzo. these houses SI want to sell cl.(dat) at high price 'These houses are wanted to be sold to him at a high price.'
Italian Restructuring Verbs
335
In this example, both queste case and gli are complements of the lower verb vendere', it is not possible to have a structure where the NP is preposed while the clitic stays behind. In fact, the clitic should also climb, as in the following grammatical sentence: (62) Queste case gli si vogliono vendere a cam prezzo. these houses cl.(dat) SI want to sell at high price 'These houses are wanted to be sold to him at a high price.' Given the similarities between the two constructions, it would be desirable to have an analysis that could account for both. In section 4.4.2,1 will show that the analysis in terms of argument composition which I have proposed for clitic climbing carries over naturally to the corresponding cases of long NP-movement. In particular, this is achieved through an interaction between the Argument Composition Lexical Rule and the Middle si Lexical Rule which I introduce in the following section.
4.4.1. MIDDLE si Italian clitic si has several different uses. It occurs with an impersonal interpretation, a middle interpretation, an ergative interpretation, and a reflexive-reciprocal interpretation, and it can also be an inherent reflexive clitic. I refer to Monachesi (1995) for an analysis of these different uses of si. In this section, I will only be concerned with middle si; in this case the clitic acts as a grammatical marker and the sentence receives a middle passive interpretation: (63) Gli spaghetti si mangiano spesso. the spaghetti SI eat often 'Spaghetti is often eaten.' This construction, which is fully productive, shares with passive the property that the direct object functions as a superficial subject; in particular it agrees with the verb and it can be replaced by a null subject.23 Therefore, it seems that clitic si acts as a passivizer and its function is similar to that of passive morphology. However, a difference between the middle si construction and the passive construction is that the former does not admit a by-phrase; being in this way similar to agentless passives.24 A lexical rule to account for middle si constructions (MIDSI-LR) can be proposed. This rule, which is similar to the lexical rule for passive, licenses
336
Paola Monachesi
middle verb forms:25 (64) Lexical rule for middle verb forms (MIDSI-LR)
The rule relates transitive verbs with middle passive verb forms.26 It has the effect of removing the subject from the SUBJ list and of promoting the object of the transitive verb to subject of the middle verb. Furthermore, it specifies that the subject in middle si constructions should receive nominative case and should be in the third person. The curs feature will trigger the presence of the clitic attached to the verb.27 4.4.2. THE ANALYSIS OF LONG NP-MOVEMENT In this section, I show that the interaction of the Argument Composition Lexical Rule with the Middle si Lexical Rule accounts in the desired way for those cases of long NP-movement, exemplified in (58), repeated here: (58)
Queste case si vogliono vendere a cam prezzo. these houses SI want to sell at high price 'These houses are wanted to be sold at a high price.'
The restructuring verb will be subject to the application of the ACLR; it will thus inherit the object of the complement verb:
The words licensed by the ACLR can in turn be the input of the Middle si Lexical Rule. The lexical rule will relate words with the COMPS list specified above with other ones where the object has been promoted to subject of the middle verb form:
Italian Restructuring Verbs
337
Therefore, in example (58), queste case which is the object of vendere will be inherited by uogliono and will be eventually promoted to subject of the restructuring verb. Since queste case is a full phrase, the Valence Principle, which is the principle of the grammar responsible for checking off subcategorization requirements that have been satisfied, can normally apply and remove the SUBJ slot. Sentence (58) receives the following representation:
As already mentioned, it is possible to have sentences where more than one restructuring verb is present. In this case, the object of the last verb is promoted to subject of the main verb, as in (60), repeated here: (60) Queste case si vogliono poter vendere a caro prezzo. these houses SI want can sell at high price 'These houses are wanted to be able to be sold at a high price.' The corresponding clitic climbing example, where the clitic climbs more than one clause, was accounted for by applying the Argument Composition Lexical Rule to the two restructuring verbs. Similarly, in the case of long NP-movement exemplified above, a double application of ACLR to the restructuring verbs makes queste case the object of vogliono. The application of the Middle si Lexical Rule makes the object queste case have the function of the subject. However, recall that in the clitic climbing cases, if more than one verb is present, it is also possible for the clitic to attach to the middle verb, as in
338
Paola Monachesi
(38b), repeated here: (38b) Anna vuole poterlo comprare. Anna wants to can cl.(acc) buy 'Anna wants to be able to buy it.' On the other hand, in a long NP-movement configuration, it is not possible for the NP to intervene between the two restructuring verbs, as shown here: (68)
*Si vogliono queste case poter vendere a caro prezzo. SI want these houses can sell at high price 'These houses are wanted to be able to be sold at a high price.'
In this case, the NP has the function of subject; therefore sentence (68) would be ruled out by the LP constraint that requires subjects to be ordered before the verb. Restructuring verbs subcategorize for a verbal complement whose CUTS set is empty; in this way sentences such as: (41 a)
* Valentino lo voleva spedirgli. Valentinocl.(acc) wanted to send cl.(dat) 'Valentino wanted to send it to him.'
are correctly ruled out, as shown in section 4.2. Similarly, sentences such as (61), repeated below, where there is an interaction among clitic climbing, long NP-movement, and middle si are also correctly ruled out in the same way: (61)
* Queste case si vogliono vendergli a caro prezzo. these houses SI want to sell cl.(dat) at high price 'These houses are wanted to be sold to him at a high price.'
The cliticized verb form uendergli does not match the requirements on the verbal complement since the curs set will not be empty; in fact it contains the information related to the dative clitic. Therefore, the restructuring verb cannot subcategorize for the object of the embedded verb; this implies that the appropriate conditions for the application of the Middle si Lexical rule are lacking. As a consequence, in this case no object preposing will be permitted and this is the desired result. On the other hand, sentences such as (62) are correctly allowed: (62) Queste case gli si vogliono vendere a caro prezzo. these houses cl.(dat) SI want to sell at high price These houses are wanted to be sold to him at a high price.' In this case, the output of ACLR will license restructuring verbs which subcategorize for the complements of the embedded verb. The following
Italian Restructuring Verbs
339
specification describes the COMPS list of the verb vogliono after the application of ACLR:
In addition, the Middle si Lexical Rule will relate words with the COMPS list specified above with other ones where the object has the function of the subject, while the clitic si is added as member of the CUTS set:
In addition, the Complement Cliticization Lexical Rule applies to words which are licensed by the lexical entry for vogliono, the relevant parts of which are shown above. Its effect is that the PP complement will be removed from the COMPS list and added as member of the CUTS set:
The sentence will receive the following structure:
340
Paola Monachesi
Like clitic climbing, long NP-movement is triggered only by restructuring verbs; therefore sentences such as (59) are not grammatical since promettere is not a restructuring verb: (59)
*Le nuoue case popolari si sono promesse di costruire. the new houses council SI are promised to build 'The new council houses are promised to be built.'
They are ruled out for the same reasons as sentences such as (42), repeated here: (42)
*Anna lo promette di comprare. Anna cl.(acc) promises to buy 'Anna promises to buy it.'
The reason is that verbs which do not belong to the restructuring class will not be subject to the application of the Argument Composition Lexical Rule. Since argument composition is not allowed, the application of the lexical rule for middle si construction will not be triggered and so there is no way that the object of costruire can become the subject of promettere. Sentences where a verb that does not belong to the restructuring class intervenes between sequences of restructuring verbs, such as: (73)
*Le nuove case popolari si vogliono promettere the new houses council SI want to promise di poter costruire. to be able to build 'The new council houses are wanted to promise to be able to be built.'
are ruled out for the same reasons.
4.5. Tough Constructions Restructuring verbs act as a class also with respect to tough constructions. While in Italian this construction is strictly bounded, it appears to be unbounded if a restructuring verb is present. In tough constructions, the embedded object surfaces as superficial subject, as in the following example from Rizzi (1982): (74) Questo problema e difficile da risolvere. 'This problem is difficult to solve.' It should be noticed that this construction differs from the English equivalent, since in English the direct object can be extracted from almost
Italian Restructuring Verbs
341
any level of embedding:28 (75)
That book will be impossible for you to convince the class to try to finish before Monday.
This is not the case in Italian, where the construction is strictly bounded: (76)
*Questo lavoro e facile da promettere difinire per domani. this work is easy to promise to finish by tomorrow
On the other hand, if a restructuring verb is present, it is possible for the object of an embedded infinitival verb to be promoted to subject of the copular construction: (77)
Questa canzone e facile da cominciare a cantare. this song is easy to begin to sing
The analysis proposed for restructuring verbs can provide an explanation for the possibility of (77) and the impossibility of (76). In example (77), the Argument Composition Lexical Rule will apply to cominciare, which will inherit the complement of the infinitival. Therefore, the object of cantare becomes the object of cominciare. This means that this case of tough construction is reduced to the simple case exemplified in (74). On the other hand, the ungrammaticality of (76) is due to the fact that promettere is not a restructuring verb and therefore the ACLR cannot apply; thus promettere cannot subcategorize for the object of the lower verb. Assuming an analysis of tough constructions along the lines of that proposed by Abeille et al. (1998) for French, adjectives such as facile will subcategorize for a VP with an object in its COMPS list whose index is shared with the subject of the copular construction:29
In example (77), the object of the lower infinitival will become the object of the restructuring verb by means of the argument composition mechanism. Given the specification for facile presented above, this object will be coindexed with the subject of the copular construction accounting in this way for the sentence.30
342
Paola Monachesi
4.6. Auxiliary Selection Restructuring verbs trigger auxiliary change; namely, restructuring verbs which have avere as auxiliary can change it into essere if the embedded verb requires essere as auxiliary: (79) a. Rocco ha voluto partire. Rocco has wanted to leave 'Rocco has wanted to leave.' b. Rocco e voluto partire. Rocco is wanted to leave 'Rocco has wanted to leave.' While this change of auxiliary is optional in the examples above, this is not the case when clitic climbing occurs. As shown in the following sentences, it is the auxiliary of the embedded verb that is realized: (80) a. * Rocco ci ha voluto andare. Rocco cl.(loc) has wanted to go 'Rocco has wanted to go there.' b. Rocco ci e voluto andare. Rocco cl.(loc) is wanted to go 'Rocco has wanted to go there.' More generally, it seems that it is the last verb in the verbal complex that determines which auxiliary is selected, as shown by the following examples where several restructuring verbs are present: (81) a. Rocco li ha voluti andare a cercare. Rocco cl.(acc) has wanted to go to find 'Rocco has wanted to find them.' b. Rocco ci sarebbe voluto cominciare ad andare. Rocco cl.(loc) would be wanted to begin to go 'Rocco would have wanted to begin to go there.' In (81a), volere and cercare require avere as auxiliary, while andare requires essere. In example (81b) both volere and cominciare require avere, while andare requires essere. It should be noted that if the other auxiliary is selected, the sentences above are not acceptable: (82) a. *Rocco li e voluti andare a cercare. Rocco cl.(acc) is wanted to go to find 'Rocco has wanted to find them.' b. *Rocco ci avrebbe voluto cominciare ad andare. Rocco cl.(loc) would have wanted to begin to go 'Rocco would have wanted to begin to go there.'
Italian Restructuring Verbs
343
However, the generalization mentioned above does not seem to hold if the restructuring verb requires essere as auxiliary: (83) Rocco lo e andato a cercare. Rocco cl.(acc) is gone to find 'Rocco has gone to find him.' In this case, cercare selects avere and it is the last verb in the verbal complex, but instead essere is realized. Therefore, it seems that in the presence of a verbal complex, which is reflected by the fact that clitic climbing occurs, the following generalizations hold with respect to auxiliary selection: •
If the restructuring verb selects essere then this is the auxiliary which is realized. • If the restructuring verb selects avere, then it is the auxiliary which is selected by the last verb, which is realized. If auxiliary selection is lexically encoded for each verb, through a feature AUX, it would be possible to account for the generalizations mentioned above. Restructuring verbs which have undergone argument composition and which select essere would simply realize this auxiliary. In the case they require avere, it should be feasible, with an appropriate system of features, to inherit the auxiliary requirements of the last verb in the verbal complex. However, the fact that it is possible to have optional auxiliary change also when the clitics do not climb, is unexpected: (84) a. Rocco ha voluto andarci. Rocco has wanted to go cl.(loc) 'Rocco has wanted to go there.' b. Rocco e voluto andarci. Rocco is wanted to go cl.(loc) 'Rocco has wanted to go there.' In fact, according to the generalizations expressed above, it seems possible to realize the auxiliary of the lower verb only in the presence of a verbal complex.31 An alternative possibility, which could be explored in order to account for the auxiliary selection facts mentioned above, is to take into consideration the role of unaccusativity.32 An analysis in this direction is that of Butt (1995), which proposes an LFG account of this property of restructuring verbs. She suggests a mechanism of argument fusion according to which, when verbs like andare (which require the auxiliary essere) are part of a complex predicate, their theme argument is fused with the highest embedded argument. Furthermore, she assumes that, when this process occurs, the auxiliary essere should be selected. Even though this analysis can
344
Paola Monachesi
account for the relevant data, it is somehow stipulative since it does not explain why it is the auxiliary essere that surfaces when argument fusion occurs and not avere.
5. AN ALTERNATIVE ANALYSIS
In the previous sections, I have presented an analysis of the different properties of restructuring verbs in terms of argument composition. The basic insight behind it is that the restructuring verb inherits the complements of the infinitival one and the two (or more) verbs form a verbal complex. Therefore, a clause union effect is obtained. In this section, I will discuss an alternative analysis of the properties of restructuring verbs, one which will take into consideration the nonlocal character of clitic climbing, long NP-movement, and tough constructions. As already mentioned, it could be possible to account for these phenomena in terms of nonlocal features and the Nonlocal Feature Principle, which have been proposed in Pollard and Sag (1994) to deal with Unbounded Dependency Constructions. In Monachesi (1993c), I have presented an analysis of clitic climbing which indeed relies on this mechanism.33 The treatment makes use of lexical rules to update the subcategorization requirements of the verbal head and of nonlocal features to encode the information that a clitic can appear at some point in the tree. Given a sentence such as (38a), repeated here: (38a) Anna lo vuole poter comprare. Anna cl.(acc) wants to can buy 'Anna wants to be able to buy it.' It will receive the following representation, if an analysis in terms of nonlocal features is assumed:
Italian Restructuring Verbs
345
A verb like comprare will be subject to the Lexical Rule for Object Clitics. Its effect is that one element will be removed from the COMPS list of the verb and added as value of the oc set. This value will percolate by means of the Nonlocal Feature Principle, the percolation will stop once a TO-BIND|OC feature is found. The TO-BIND|OC feature will be present on the cliticized verb form. If the values of the TO-BIND|OC feature and the INHER|OC feature are identical, the nonlocal dependency is bound off according to the Nonlocal Feature Principle: it will be subtracted from the set of nonlocal feature values that are passed up to the mother.34 However, this type of account faces certain problems which are avoided in an analysis in terms of argument composition. An obvious problem is that the analysis in terms of nonlocal features overgenerates, allowing clitic climbing also with verbs that do not trigger it. Therefore, a sentence such as the following, where the main verb is not a restructuring verb, will be accepted, unless specific conditions are added: (42) *Anna lo promette di comprare. Anna cl.(acc) promises to buy 'Anna promises to buy it.' The approach does not account in a straightforward way for the fact that only restructuring verbs can trigger clitic climbing. This is because Unbounded Dependency Constructions are subject to different locality constraints than clitic climbing constructions. In order to rule out sentences such as (42), it is necessary to add specific constraints on the path of the oc feature. The percolation of the feature must be stopped if there is a verb that does not trigger clitic climbing and must be allowed if there is a verb that triggers it. In Monachesi (1993c) this was achieved by imposing the condition that verbs that do not allow clitic climbing should select for a complement with INHER|OC { }. While this could be a possible way to obtain the correct result, it also suggests that this approach does not naturally capture the fact that clitic climbing is triggered only by a specific class of verbs since specific conditions need to be imposed to obtain the right result.35 Furthermore, the constraints need to be imposed both on verbs that do not trigger clitic climbing and on those that trigger it; this is clearly not very economical. In addition, some generalizations are missed if the analysis is extended to account for long NP-movement: (58) Queste case si vogliono vendere a cam prezzo. these houses SI want to sell at high price These houses are wanted to be sold at a high price.'
346
Paola Monachesi
As already mentioned in section 4.4, long NP-movement and clitic climbing are subject to similar constraints; it is thus desirable to have an analysis that could account for both constructions. Therefore, sentence (58) should be analyzed by means of an interaction between the mechanism based on nonlocal features and the rule proposed in section 4.4.1 for middle si constructions. However, there are some problems connected to this. Recall that MIDSI-LR relates two sets of words with different subcategorization requirements; its effect is to promote the direct object of a transitive verb to the subject of the middle verb form. However, given the nonlocal approach presented above, the information about the object of the embedded transitive verb, namely, queste case, would be encoded in the INKER |oc of the verb volere. One could postulate another rule for middle si constructions which relates restructuring verbs; the rule should operate on information encoded in a nonlocal feature and change the grammatical function of the element present in the INHER|OC set, making it a subject. Therefore, there would be two lexical rules in order to account for middle si constructions. One would relate restructuring verbs and would operate on material encoded in the nonlocal feature while MIDSI-LR would relate verbs that do not belong to this class, as in (63), repeated here: (63) Gli spaghetti si mangiano spesso. the spaghetti SI eat often 'Spaghetti is often eaten.' In this case, the rule would effect the subcategorization requirements of the verb; thus, the same phenomenon would be analyzed in two different ways, which is not a desirable result. A similar problem would occur in the analysis of tough constructions: (77) Questa canzone e facile da cominciare a cantare. this song is easy to begin to sing Given an analysis of tough constructions such as that sketched in section 4.5, the adjective facile subcategorizes for a VP with an object in its COMPS list whose index is shared with the subject of the copular construction. However, if nonlocal features are used to account for the properties of restructuring verbs, cominciare will encode the information relative to the object of cantare, not in the COMPS list, but in the INHER|OC set. Therefore, an alternative entry which is associated with the adjective facile should be foreseen to account for the interaction of restructuring verbs with tough constructions. In this entry, there should be a coindexation between the subject of the copular construction and the element contained in the INHER|OC set.
Italian Restructuring Verbs
347
It seems therefore that an approach in terms of nonlocal features does not account for clitic climbing in an adequate way since it does not naturally capture the fact that only a specific class of verbs triggers it. Furthermore, it does not account for long NP-movement and for the interaction of restructuring verbs and tough constructions in an elegant way. On the other hand, an analysis of restructuring verbs which is based on the mechanism of argument composition provides a more adequate treatment since it can naturally account for the intermediate distance character of the constructions involved.36
6. RESTRUCTURING VERBS IN SALENTINO The second part of this paper will be concerned with dialectal variation. I will take into account restructuring verbs in Salentino, a southern Italian dialect discussed in Calabrese (1993). I will focus on one property of restructuring verbs, namely, clitic climbing. In particular, I will consider the behavior of the verb ulire 'want' in that variety of Salentino spoken in the provinces of Brindisi and Taranto. This verb has the peculiar property of triggering clitic climbing out of a finite clause. I will show that, within HPSG, the parameterization of linguistic variation occurs in the lexicon and, in particular, that a lexical approach to clitic climbing allows for a proper treatment of the variation represented by this dialect.
6.1. Clitic Climbing in Salentino As already discussed, clitic climbing with restructuring verbs is optional in Standard Italian, but this is not the case in certain southern Italian dialects, such as Salentino, which is spoken in certain parts of Puglia. In particular, in this section I will be concerned with that variant of the dialect spoken in the northern part of the province of Lecce. In this dialect, restructuring verbs trigger obligatory clitic climbing, as can be seen from the following example:37 (86) a. *Lu pottsu kkattare krai. cl.(acc) canlsgbuyinf tomorrow b. *Pottsu kkattarlu krai. canlsg buyinj cl.(acc) tomorrow 'I can buy it tomorrow.'
348
Paola Monachesi
These data can receive an account under the argument composition analysis proposed for restructuring verbs in Standard Italian. The same mechanism will account for clitic climbing in both cases. Verbs that trigger clitic climbing will thus subcategorize for a verbal complement and for the arguments of the latter, as shown in the following lexical entry for pottsu:38
Therefore, while in Standard Italian, it is necessary to formulate a lexical rule in order to account for the optionality of clitic climbing, in Salentino a lexical entry would suffice. Each clitic climbing trigger verb will be described by a lexical entry similar to the one presented above. I assume the Complement Cliticization Lexical Rule, which I have adopted for Standard Italian and which licenses cliticized verb forms to be operative in this dialect as well. The restriction on curs in the lexical entry (87), namely, curs must be empty, accounts for obligatory climbing. A sentence such as (86b) where the clitic is attached to the lower verb is ruled out. In fact, in that case the CLTS set will contain the information related to the clitic; therefore it will not be empty and will thus not fulfill the condition imposed by the lexical entry above. It should be mentioned that there is variation with respect to which elements belong to the class of restructuring verbs in Standard Italian and in Salentino. For example, while a verb like volere 'want' is a restructuring verb in Standard Italian, this is not the case in Salentino (at least in the variant spoken in the province of Lecce). However, a crucial difference between Standard Italian and Salentino is the fact that in Italian volere allows for the presence of an infinitival complement if there is coreference between the subject of the matrix clause and that of the embedded clause: (88)
Carlo uuole venire domani. Carl wants to come tomorrow 'Carl wants to come tomorrow.'
Italian Restructuring Verbs
349
This is not the case in Salentino, where a verb like volere cannot subcategorize for an infinitival complement: (89)
*Lu Karlu ole inire krai. the Carl wants 3sg to comeinf tomorrow 'Carl wants to come tomorrow.'
More generally, in Salentino there is a class of verbs, namely, verbs of ordering, desiring, warning, etc. (of which ulire 'want' is a member), that cannot be followed by an infinitival complement. They select instead a finite clause which is introduced by the element ku, as can be seen in sentence (90):39 (90) Lu Karlu ole ku bbene krai. the Carl wants35g KU comes inrf3ig tomorrow 'Carl wants him to come tomorrow.' In particular, Calabrese (1993) suggests that ku introduces the complement of verbs which express an attitude toward, or an attempt to bring about, an event which is yet to come. The status of the element ku, which introduces finite clauses after this particular class of verbs, is rather controversial since it exhibits both properties of inflectional affixes and of complementizers.40 The idiosyncratic behavior of ku can be easily understood if diachronic data is taken into consideration; in fact ku was originally a complementizer. As Calabrese notices, ku derives from Latin quod, which was a complementizer: (91) Luvat me quod studia vigent. pleases me that studies flourish 'I am pleased that studies are flourishing.' However, as will become clear from the following discussion, synchronically, ku is becoming an inflectional affix connected to tense morphology. Yet, the process is not completed; therefore it still maintains some of its original properties of complementizers. 6.1.1. ON THE STATUS OF THE ELEMENT ku Evidence in favor of the status of ku as inflectional affix comes from the following properties: • It attaches to the verb and only clitics can separate it from the verb. • It constrains tense morphology. • It cannot have wide scope over coordination.
350
Paola Monachesi
The first property is exemplified as follows: (92) a. *Oyyu ku lu Maryu bbene krai. want lig KU the Mario comesind, 3sg tomorrow 'I want Mario to come tomorrow.' b. Oyyu ku nnu le kkatti. want lig KU not cl.(acc) buyind>2sg 'I want you not to buy them.' Sentence (92a) shows that no phrasal complement can separate ku from the verb; only clitics may intervene, as in (92b).41 The element ku constrains tense morphology; in fact, when it is present, there is a violation of the morphological agreement of tenses between the main clause and the embedded clause, which is generally assumed to be obeyed in Standard Italian. For example, the presence of a past tense in the main clause does not imply a past tense in the clause introduced by ku; in fact, only the present and the perfect may be used in this case, as shown in the following example: (93) la ulutu la Maria ku bbae/ia ™*cond.past, 1sg wanted the Maria KU go P r e s . / p e r f , 'I wanted Maria to have gone there before.'
3sg
f/iuta ddai. there
More generally, ku seems to introduce clauses characterized by a nondeictic tense; in fact, as noticed by Calabrese (1993), the tense of the embedded clause must be bound by the tense of the matrix clause in order to get a time reference.42 Additional evidence in favor of the affixal status of ku comes from coordination; in fact, in this case, the element must be repeated in front of each conjunct, as the following example from Melillo (1975) shows: (94)
Tokka ku ffacimu festa e kku ne prisamu. must KU make^ip, party and KU cl. enjoyind, 1pl 'We should have a party and enjoy it.'
However, ku exhibits also additional properties which could argue in favor of its status of complementizer: •
It is in complementary distribution with the complementizer ka.
•
It is incompatible with a wh-expression.
•
It precedes negation.
Italian Restructuring Verbs
351
The first property is shown by the following example, which demonstrates that ku and the complementizer ka cannot cooccur in the same sentence: (95)
*Voleva ka la Maria ku bbene. wanted 3ig that the Mary KU comesind >3sg 'He wanted that Mary would come.'
This seems to suggest that they are in complementary distribution because they are both complementizers. The position of negation also argues in favor of the complementizer status of ku, as can be seen from example (92b), repeated here: (92b) Oyyu ku nnu le kkatti. want ug KUnot cl.(acc)buyjW>2jg 'I want you not to buy them.' It should be noticed that in Standard Italian negation follows the complementizer che: (96) Martina dice che non vena alia festa. Martina says that not will come to the party 'Martina says that she will not come to the party.' Furthermore, ku is incompatible with a wh-expression: (97) a. *Me sta ddumannu nt/e ku add3u fare. cl. aux. wonder1sg what KU have.nrf ls_ to do lagind,1sgb. Me sta ddumannu ntje add3u fare. cl. aux. wonder what have to do 'I am wondering what to do.' In this respect it shares the same behavior as the complementizer ka: (98)
*Me sta ddumannu nt je ka add3u fare. cl. aux. wonder1sg what that have lcff to do 'I am wondering what to do.' 13
6
13
5
The following table summarizes the properties of ku:
(99) Inflectional affix properties Attaches to the verb; only clitics can intervene Constrains tense morphology No wide scope over coordination
Complementizer properties Complementary distribution with the complementizer ka Incompatible with wh-expression Precedes negation
352
Paola Monachesi
Therefore, it seems that ku has an intermediate status between that of an inflectional affix and of a complementizer. 6.2. An HPSG Analysis In this section, I propose an analysis of the data presented above. In particular, I will be concerned with the status of the element ku and how it should be represented within HPSG. Given that it shares both properties of inflectional affixes and of complementizers, I can see three possible ways of accounting for it: 1. ku is considered an inflectional affix; therefore the analysis proposed for object clitics in Standard Italian can be extended to account for it. 2. ku is considered a complementizer; therefore it will have the status of a marker and it attaches to a VP. 3. ku is a complementizer, thus a marker, but it attaches to a v. If the first possibility is considered, it would be easy to account for the fact that ku attaches to the verb and nothing but clitics can intervene. However, it is not evident how to account in a principled way for the fact that ku is in complementary distribution with the complementizer ka. On the other hand, if ku is considered a VP marker it would be straightforward to account for the complementary distribution with ka, but one would predict that it is possible for adverbs to intervene between ku and the verb, while this does not seem to be the case.43 The third option, which views ku as a complementizer, thus a marker which attaches to v, seems thus the most appropriate way to account for the hybrid status of this element. In fact, it incorporates its complementizer properties and would also account for the fact that ku is a verbal affix. According to Pollard and Sag (1994), complementizers are considered markers.** In fact, they mark the constituent in which they occur; they are distinguished by the attribute MARKING which they structure share with the mother node. Markers can select the phrase with which they combine through the feature SPEC. Therefore, the following will be the entry for ku:
In addition, verbs of order, desire, etc. should subcategorize for a finite sentence introduced by ku. However, it should be noted that, in Salentino,
Italian Restructuring Verbs
353
finite clauses introduced by ku (or ka) cannot be considered control structures, as argued in Calabrese (1993). He provides examples such as the following, where he shows that the empty category in the subject position of these clauses doesn't need to be bound by an antecedent in the matrix clause: (101) Lu Maryu ole ku ej bbete bbona. the Mario wants3sg KU is3sg good3F5g 'Mario wants her to be good.' Calabrese considers it a free pronominal that can be related to a referent in the extralinguistic context or to an antecedent in the discourse. Furthermore, he argues that sentences such as the following cannot be cases of Exceptional Case Marking: (102) Oyyu lu Maryu ku le kkatta krai. want lig the Mario KU cl.(acc) buysind, 3sg tomorrow 'I want Mario to buy them tomorrow.' In fact, the subject preceding ku must be assigned nominative case, as shown in the following example where the nominative pronoun idda 'she' occurs: (103) Oyyu idda ku bbene krai. want ljg she (nom) KU comes/nrf 35g tomorrow 'I want her to come tomorrow.' Therefore, the lexical entries for verbs of order, desire, etc. should contain the following description:
A sentence such as (102) will thus receive the following structure:
354
Paola Monachesi
It should be mentioned that the percolation of the MARKING value from the marker up to the s is ensured by the Marking Principle from Pollard and Sag (1994): (106)
MARKING Principle In a headed phrase, the MARKING value is token identical with that of the marker daughter if any, and with that of the head daughter otherwise.
Therefore, the resulting sentence will have ku as value for MARKING and will thus satisfy the condition in the lexical entry for ulire in (104). As already mentioned, no overt subject can intervene between ku and the verb: (92a) *Oyyu ku lu Maryu bbene krai. want 1sp KU the Mario comesind, tomorrow lig i r i U j3sg JAg 'I want Mario to come tomorrow.' This fact can be easily accounted for under the analysis proposed. Since ku selects for a v and not for an s, it will not be possible to have the subject in that position. In addition, under this treatment no adverbs are expected to occur between ku and the verb, which is the correct result. It should be recalled that ku is in complementary distribution with the complementizer ka; therefore a sentence such as (95), repeated here, where ku which is introduced by the verb ulire and ka cooccur, is not grammatical: (95) *Ulia ka la Maria ku bbene. wanted3ig that the Mary KU comesind, 3sg 'He wanted that Mary would come.' The complementizer ka will be also considered a marker which has the following lexical entry:
Since ka subcategorizes for an s with the value for MARKING equal to unmarked, the sentence above would be ruled out because it will have ku
Italian Restructuring Verbs
355
as value. The following representation exemplifies this:
Therefore, examples such as (95) would be ruled out in the same way as an English sentence such as the following: (109)
*/ think that that John left.
In a similar way, it is possible to rule out sentences such as the following, where ku and ka cooccur, but where the main verb subcategorizes for an embedded clause introduced by ka: (110)
*Sperava ka la Maria ku bbene. hoped3sg that the Mary KU comesind, 3sg 'He hoped that Mary would come.'
As discussed in Monachesi (1995) it would be more problematic to rule out sentences such as the one above if ku is considered an inflectional affix. 6.3. Clitic Climbing and Finite Clauses Unlike Standard Italian, clitic climbing does not occur with a verb such as ulire 'want'; in fact a sentence such as the following would be ungrammatical in Salentino of Lecce: (111)
*Le ole ku kkatta. cl.(acc) wants3igKU buyind, 3sg 'He wants to buy them.'
The ungrammaticality of this sentence could in some sense be expected since generally clitic climbing occurs out of a nonfinite clause, but in this case there is a finite one. However, as will become clear in the next section, this cannot be the reason. Instead, I suggest that clitic climbing cannot occur in (111), because there is no obligatory control. In fact, as I mentioned in the previous section, in Salentino, finite clauses introduced by ku cannot be considered control structures. I show in the next section that the reason for not having clitic climbing in sentence (111) cannot be related to the presence of a finite verb and that indeed obligatory control is a crucial condition for the occurrence of clitic climbing.
356
Paola Monachesi
6.3.1. SALENTINO OF BRINDISI AND TARANTO Calabrese (1993) mentions in a footnote data from a variety of Salentino which is spoken in the provinces of Brindisi and Taranto. This variety of Salentino is similar in all relevant respects to the variety previously discussed, but it has an additional peculiar property. In fact, if the matrix verb is ulire 'want' and if there is coreference between the subject of the matrix clause and the subject of the embedded clause, the element ku can be omitted: (112) a. Voggyu ku kkattu nu milune. wantl5g KU buy,^
Italian Restructuring Verbs
357
Therefore, clitic climbing with the verb ulire is optional in this variety. On the other hand, if ku is present, clitic climbing is not allowed (115b): (115) a.
Voggyu want lsg 'I want b. *Lu cl.(acc) 'I want
ku lu kkattu. KU cl.(acc) buy,. nd>ljg to buy it.' uoggyu ku kkattu. wantj^ KU buyind>lxg to buy it.'
The data presented above seem to suggest that, in this variety of Salentino, clitic climbing is allowed with the verb ulire when there is obligatory control. The obligatory control configuration is reflected in the omission of the element ku. The lexical analysis of clitic climbing in terms of argument composition proposed for Standard Italian can be adopted to account for the data presented above. The fact that, in this variant of the dialect, ulire 'want' is the only verb in its class to trigger clitic climbing is accounted for by the fact that argument composition is lexically constrained. Parametric variation is thus handled in the lexicon. Therefore, if there is coreference between the subject of ulire and that of the embedded verb, the lexical entry should contain the following description:
The verb will subcategorize for a finite verb and its complements; this operation of argument composition is indicated by the tag 2. The coreference between the subjects is represented by the presence of the tag CD. It should be noted that the conditions imposed in the lexical entry above are similar to those in the output of the Argument Composition Lexical Rule proposed for Standard Italian.
358
Paola Monachesi
In this way, it is possible to account for a sentence such as (114a) where clitic climbing occurs: (114a) Lu uoggyu kkattu. cl.(acc) want^buy^^ 'I want to buy it.' It will receive the following structure:
This representation is similar to that provided for cases of clitic climbing in Standard Italian; the only crucial difference with respect to Standard Italian is that in this case two finite verbs will form a verbal complex.47 On the other hand, clitic climbing is blocked if ku is present: (115b) *Lu uogguy ku kkattu. cl.(acc) want1sg KU buy^^ 'I want to buy it.' Sentences such as (115b) are ruled out by the condition MARKING unmarked specified in (116). In other words, argument composition, thus clitic climbing, is possible only if the verbal complement has not combined with the marker ku. Clitic climbing with the verb ulire is optional; in fact clitics can also attach to the embedded verb, both if ku is present and if it is omitted. Therefore, there will also be a lexical entry where ulire subcategorizes for a saturated VP:
Italian Restructuring Verbs
359
In this case, it is required that the subject of the embedded verb and that of the main verb should be referential. It should be noted that the conditions in the lexical entry above are similar to those imposed on words licensed by the input of the Argument Composition Lexical Rule proposed for Standard Italian. The entry will account for those cases where the clitic has not climbed:48 (114b) Voggyu lu kkattu. want ljg cl.(acc) buylW>1,g 'I want to buy it.' (115a) Voggyu ku lu kkattu. want 1sg KU cl.(acc) buy/B
As already mentioned, the marker ku can be optionally omitted only in a control configuration; therefore a sentence such as the following would not be grammatical since there the element ku is not present and there is no subject coreference: (120)
*Maryu uoli vueni Karlu. Mario wants,.. comesind, Carl ind ,.„ JAg 3sg 'Mario wants Carl to come.'
This sentence would be ruled out because it does not satisfy the conditions in either of the entries proposed for ulire; in fact it is not possible to license the presence of the subject. If there is no coreference between the subjects, the element ku must be present, as shown in the following example: (121) Maryu voli ku vveni Mario wants 3sg KU comesind 'Mario wants Carl to come.'
3sg
Karlu. Carl
360
Paola Monachesi
This suggests that ulire subcategorizes also for a sentence introduced by ku, as in the other variety of Salentino taken into consideration. Therefore, also in this dialect, the verb ulire like other verbs of order, desire, etc. will be associated with the following lexical entry, proposed in section 6.2:
The entry requires that the verb subcategorizes for a finite s introduced by the complementizer ku and it will account correctly for sentence (121).
7. CONCLUSION In this article, I have proposed an analysis of Italian restructuring verbs in terms of argument composition. It is based on the idea that the complements of the infinitival verb are raised to become complements of the restructuring verb. I have shown that this mechanism allows for a uniform account of the different properties exhibited by these verbs: clitic climbing, long NP-movement, tough construction, and auxiliary selection. In particular, I have shown that the analysis in terms of argument composition interacts in the appropriate way with the analysis of cliticization, providing an adequate account of the different aspects of clitic climbing. The differences and similarities of monosyllabic clitics and of loro with respect to climbing are derived by assuming that they have a different status. Monosyllabic clitics show affixlike behavior, while loro exhibits wordlike behavior. Furthermore, I have provided an account for the impossibility of having split clitics in Italian, a fact which has never received a clear treatment under traditional generative approaches such as that of Rizzi (1982). I have also shown that the analysis of restructuring verbs in terms of argument composition interacts with the treatment proposed for middle si constructions, allowing for an elegant account of long NP-movement. In particular, the analysis can capture in a natural way the parallelism attested between clitic climbing and long NP-movement. Furthermore, the analysis proposed can provide an explanation for the fact that the Italian tough construction, which is strictly bounded, appears to be unbounded in the presence of restructuring verbs. I have also shown that the argument composition approach is superior to an approach in terms of nonlocal features. In the former, no special
Italian Restructuring Verbs
361
constraints need to be postulated in order to account for the fact that only restructuring verbs exhibit the properties mentioned above. An advantage of the approach proposed is that it can be easily extended to account for dialectal variation. In the second part of the article, I have considered restructuring verbs in Salentino, focussing mainly on clitic climbing. I have shown that the argument composition mechanism can be employed to account for the peculiar fact that in this dialect, the verb ulire 'want' allows for clitic climbing out of a finite clause.
ACKNOWLEDGMENTS This paper is based on my dissertation and thus the acknowledgments made there hold here as well. In addition, I would like to thank Tilman Hohle and Detmar Meurers for illuminating discussions about the use of lexical rules. Thanks also to Erhard Hinrichs, Tilman Hohle, Adam Przepiorkowski, Frank Richter, and two anonymous reviewers for the valuable comments they made on an earlier draft of this paper.
NOTES In general there is no difference in meaning between the two sentences. However, see Napoli (1981), for a potential counterexample. 2 This is the generalization which has been assumed in traditional grammars and in previous studies of this clitic. However, a corpus analysis has revealed that loro can optionally precede the verb in certain contexts, namely, within a relative clause and in the presence of an adnominal past participle. See Monachesi (1995) for further details. 3 While the second example is considered ungrammatical in Rizzi (1982), it seems acceptable to me. On the other hand, it is not clear if RNR can be considered a reliable test for constituency. See Abbott (1976) for English. 4 A relevant question in relation to the flat structure is whether it makes the right predictions with respect to the scope of adjuncts. Given that the treatment of adjuncts in HPSG is still a topic of much debate and that a clear theory is lacking, it is difficult to provide an adequate answer. 5 Examples are from Moore (1991). 6 Compare Abeille and Godard (1994) and Emonds (1996), who also propose a flat structure for French auxiliary verbs. 7 Lexical rules have been used quite extensively in the analysis of several phenomena in the HPSG literature; however, their formalization is still a matter of debate. There are at least two possible ways in which the functionality of lexical
362
Paola Monachesi
rules could be formalized: either as meta-descriptions relating lexical entries (Calcagno, 1995) (MLR) or as descriptions relating word objects (Meurers, 1995) (DLR). Hohle (1995) has shown that the metalevel formalization leads to wrong results if the Complement Extraction Lexical Rule (CELR) (Pollard and Sag, 1994) is used to extract complements from verbs which have undergone argument composition. Similar problems would arise if the Complement Cliticization Lexical Rule (CCLR), adopted in this work, applies to verbs which have undergone argument composition. However, Meurers (1995) has shown that under the DLR approach it is possible to avoid the unwanted results. It is this interpretation of lexical rules that is assumed in this paper. In addition, it should be noted that since in the DLR approach word objects are related, issues of instantiation do not play a role. Therefore restructuring verbs should be marked by a special feature which should be mentioned also in the left-hand side of the ACLR. In this way, it is possible to ensure that the rule applies only to restructuring verbs. However, an alternative worth exploring is that of replacing the lexical rule by an underspecified lexical entry for each verb belonging to the restructuring class, while the relevant generalizations are expressed by means of lexical principles. 9 1 disagree with Burzio's judgment; in fact, I would consider example (25b) grammatical. 10 See Fontana (1996) for a more recent attempt to derive the class of trigger verbs semantically. 11 This implies that there will be no entries related to the clitics in the lexicon. 1 A previous version of this rule which was adopted in Monachesi (1993a,b) was proposed in Miller and Sag (1993). 13 It should be mentioned that the notation [U clitic in (27) stands for a SYNSEM of sort clitic. This information plays a role in the account of past participle agreement which is triggered by object clitics and in the case of binding theory. 14 Given the lexical rule, two structures will be created if the subcategorization slots are filled by complements instead of clitics: a flat one and one where the lower verb and its complements form a constituent. However, data concerning long NP-movement, tough constructions, and auxiliary selection show that they are both necessary also when clitics are not present and that it is not possible to restrict the application of the rule so that only arguments which are realized as clitics are raised. Skytte et al. (1991) notice that there are regional differences in the use of the two structures. In the dialects of central Italy and Toscana, both structures are present, while in the Northern Italian dialects only the hierarchical structure is attested and in the Southern Italian dialects, only the flat one. 15 It should be mentioned that the argument composition approach and the flat structure which I have assumed in the case of clitic climbing are independent of each other. Therefore, it would be possible to maintain an analysis of clitic climbing in terms of argument composition even if new evidence were to reveal that the flat structure is not the correct one. 16 This is not always the case in other Romance languages. For example, in French causative constructions, a reflexive clitic remains attached to the infinitival
Italian Restructuring Verbs
363
verb, while object clitics, which correspond to complements of the infinitival, attach to the causative verb. Kayne (1991) mentions that split clitics are attested also in Franco-Provengal. In constructions where there is an auxiliary and a past participle, it is possible to have one clitic attached to the former and another one attached to the latter. 17 The exceptional behavior of low could be related to its phonological shape. In fact, loro is bisyllabic while all the other Italian clitics are monosyllabic. See Wanner (1987) and Nespor (1994) for a similar assumption. See Monachesi (1995) for a detailed comparison between monosyllabic clitics and loro. 19 Cardinaletti (1991) points out that loro can be marginally doubled by a monosyllabic clitic, behaving in this respect as a maximal projection. The feature could be employed also in the analysis of similar elements, which have been attested in other languages. See Cardinaletti and Starke (1996) for a typological study. 21 An additional LP constraint should be specified to order the subject before a phrasal head. This treatment produces spurious ambiguity since a sentence such as (50b) will be also licensed by the input of the ACLR. 23 See Manzini (1983) for a more detailed discussion of the subject status of the element in preverbal position. 24 Cinque (1988) notes that in more rhetorical styles of Italian it is possible to find a by-phrase cooccurring with si, and Zubizarreta (1987) notices that there are languages in which passive cannot take an agentive by-phrase. 25 See also Grimshaw (1982) for an analysis in terms of lexical rules of the French se moyen within the LFG framework. The rule applies to finite verbs and generally it cannot apply to infinitivals. However, it can apply to infinitivals in raising constructions, as in the following example: (i) Possono trovarsi nuovi articoli. can find SI new articles 'It is possible to find new articles.' Therefore, additional conditions need to be specified in order to avoid the application of the rule in the general cases where an infinitival is present. 2/ As for the semantics of middle verb forms, it should be mentioned that the superficial subject is linked to the object argument position, while the subject argument position is filled by an external agent which, however, does not surface syntactically. 28 However, Grover (1995) argues against the idea that English tough constructions should be considered nonlocal dependencies. 29 The unsaturated VP complement should be licensed by an appropriate ID schema. 30 An alternative approach which could also account for Italian tough constructions is presented in Grover (1995) for English. She argues against a nonlocal
364
Paola Monachesi
treatment of English tough constructions and she proposes an account where the object is promoted by a lexical rule and becomes the subject of the infinitival verb. In this way, the infinitival complement will have two elements in the SUBJ list, the most oblique of which (the promoted object) will be structure-shared with the subject of the adjective. However, while I think that an analysis in terms of nonlocal features is completely unmotivated for Italian and that an approach in terms of object promotion is more appropriate, I think that the option of having more elements in a SUBJ list is rather controversial. However, it should be noticed that the grammatical status of this example is questionable (cf. also Rizzi, 1982, footnote 26). In fact, it is given as acceptable in Skytte et al. (1991) and as ungrammatical in Calabrese (1988). Since there is quite a lot of dialectal variation associated with auxiliary selection, the contrasting judgments could be attributed to it. It should be noted that other attempts to account for the auxiliary selection of restructuring verbs, such as Rizzi (1982) and Burzio (1986), have not been able to provide an explanation for the example (84b). 32 However, Centineo (1986) challenges the Unaccusative Hypothesis Burzio (1986) and argues that auxiliary selection in Italian is determined by the lexical semantic properties of the verb and not by their configurational properties. 33 See Miller (1992) for a similar analysis of clitic climbing triggered by French auxiliary and causative verbs within a GPSG/HPSG framework. 34 In Monachesi (1993c) I have assumed a hierarchical structure in case of clitic climbing, as in the representation sketched here. However, this position has been abandoned in subsequent work, given the motivations presented in section 3. 35 The analysis of clitic climbing presented in Miller (1992) in terms of FOOT features runs into a similar problem. Therefore also in his account specific constraints are added on the FOOT feature specifications to obtain the correct result. 36 A reviewer suggests that there could be a lexical alternative to the structural percolation of nonlocal features, which has been employed in the analysis discussed here. In such an account, the flow of information would be regulated by the restructuring verb. I think that this could be an interesting possibility worth exploring; however, it will imply a revision of the way nonlocal features are introduced and propagated, which is not immediate. I believe that an analysis along these lines is that proposed by Kraak (1998) in the framework of multimodal categorial grammar. 37 A11 the examples of Salentino discussed here are from Calabrese (1993). 38 This entry is similar to the one proposed for auxiliaries in Standard Italian which also trigger obligatory clitic climbing. I refer to Monachesi (1993b, 1995) for an analysis of auxiliaries in Italian. 39 The finite complement can either have an indicative or a subjunctive verb. 40 It should be mentioned that an equivalent of ku can be found in other languages, such as Greek (na), Albanian (te), and Rumanian (sa) (cf. Terzi, 1992). 41 Sentence (92a) also shows that ku behaves differently from complementizers, since no overt subject can intervene between it and the verb. Instead, the subject
Italian Restructuring Verbs
365
must precede ku or it can be postverbal. Therefore, ku behaves differently from the element ka, whose status is uncontroversially that of a complementizer. This element introduces finite clauses which are embedded under a verb of saying or believing, and in this case the subject can intervene between ka and the verb, as expected with a complementizer. The characteristics of ka are therefore similar to those of Italian che and of English that. 42 The behavior of ku is different from that of the complementizer ka; if the latter is present, there is always morphological agreement of tenses. 43 I refer to Monachesi (1995) for a more detailed discussion of the advantages and disadvantages of the two solutions sketched here. 44 However, see Sag (1997) for a different analysis, where the complementizer is viewed as a head. 45 It should be noted that elements equivalent to ku across Balkan languages cannot be omitted, suggesting that ku might indeed be different from them. Terzi also mentions that the Greek equivalent of ku, the particle na, is always obligatory while its antecedent in earlier stages ina which was uncontroversially a complementizer could be omitted. 46 The peculiarity of this variation of Salentino with respect to clitic climbing is also recorded in AIS 1086 (Jaberg and Jud, 1940), where the variants la voggyu ttakku and voyu lla ttakku 'I want to attach it' are mentioned. 47 At first sight, one might think that there could be a relation between this construction and serial verb constructions. As Veenstra (1996) points out, serial verb constructions are characterized by the fact that they contain two verbs which have: • only one expressed subject; • one specification for tense/aspect; • only one possible negator; • no intervening coordinating conjunction; • no intervening subordinating conjunction; • no intervening pause possible. The Salentino construction discussed in section 6.3.1, complies with the properties just mentioned; however, there is a crucial difference with respect to the way complements are ordered. In fact, serial verb construction have typically the format: (i)
N?! Y! NP2 V2 (XP3)
while this is not the case for the Salentino construction. Another crucial difference is that in Salentino the verb ulire subcategorizes for a verbal complement, while this is not the case in serial verb constructions. 48 It should be noted that since in the entry for ulire presented in (118) the mechanism of argument composition is not triggered, a sentence such as (115b) would not be allowed by this entry. This is the correct result.
366
Paola Monachesi
REFERENCES Abbott, B. (1976). Right node raising as a test for constituenthood. Linguistic Inquiry 4, 639-642. Abeille, A., and D. Godard (1994). The complementation of French auxiliaries. In Proceedings of the Thirteenth West Coast Conference on Formal Linguistics, 157-173. CSLI Publications, Stanford, CA. Abeille, A., D. Godard, P. Miller, and I. Sag (1998). French bounded dependencies. In S. Balari and L. Dini (eds.), HPSG in Romance. CSLI Publications, Stanford, CA. Aissen, J., and D. Perlmutter (1983). Clause reduction in Spanish. In D. Perlmutter (ed.), Studies in Relational Grammar, 360-403. Chicago University Press, Chicago. Burzio, L. (1986). Italian Syntax. Reidel, Dordrecht. Butt, M. (1995). The Structure of Complex Predicates in Urdu. Dissertation in Linguistics, CSLI, Stanford, CA. Calabrese, A. (1988). I pronomi clitici. In L. Renzi (ed.), Grande Grammatica Italiana di Consultazione, Vol. 1, 549-592. II Mulino, Bologna. Calabrese, A. (1993). The sentential complementation of Salentino: A study of a language without infinitival clauses. In A. Belletti (ed.), Syntactic Theory and the Dialects of Italy, 28-98. Rosenberg & Sellier, Torino. Calcagno, M. (1995). Interpreting lexical rules. In Proceedings of the Conference on Formal Grammar, Barcelona. Cardinaletti, A. (1991). On pronoun movement: The Italian dative low. Probus 3(2), 127-153. Cardinaletti, A., and M. Starke (1996). The typology of structural deficiency. On the three grammatical classes. In H. Van Riemsdijk (ed.), Clitics in the Languages of Europe. Language Typology, Vol. 8. de Gruyter, Berlin. Centineo, G. (1986). A lexical theory of auxiliary selection in Italian. In Davis Working Papers in Linguistics, 1-35. University of California, Davis. Cinque, G. (1988). On Si constructions and the theory of Arb. Linguistic Inquiry 19, 521-581. Emonds, J. (1996). How clitics license null phrases: A theory of the lexical interface. In H. Van Riemsdijk (ed.), Clitics in the Languages of Europe. Language Typology, Vol. 8. de Gruyter, Berlin. Fontana, J. (1996). On the Sources of Variation in Clause Reduction Phenomena. Talk given at the ESSLLI workshop on Surface-Based Syntax and Romance Languages, Prague. Grimshaw, J. (1982). On the lexical representation of Romance reflexive clitics. In J. Bresnan (ed.), The Mental Representation of Grammatical Relations, 87-148. MIT Press, Cambridge, MA. Grover, C. (1995). Rethinking some empty categories: Missing objects and parasitic gaps in HPSG. Ph.D. thesis, University of Essex.
Italian Restructuring Verbs
367
Hinrichs, E., and T. Nakazawa (1990). Subcategorization and VP structure in German. In Hughes, Shaun and Salmons (eds.), Proceedings of the Third Symposium on Germanic Linguistics. Benjamins, Amsterdam. Hohle, T. (1995). The Complement Extraction Lexical Rule and Variable Argument Raising. Talk given at the HPSG workshop, Tubingen. Jaberg, K., and J. Jud (1940). Sprach- und Sachatlas Italiens und der Sudschweiz. Ringier, Zofingen. Kayne, R. (1991). Romance clitics, verb movement and PRO. Linguistic Inquiry 22(4), 647-686. Kraak, E. (1998). A deductive account of French object clitics. In E. Hinrichs, A. Kathol and T. Nakazawa (eds.), Complex Predicates in Nonderiuational Syntax. Syntax and Semantics, Vol. 30, 271-312. Academic Press, San Diego. Manzini, M. R. (1983). Restructuring and Reanalysis. Ph.D. Thesis, Massachusetts Institute of Technology, Cambridge, MA. Melillo, M. (1975). Le Strutture Verbali dei Dialetti di Puglia. Universita degli studi di Bari. Meurers, D. (1995). Towards a semantics for lexical rules as used in HPSG. In Proceedings of the Conference on Formal Grammar, Barcelona. Miller, P. (1992). Clitics and Constituents in Phrase Structure Grammar. Garland, New York. Miller, P., and I. Sag (1993). French Clitic Movement without Clitics or Movement. Paper presented at the LSA annual meeting, Los Angeles. Miller, P., and I. Sag (1995). French Clitic Movement without Clitics or Movement. Unpublished manuscript, Lille and Stanford. (French version in Revue Quebecoise de Linguistique) Monachesi, P. (1993a). Restructuring verbs in Italian HPSG grammar. In K. Beals, G. Cooke, D. Kathman, S. Kita, K. McCullough, and D. Testen (eds.), Proceedings of the 29th Regional Meeting of the Chicago Linguistic Society, Chicago, 281-295. Monachesi, P. (1993b). Object clitics and clitic climbing in Italian HPSG grammar. In Proceedings of the Sixth Conference of the European Chapter of the Association for Computational Linguistics, Utrecht, 437-442. Monachesi, P. (1993c). The use of nonlocal features in the analysis of Italian object clitics and clitic climbing. In W. Sijtsma and O. Zweekhorst (eds.), Proceedings of the Third CLIN Meeting, Tilburg, 71-82. Monachesi, P. (1995). A Grammar of Italian Clitics. ITK Dissertation Series 1995-3 and TILDIL Dissertation Series 1995-3, Tilburg University. Moore, J. (1991). Reduced Constructions in Spanish. Ph.D. thesis, University of Southern California. Moortgat, M. (1988). Categorial Investigations: Logical and Linguistic Aspects of the Lambek Calculus. Foris Publications, Dordrecht. Napoli, D. (1981). Semantic interpretation vs. lexical governance: Clitic climbing in Italian. Language 57(4), 841-887.
368
Paola Monachesi
Nespor, M. (1994). The phonology of Clitic Groups. In L. Hellan and H. Van Riemsdijk (eds.), Clitic Doubling and Clitic Group. EUROTYP working papers, 67-90. Nishida, C. (1991). A non-transformational analysis of clitic climbing in Spanish. In A. Halpern (ed.), Proceedings of the Ninth West Coast Conference on Formal Linguistics, Stanford, 395-409. Pollard, C., and I. Sag (1987). Information-Based Syntax and Semantics. Fundamentals, Vol. 1 CSLI Publications, Stanford, CA. Pollard, C., and I, Sag (1994). Head-Driven Phrase Structure Grammar. University of Chicago Press, Chicago, and CSLI Publications, Stanford, CA. Przepiorkowski, A., and A. Kupsc (1997). Negative concord in Polish. Unpublished manuscript, Tubingen and Warsaw. Rentier, G. (1994). A lexicalist approach to Dutch cross serial dependencies. In K. Beals, J. Denton, R. Knippen, L. Melnar, H. Suzuki, and E. Zeinfeld (eds.), Proceedings of the 30th Regional Meeting of the Chicago Linguistic Society, Chicago, 376-390. Rizzi, L. (1982). Issues in Italian Syntax. Foris Publications, Dordrecht. Sag, I. (1997). English relative clause constructions. Journal of Linguistics 33(2), 431-484. Simpson, J., and M. Withgott (1986). Pronominal clitic clusters and templates. In H. Borer (ed.), The Syntax of Pronominal Clitics. Syntax and Semantics, Vol. 19, 149-174. Academic Press, Orlando, FL. Skytte, G., G. Salvi, and M. R. Manzini (1991). Frasi subordinate all' infinito. In L. Renzi, and G. Salvi (eds.), Grande Grammatica Italiana di Consultazione, Vol. 2, 483-569. II Mulino, Bologna. Terzi, A. (1992). PRO in Finite Clauses. A Study of the Inflectional Heads of the Balkan Languages. Ph.D. Thesis, New York. Terzi, A. (1995). Two Types of Clitic Climbing out of Finite Clauses. Unpublished manuscript, CUNY and Rutgers University, van Noord, G., and G. Bouma (1994). Adjuncts and the processing of lexical rules. In Proceedings of COLING 94, Kyoto, 250-256. Veenstra, T. (1996). Serial Verbs in Saramaccan: Predication and Creole Genesis. HIL dissertation, Ph.D. Thesis, Universiteit van Amsterdam. Wanner, D. (1987). The Development of Romance Clitic Pronouns, de Gruyter, Berlin. Zubizarreta, M. L. (1987). Levels of Representation in the Lexicon and in the Syntax. Foris Publications, Dordrecht.
CAUSATIVES AND THE NATURE OF ARGUMENT STRUCTURE GERT WEBELHUTH Department of Linguistics University of North Carolina at Chapel Hill Chapel Hill, North Carolina
1. INTRODUCTION1
This article concerns itself with the question of the "unithood" of causatives in a number of senses of this term. Inspection of causatives from several languages shows that (i) causative predicates can be expressed either synthetically (i.e., as single morphological words) or analytically (i.e., through a combination of words that do not form a morphological constituent together) and (ii) causative predicates can behave as if all of their grammatical arguments form a single clause with respect to binding, passivization, etc. or as if they form two different clauses, and they can show mixed behavior; i.e., some causatives behave in some respects as if a single clause is present whereas in other respects a biclausal analysis is indicated. As will be shown, the data available in the linguistic literature indicate that all the logically possible combinations of surface expressions and clausality behaviors that were discussed in the previous paragraph are instantiated in some natural language. The table in (1) gives an overview of the languages that data will be cited from in support of these typological Syntax and Semantics, Volume 30 Complex Predicates in Nonderivational Syntax
369
Copyright © 1998 by Academic Press All rights of reproduction in any form reserved. 0092-4563/98 $25.00
370
Gert Webelhuth
generalizations: (1)
Monoclausal Biclausal Mixed
Analytic German I German II Italian
Synthetic Malayalam Chi-Mwi:ni Turkish
The picture in (1) is of great theoretical relevance for the design of syntactic theories. First of all, the existence of mixed causatives raises the obvious issue of how a syntactic theory can simultaneously ascribe monoclausal and biclausal properties to a construction. One way in which this challenge has been met traditionally is through the postulation of various strata of syntactic analysis that make it possible for a construction to destructively alter its properties from one syntactic level of representation to the next. For both conceptual and computational reasons such nonmonotonic operations are problematic, however, and ideally we would like to do without them. From the point of view of the theoretical linguist the data that (1) sums up is also important in that each of the three causative types can be projected both from synthetically and analytically expressed predicates. As Ackerman and Webelhuth (1998) demonstrate in detail, causatives in this regard are representative of many other notions expressed by natural language predicates, e.g., tense, aspect, passive, middles, and modalities. The regularity with which the small class of notions including the ones mentioned in the previous paragraph recurs in the world's languages suggests that they have a privileged status in the grammars of individual languages. Even if one knows nothing else about some natural language, one does not run a big risk by predicting that it will have predicates whose meaning involves the notions of tense-aspect and that it is likely to express various modalities and causation. The same degree of predictability does not attach to the way the predicates carrying these notions will be expressed, however. Thus, even though a language may express tense, aspect, or causatives, if one knows nothing else about the typological properties of the language, one will not be able to predict whether these notions are expressed with the use of affixes or by adding auxiliaries to a main verb. This asymmetry between a set of notions expressed by predicates that recur regularly in the world's languages and our inability to predict whether the relevant predicate will be expressed synthetically or analytically in any given language leads us to the following considerations. Linguistic theory would miss an important generalization if it failed to
Causatives and Argument Structure
371
capture the fact that certain notions are likely to be expressed in natural languages. Moreover, as we shall see below, while the surface expression of a predicate expressing a certain notion is not in general predictable, other grammatical properties of the clauses containing such a predicate may be predictable! Again, to be explanatorily adequate, linguistic theory should capture these generalizations. Finally, linguistic theory should explain why it is that causatives in particular and predicate notions in general can be freely expressed either in synthetic or in analytic surface forms. The issues raised in the last paragraph thus lead to the following practical design problems to be solved by the grammar writer: (2) • How can the predicate notions which recur with systematic frequency in the world's languages be given a privileged status that explains their systematic recurrence with the same or similar properties in language after language? • How can grammars be designed so that languages can make use of the same predicate notion but are free to express the notion either synthetically or analytically in a language-particular fashion (or following the typological tendencies of the language)? The remainder of this paper is an attempt to find plausible answers to these two questions in the domain of causatives. The solution that will be sketched for this circumscribed grammatical domain is not dependent on any properties of causatives, however. It can therefore be seen as one special application of a general theory of predicate formation that deals with the issues in (2). Such a general theory is worked out in formal detail in Ackerman and Webelhuth (1998), to which the reader is referred. The present article will limit its scope to causatives and will also not attempt a full formalization of the theory. Instead, we will work out an informal sketch of the theoretical commitments that are necessary in a satisfactory approach to causatives in light of the data in (1) and the theoretical considerations in (2).
2. THE DATA FROM CAUSATIVES 2.1. Terminology and Pretheoretical Diagnostics for Clausality It will be best to agree on some terminological conventions and some pretheoretical diagnostics for monoclausality vs biclausality before we approach different causative constructions and compare them. Since we
372
Gert Webelhuth
will primarily be dealing with causatives formed from transitive verbs, the resulting causative predicate will contain information about up to three semantic arguments. We will base our terminology on causatives that embed a predicate with a clearly distinguished active human agent argument and a passive nonhuman patient argument: (3)
Sue Human eats the
fishNon.human.
In a causative construction such as (4) we have an additional argument: (4) MaryCauser causes SueCausee to eat fishPatient. We will refer to Mary as "the causer," to Sue as "the causee," and to fish as the "patient," "the patient of the lower verb," or "the lower patient." In constructions where the lower predicate's arguments are not clearly distinct in terms of agentivity or humanness, we will apply the terms "causee" and "patient" in analogy with the behavior of similarly expressed predicates that draw the distinctions clearly. As an added typographic signal, we will underline the causee and use boldface roman for the patient in all relevant examples. We will apply three pretheoretical diagnostics of monoclausality and four diagnostics of biclausality in the creation of clausality profiles for different causative constructions.2 The first diagnostic of monoclausality is that the causative predicate has the grammatical-functional profile of a generic monoclausal three-place predicate that denotes a single state of affairs. The prototype of such a verb is give, which in many languages governs three distinct functions: a subject, a direct object, and an indirect object-oblique. If causative predicates behave like typical verbs give in this respect, this diagnostic will identify them as monoclausal. Second, there are many empirical reasons for postulating that some semantic and/or grammatical arguments of a predicate are in some sense more closely related to the predicate than others.3 This can be reflected in phrase-structural differences; e.g., the verb in English forms a phrase with its objects which the subject argument is external to, making the subject argument less closely related to the verb than its objects. It can also be reflected in terms of idiom formation in that patient arguments typically form idioms more easily with verbs than agent arguments, etc. As a whole, these criteria converge on an ordering that establishes agentive subjects as being less closely bound to the predicate than patient objects. From this point of view, the three arguments of causativized predicates can be ranked as follows: the patient is the innermost argument, the causee is less closely bound to the embedded predicate, since it is an agent, and the
Causatives and Argument Structure
373
causer is least closely bound, since it is not even an argument of the embedded predicate. The two arguments farthest away from each other in this hierarchy are the causer (an argument of CAUSE) and the lower patient (the innermost argument of the embedded predicate). We will use the possibilities of binding-theoretic interaction between these two arguments as another diagnostic. Since anaphoric binding is typically constrained to applying to arguments of the same clause in the world's languages, if in a causative construction the two arguments of the causative that are most distant from each other can enter into an anaphoric binding relationship, then we will take this to be a symptom of monoclausality. (5) illustrates this case: (5)
tciause-l NPCauser-i V NPCausee (V) NP patient .i]
Assuming the final NP to be an anaphor that is bound by the causer across the intervening causee, this is tentative evidence that all three NPS appear in one and the same clause. Finally, we invoke the theoretical assumption that passive is universally defined as the demotion of an underlying subject. In the unmarked case, the underlying direct object of the same predicate becomes the surface subject of the passive predicate. In other words, the unmarked passivization construction creates single clauses. Consider a causative construction such as (5) in this light: if the causative in (5) can be passivized with the consequence that the causer is demoted from a subject to an oblique and the patient is promoted to the new surface subject, then this is evidence that all three arguments of the original causative predicate appear in one clause. These, then, are our three diagnostics for monoclausality. We will sum them up in a table once we have discussed the tests for biclausality. If each of these diagnostics returns a positive result for a causative construction, we take that to be strong evidence that the construction is monoclausal. The first of the four diagnostics of biclausality acts as a counterpart of the first diagnostic for monoclausality: if we find doubled grammatical functions in the causative predicate, e.g., two direct objects or two indirect objects, then we will take that to be an indication that two clauses are present, under the assumption that each clause contains only a single instantiation of each nonoblique grammatical function. According to this diagnostic, the English construction in (6) would be biclausal, because both the causee and the patient can be shown to behave like direct objects: (6) / made her read it. (6) is an instance of the two-predicate raising-to-object construction, in which the subject of the underlying predicate read is structure-shared
374
Gert Webelhuth
("has been raised") with the object function of the causative predicate. While the two clauses overlap because of this functional identification, they have not been collapsed into a single clause and the lower "caused" clause contains a subject distinct from the subject of the "cause" clause. The presence of such a biclausal structure with a discrete subject in the lower clause may also be indicated by binding-theoretic facts. Thus, if in a structure similar to (5) the causer is unable to bind the patient argument, then the reason is possibly that the structure needs to be parsed into two clauses as in (7) rather than just one as in (5): '')
Lciause-l
NP
Causer-i
V
l-Clause-2
NP
Causee-k 'V'
NP
Patient-*i/k-U
Since binding is typically clause-bound, the structure in (7) is one way of naturally accounting for the inability of the causer to bind the patient. We thus again get the picture typical of subject-to-object raising sentences: (8)
*The woman i made Kim call herself
The English causative construction thus behaves biclausally both with respect to the doubling of grammatical functions and the binding theory. The third diagnostic for biclausaliry is the counterpart of the third diagnostic for monoclausality. Recall that we argued that passivization is clause-bound and hence that if the patient as the innermost argument of the causative predicate can become the surface subject of the passivized causative, this suggests that the causative predicate projects all its semantic arguments into a single clause. Conversely, if that patient argument cannot become the surface subject of the passivized causative, then one reason for this may be that there really are two clauses and what is being passivized is the upper clause. Since the patient does not bear a grammatical function in the upper clause, it is barred from promoting to the subject of the passivized causative. We can illustrate this feature again with the English subject-to-object raising causative construction. Compare the following three examples: (9)
Mary made S«eCausee criticize HelenPatienl.
(10)
SweCausee was made to criticize Helen Patient .
(11)
*HelenPatient was made SueCausee to criticize.
Passivization of the active causative (9) yields the result in (10) rather than (11): the causee and not the patient of the embedded predicate becomes the surface subject of the passivized causative. Postulating a biclausal structure for (9) immediately accounts for this paradigm: since
Causatives and Argument Structure
375
the causee but not the patient bears the grammatical function of direct object in the upper clause, it will become promoted to the subject function if the upper clause is passivized. Finding a paradigm of this kind in a causative construction thus is another diagnostic for biclausality. The fourth and final diagnostic for biclausality comes from binding theory once again. It is well known that subjects make very good binders cross-linguistically. Consequently, there are languages in which subjects are the only grammatical function that can antecede anaphors in constructions where there is no doubt about how many clauses are present or what grammatical functions the arguments present in the sentence bear. Dalrymple (1993:11) cites the Marathi reflexive swataah as an example of an anaphor requiring a subject as its antecedent. Assume, then, that we have a language with an anaphor that must be anteceded by a subject and we lexicalize the patient argument of a causative construction with that anaphor. If only the causer can antecede the anaphor, then we will assume that only one clause is present and that the causer is the subject of that clause. On the other hand, if either only the causee or both the causee and the causer can antecede the anaphor in the innermost argument position, then we have reason to believe that the causee is a subject. Since the causer is always a subject of active causative constructions, the construction must thus have two subjects. Given that we are assuming that nonoblique grammatical functions cannot be doubled within clauses, this state of affairs indicates the presence of two clauses with the causer the subject of the higher clause and the causee the subject of the lower one. From here on, we will frequently abbreviate "monoclausal property" as "1CLP" and biclausal property as "2CLP." The reader should be aware that sometimes not all diagnostics can be applied to a given construction for independent reasons and that we also sometimes lack relevant data. It should also be kept in mind that the diagnostics are just pieces of circumstantial evidence meant to help us develop an analysis. The more diagnostics converge on the same result, the more confident we can be that a construction is really monoclausal or biclausal. If only a single diagnostic gives us a positive result in one direction, then we should use this piece of evidence only very cautiously. The following three sections will be concerned with purely monoclausal, purely biclausal, and mixed causatives, respectively. It will be shown that the functional property of clausality is independent of surface realization, since for each of the three clausality types we can find languages that express that type through a synthetic predicate and other languages that express the type analytically.
376
Gert Webelhuth
2.2. Purely Monoclausal Causatives This section has two subsections that each deal with a purely monoclausal causative construction. They differ in that in German the predicate of this construction is expressed through an auxiliary-main verb combination whereas Malayalam uses a single synthetic word form. 2.2.1. GERMAN I Compare the regular noncausative transitive sentence (12) with the causativized counterpart in (13) which is one of the two German causative constructions: (12) Ellen hat den Jungen gekammt. Ellen has the boy combed 'Ellen combed the boy.' (13) Ellen liefl den Jungen von seiner Schwester kammen. Ellen caused the boy by Els sister comb 'Ellen had the boy combed by his sister.' According to our terminological convention, Ellen is the causer, von seiner Schwester is the causee, and den Jungen is the lower patient. (13) has the grammatical-functional profile of a three-place verb: the mapping from semantic roles to grammatical functions is as in (14): (!4)
sucauser OBLCausee OBJ Patient (= 1CLP)
The absence of any doubled grammatical function in a causativized transitive is a monoclausal property. In all other respects, the construction in (13) behaves like a single clause as well. We illustrate this with anaphoric binding. As Reis (1976) has shown, the German anaphor sich must be anteceded by the subject of the minimal clause containing it:4 (15)
*Maria glaubt [daft JEllen *die Frau Maria believes that Ellen-su the woman-Do sich gezeigt hat]. herself-io shown has 'Maria believes that Ellen showed the woman to herself.'
(16)
*Maria glaubt [daft J Ellen *der Frau Maria believes that Ellen-su the woman-IO sich gezeigt hat]. himself-vo shown has 'Maria believes that Ellen showed the woman herself.'
Causatives and Argument Structure
377
(15) and (16) both contain the ditransitive predicate zeigen 'to show.' In the first example, we have lexicalized the indirect object of this predicate with the reflexive sich. Of the three underlined functions in (15), only the subject of the embedded clause can be understood as binding sich, whereas the direct object of the lower clause as well as the subject of the main clause are not possible antecedents. In (16) we reflexivized the direct object of the predicate of the lower clause and as before, only the subject of its own clause can antecede the anaphor. Assuming that the anaphor sich obeys Reis' Generalization, this data is immediately accounted for: the two objects in (15) and (16) cannot antecede the reflexive because the reflexive must be bound by a subject and the main clause subject cannot do so because it is not contained in the same minimal clause that contains the reflexive. Armed with this generalization about sich, we can now return to the causative construction and see what happens when we lexicalize the lower patient with sich: (17) Elleni liei sich i/*k von ihrer Schwesterk kdmmen. Ellen caused herself by her sister comb 'Ellen caused her sister to comb herself.' (18)
Ellent heft von ihrer Schwesterk sich;/*k kdmmen. Ellen caused by her sister herself comb 'Ellen caused her sister to comb herself.'
Independently of the word order we choose for the causee and the lower patient, only the causer can antecede sich. According to our diagnostics, this is double evidence for a monoclausal analysis of this construction: since the causer Ellen can bind sich, these two expressions must occur in the same clause if sich behaves in causatives as it does elsewhere. Second, the fact that the causee her sister cannot antecede sich suggests that the causee is not a subject because otherwise it would have to be the subject of the minimal sentence containing sich in (17)-(18). The binding facts thus unambiguously favor a monoclausal analysis of this German causative construction. In sum, the causative construction we have called German I passes two tests for monoclausality and none for biclausality. We therefore take it to be an instance of a purely monoclausal causative construction. As all the examples show, this construction is formed by a combination of the auxiliary lassen and the bare infinitive of the main verb. Moreover, the auxiliary and the main verb can be separated from each other in phrase structure, e.g., wnen the auxiliary that serves as the categorial head of the predicate must appear in the second position of main clauses, as in (17).5
378
Gert Webelhuth
2.2.2. MALAYALAM6
Malayalam has a purely monoclausal causative construction as well. Interestingly, in this language the causative predicate is expressed by one single morphological word. In (19) we provide an example of a simple transitive sentence: (19) kutti aanaye nulli child-NOM elephant-Ace pinched The child pinched an elephant.' As in the German construction just discussed, the causative predicate formed from the predicate of (19) governs three distinct grammatical functions, namely, a subject, a direct object, and an oblique. This is a monoclausal property: (20) amma kuttiyekkonte aanaye nulliccu mother child-iNST elephant-Ace pinch-CAUSE-PAST 'Mother made the child pinch the elephant.' Second, the causative predicate can be passivized and under passivization the lower patient rather than the causee becomes the subject of the passive predicate: (21) ammayaal aana nullikkappettu mother-iNST elephant-NOM pinch-CAUSE-PASS-PAST 'The elephant was caused by mother to be pinched.' Since the lower patient, the innermost argument of the embedded predicate, can become the subject of the passive clause and we assume in general that passive is clause-bound, this is further evidence for the monoclausal nature of the Malayalam causative construction. Finally, like German, Malayalam has an anaphor that is subjectoriented. In the sentence below, the possessive reflexive can only take the subject Johnny as its antecedent and cannot be bound by the direct object Mary: (22) jooni meehye swantam wiittil weccs umma weccu Johnny Mary self s house-LOC at kiss put 'Johnny kissed Mary at Johnny's/*Mary's house.' The binding facts in (23) fall out from this generalization and the assumption that the causative consists of only one clause with only one
379
Causatives and Argument Structure
subject which expresses the causer mother: (23) amma kuttiyekkonte aanaye mother child-iNST elephant-Ace nulliccu
swantam wiittil self s house
weccd at
pinch-CAUSE-PAST
'Mother made the child pinch the elephant at mother's/*child's/*elephant's house.' The causee and the lower patient cannot antecede the possessive reflexive because the single clause cannot have more than one subject and the subject function is already taken by the causer. The Malayalam causative construction thus displays only monoclausal properties. Functional-semantically it behaves like German I, but there is one crucial difference between the two constructions: in Malayalam the causative predicate is expressed by one single morphologically integrated word form whereas German uses a combination of two phrase-structurally independent word forms.
2.3. Purely Biclausal Causatives Now we will encounter causative constructions that behave in all respects as if they contain two clauses. They show the typical properties of raising-to-object constructions which identify the object of the higher clause with the subject of the clause immediately embedded under the higher clause. We will refer to the higher clause as the "cause clause" and the lower one as the "caused clause." An analytic version of this construction is presented in section 2.3.1 and a synthetic version in section 2.3.2.
2.3.1. GERMAN II German has a second causative construction apart from the one we saw in section 2.2.1. Like the first construction, however, the present one expresses its predicate with the auxiliary lassen and the bare infinitive of the main verb. Note the difference between (24) and the earlier example (13), both of which are causatives based on the predicate of the simple noncausative sentence (12): (24) Ellen liefi den Vater den Jungen kdmmen. Ellen made the father the boy comb 'Ellen made the father comb the boy.'
380
Gert Webelhuth
Unlike (13), (24) expresses both the causee and the lower patient as noun phrases [(13) expressed the causee as an oblique], which is a first hint that these two elements bear the same grammatical function, albeit in different clauses. Second, both NFS in (24) are accusative, the case typical of direct objects in German, and they show the same word order behavior relative to other constituents. There is thus reason to assume that this causative construction contains two direct objects, one that expresses the causee and one for the lower object. Assuming that each clause can have at most one direct object, this is evidence of the presence of two clauses in this construction which are related by the functional-semantic analog of subject-to-object raising. If the conclusion in the previous paragraph is correct, then the subjectbound anaphor sich should behave differently in the construction we are calling German II than in the previously discussed construction German I. For, if the overt direct object of the cause clause really is identified with the subject of the caused clause, then the subject of the minimal clause containing sich in the lower object position should be the causee rather than the causer. In other words, the construction should behave like (8) with the lower object anteceded by the surface direct object but not the surface subject which functionally is "one clause too far away" from the anaphor. This prediction is borne out: (25) Hansi Heft den Jungenk sichk/* kammen. Hans made the boy himself comb 'Hans made the boy comb himself.' Compare (25) with (17) and (18), where the binding facts were just the opposite. 2.3.2. CHI-MWI:NI7 The Chi-Mwi:ni causative is interesting because functional-semantically it behaves exactly like the purely biclausal construction we just saw but differs from it in that the predicate of the construction is expressed on the surface as one single morphological word consisting of a verb stem, a causative affix, and various other pieces of morphology. As we understand it, the components of this word are not separable under any syntactic or phrase-structural conditions. As a first indication of biclausality, note that the causee and the lower patient are both expressed as simple noun phrases in the causative (26): (26) Mwa:limu 0-wa-andik-ish-iz-e wa\na xati. teacher SP-OP-write-CAUSE-T/A children letter 'The teacher made the children write a letter.'
381
Causatives and Argument Structure
This is compatible with (26) being an instance of subject-to-object raising in which the causee is expressed as the direct object of the cause clause and the lower patient as the direct object of the caused clause.8 Further evidence for this structure is provided by passivization. As the difference between (27) and (28) indicates, only the causee but not the lower patient can become the subject when the predicate in (26) is passivized: (27)
(28)
Wa:na wa-dndik-ish-iz-a: xati na mwailimu. children SP-write-CAUSE-PASS-T / A letter by teacher 'The children were made to write a letter by the teacher.' *Xati a-dndik-ish-iz-a wa:na na mwailimu. letter write children by teacher 'The letter was made to be written by the children by the teacher.'
(26)-(28) are parallel to the English sentences (9)-(11) whose grammatically judgments we had assumed to follow from the presence of two clauses with semantic identity between the object of the higher and the subject of the lower clause. Passivization thus provides powerful evidence for the biclausality of the Chi-Mwi:ni causative construction. The binding theory, finally, falls into place as well. In both (29) and (30) we have lexicalized the lower patient with an anaphor that reportedly follows the same binding principle as the German reflexive sich: it must be bound by the subject of the minimal clause containing it. In a sentence such as (29), only the surface direct object causee which acts as the subject of the minimal clause containing the anaphor can be interpreted as anteceding the reflexive. The surface subject is "one clause too far away": (29)
M/j [ni-m-big-ish-iz-e] mwa'.na^ I-SUBJ SP-OP-hit-CAUSE-T/A child-OBJ T made the child hit himself.'
ru:hu-y-e*i/k. himself
If we replace the third person anaphor in (29) with a first person anaphor to make it featurally compatible only with the first person subject of the cause clause and incompatible with the third person direct object of the cause clause, then the sentence becomes ungrammatical: (30)
M/J [ni-m-big-ish-iz-e] Alik ru:hu-y-a*i/*k. PSUBJ SP-op-hit-CAUSE-T/A Ah-OBj myself 'I made Ali hit myself."
382
Gert Webelhuth
The reason is obvious: mi bears the right features to bind the anaphor but is not close enough to bind it structurally; AH, on the other hand, is structurally in the right position to bind the anaphor [cf. (29)] but conflicts with it featurally. In sum, like the second German causative construction, the Chi-Mwi:ni causative construction discussed in the present section behaves biclausally in every respect. Interestingly, however, the predicate that projects these two clauses is expressed by one single morphological word.
2.4. Mixed Causatives The causative constructions whose data are introduced in this section are very interesting as well. Both constructions differ from the previous four in that they simultaneously evince monoclausal and biclausal properties. But like the previous two clausality types, this mixed type can be expressed analytically or synthetically.
2.4.1. ITALIAN9
The Italian causative, like Romance causatives in general, is very complex and consists of a number of different constructions with different properties. The reader should be aware that our goal here is not to provide anything like a comprehensive analysis of these constructions. Our interest is in showing that there exist some causative constructions which display a mixture of monoclausal and biclausal behaviors, i.e., that this construction type exists and has to be recognized as one type of complex predicate. The Italian causative under discussion behaves biclausally with respect to the binding theory but monoclausally in all other respects. Thus, note that the functional profile of the causative sentence in (31) is similar to that of a sentence headed by a ditransitive verb of the give type. The causer is expressed as a subject, the causee as an indirect object marked by the preposition a, and the lower patient is expressed as a direct object: (31) Fara riparare la macchina a Giovanni. he will make repair the car to Giovanni 'He will make Giovanni repair the car.' The absence of functional doubling is a monoclausal property. Moreover, the patient direct object becomes the subject when the predicate of
Causatives and Argument Structure
383
(31) is passivized: (32) La macchina sara fatta riparare a Giovanni. the car will be made repair to Giovanni. 'The car will be made to be repaired by Giovanni.' In this respect, the Italian construction patterns with the purely monoclausal causative construction of Malayalam [cf. (21)] and differently than the purely biclausal causative of Chi-Mwi:ni [cf. (27)-(28)]. Despite this evidence from relational uniqueness and passivization, however, there is evidence for the biclausality of the construction from binding theory. The argument is based on the anaphor se stess- (which alters its shape depending on the inflectional features of its antecedent). This element occurs in the following noncausative sentence that we use as a control. (33) is of the subject-to-object raising type and the inability of the subject of the main clause to bind the anaphor suggests that this anaphor must find an antecedent within the minimal clause containing it: (33)
* Maria i considerava [Giovanni orgoglioso di se stessai]. Maria considered Giovanni proud of herself 'Maria considered Giovanni proud of herself.'
Lexicalizing the lower patient position of our Italian causative with the inflectionally correct form of se stess- shows that the causative construction behaves like the biclausal raising-to-object construction (33) with respect to the binding of this anaphor: (34)
*Con le minacce fecero accusare se stessii a Giovanni. with threats(they i) made accuse themselves to Giovanni 'With threats they made Giovanni accuse themselves.'
(35)
Con le minacce fecero accusare se stesso i a Giovannii. with threats- (they) made accuse himself to Giovanni 'With threats they made Giovanni accuse himself.'
In both sentences, the subject causer is left unexpressed and signaled through the finite morphology on the causative auxiliary. (34) shows that this understood subject is unable to antecede se stessi, and (35) shows that, once the anaphor is put into the singular, it can be anteceded by the causer which is expressed as an indirect object on the surface. In sum, Italian has a causative construction that behaves monoclausally with respect to non-binding theoretic properties but biclausally with respect to at least one binding theoretic property.10
384
Gert Webelhuth
Finally, note that this Italian mixed causative is expressed analytically by the combination of the auxiliary fare and the bare infinitive of the main verb. 2.4.2. TURKISH 11 This brings us to the final empirical option we will illustrate here: a mixed causative construction whose predicate is expressed by a single synthetic word form. This case is found in Turkish, which has a causative construction that behaves very similarly to the Italian construction we just examined. (36) provides a simple noncausative transitive sentence of this language: (36) Hasan bauul-u acti. Hasan suitcase-Ace opened 'Hasan opened the suitcase.' The causativized version of (36) in (37) expresses the causee as an indirect object marked by dative case and the lower patient by a direct object marked by accusative case. The presence of three distinct grammatical functions in this causative is a monoclausal property: (37) Mehmet Hasani bavul-u ac-tir-di. Mehmet Hasan-DAT suitcase-ACC open-CAUSE-PAST 'Mehmet had Hasan open the suitcase.' In further support of a monoclausal analysis, the lower patient can become the surface subject when (37) is passivized, but the causee cannot: (38)
Bavul Mehmet tarafindan Hasan-a aq-tir-il-di. suitcase Mehmet by Hasan-DAT open-CAUSE-PASS-PAST 'The suitcase was caused by Mehmet to be opened by Hasan.'
(39)
* Hasan Mehmet tarafindan bavul-u aq-tir-il-di. Hasan Mehmet by suitcase-Ace open-CAUSE-PASS-PAST 'Hasan was caused by Mehmet to open the suitcase.'
As in Italian, some binding-theoretic facts spoil a purely monoclausal picture of the Turkish causative. For some speakers the first person anaphor kendim cannot be bound by an indirect object: (40)
*Hasan banai kendim-ii anlatti. Hasan me-DAT myself-ACC explained 'Hasan explained me to myself.'
385
Causatives and Argument Structure
Yet, apparently even those speakers who find (40) ungrammatical accept both (41) and (42): (41) Beni Hasan-a kendim-ii I Hasan-DAT myself-ACC 'I made Hasan wash me.' (42)
yika-t-ti-m. wash-CAUSE-PAST-lSG
Banai kendim-ii anlat-tir-di. me-DAT myself-ACC explain-CAUSE-PAST 'He made me explain myself.'
(41) is expected under a monoclausal analysis. If there is only one clause, then in the active the causer should be its subject and as a subject should be able to antecede the anaphor in the direct object position of the same clause. (42) presents a problem, however. Here the causee in the grammatical function of an indirect object binds the lower patient, even for speakers who do not allow indirect objects to bind that same anaphor in noncausative environments such as (40). One obvious way of explaining the difference between (40) and (42) is that in (40) the dative phrase is a pure indirect object whereas in (42) it is an indirect object semantically identified with a subject of a lower clause. If kendim can take any subject as its antecedent, then both the surface subject and the indirect object causee in the causative construction qualify as possible binders. For this explanation to succeed, the construction has to be biclausal, however, in contradiction to the arguments for monoclausality from surface grammatical functions and passivization presented earlier. The Turkish causative thus displays a mixed clausality profile, like the Italian causative discussed in the previous subsection. Unlike that construction, however, the Turkish causative expresses its predicate as one single morphological word.
3. SOME ANALYTIC PREREQUISITES We will begin our analysis of causatives by more specifically spelling out what properties a theory of causative predicates should have in order to best capture the empirical generalizations presented in the previous section. At the beginning of the article we already made one decision: the theory should not have to take recourse to destructive structural operations such as transformations. In essence this precludes all grammatical
386
Gert Webelhuth
function changing operations in the syntactic component through the application of tree modifying operations (i.e., "Move Alpha."). We might take this prohibition against tree-modifying function changes to be a symptom of a more general conceptual prohibition against all grammatical function changes in the syntax. As I will show next, this opens the door (i) to a very clean and conceptually attractive division of labor between the syntactic and the lexical components of natural language grammars and (ii) to an empirical approach to the six types of causatives discussed earlier that captures more generalizations than any other approach to the syntax of causatives that I am aware of.
3.1. A Strong Form of Lexicalism Preventing the syntax from effecting any kind of grammatical function changing operations means to identify the lexicon as the sole source of functional alternations. The correctness of this kind of grammar design is strongly suggested by the fact that predicates are typically associated with morphological categories in the lexicon and grammatical function changes are typically registered through morphological changes of predicates or through the addition of other morphological units (such as auxiliaries) to the base predicate. The previous section presented examples of both kinds: Malayalam adds an affix to a noncausative predicate to causativize it and German adds a causative auxiliary for the same purpose.12 We are led then to postulate two principles that give the lexicon clearly defined analytical privileges over the syntactic component. The first principle strongly ties the morphological component to the lexicon and separates the morphology from the syntactic component: (43) The Principle of Morphological Lexicalism No syntactic process creates morphological words, and no syntactic constraint mentions the features of the morphological bases of any morphologically derived syntactic expressions. (43) privileges the lexicon over the syntax in two ways. First, it makes it possible only for the lexicon but not for the syntactic component to form new morphological words. This restriction has been one of the pillars of grammatical frameworks such as LFG, GPSG, and HPSG. In GB, word formation in the syntax has traditionally been possible (see, for example, Baker, 1988), but with the advent of the Minimalist Program words have to enter the syntax fully formed in this theory as well, in accordance with the Strong Lexicalist Hypothesis.
Causatives and Argument Structure
387
The second consequence of the Principle of Morphological Lexicalism is that morphological derivations are opaque to the well-formedness constraints on syntactic representations. In other words, only the output of a morphological derivation is relevant for syntactic well-formedness. Without such a constraint, the syntax could "look" into a morphological expression at an arbitrary depth and impose constraints that are not found anywhere in the world's languages, e.g., "a sentence containing a denominal verb is grammatical if that verb at any level of morphological derivation is derived from an adjective whose most oblique argument is a subject at some other level of morphological derivation." Like Baker's theory of "incorporation," this approach makes far too much morphological information available to the syntactic component to be typologically adequate.13 The second leading principle of lexicalism I propose limits all predicate formation to the lexicon: (44) The Principle of Functional Lexicalism No syntactic schema admits a predicate whose argument structure is unspecified or underspecified. This principle, obviously inspired by early LFG,14 runs counter to the current trend of work on complex predicate formation in LFG and HPSG. Within LFG, Alsina (1993), Butt (1995), Nino (1995), and Nordlinger (1995) have modified the position of traditional LFG that all predicate formation is accomplished in the lexicon. As I understand them, these works now restrict the designated role of the lexicon with respect to lexicalism to the morphological side of lexicalism. As a result both the lexicon and the syntax can contain predicates with unspecified or underspecified argument structures and the syntax has the power to equip predicates with function inventories. At least in part in reaction to this extension of the power of the syntax at the cost of the lexicon, different assumptions about syntax ("Extended X-bar Theory") and morphology have become necessary within LFG. HPSG has been in open violation of Functional Lexicalism at least since the highly influential paper of Hinrichs and Nakazawa (1989) in which the authors presented an analysis of the German verb cluster in terms of a theory of argument attraction (also known as argument inheritance). In argument attraction, a head subcategorizes for another head (i.e., before that subordinate head has had a chance to combine with its subcategorized arguments) and by combining with that head in phrase structure "attracts" or "inherits" the arguments of the subordinate head. The diagram below provides a schematic example: note that the outer expression as well as the
388
Gert Webelhuth
final element on its SUBCAT list are lexical verbs. The SUBCAT list of the outer expression consists of the SUBCAT list of its final argument plus the final argument itself:
Given that the adicity of the subordinate head typically is not known in the lexicon, the adicity of the "attracting" head cannot be known in the lexicon either and is only determined in the syntactic component after lexical insertion (once [1] has been instantiated). This insertion of an expression with a lexically unspecified or underspecified argument structure is what the Principle of Functional Lexicalism prohibits. Hinrichs and Nakazawa's concept of argument attraction has been highly influential in the analysis of the verb clusters of various Germanic languages and has found application outside of that linguistic domain as well. See, for instance, Hinrichs and Nakazawa (1994), Heinz and Matiasek (1994), Kathol (1994, 1995), and many others. There can be no doubt that certain syntactic analyses of complex predicates have become easier to formulate once the restriction against predicate formation in the syntactic component was eliminated. Once the floodgates were opened, it did not take long for researchers to take advantage of the new analytical possibilities the more powerful syntactic component offered. Nevertheless, I believe that a careful weighing of both the conceptual and the empirical evidence favors a grammatical model where predicate formation is not distributed over the lexicon and the syntax but takes place exclusively in the lexicon. From a conceptual point of view, this leads to a very clean and attractive division of labor between the lexicon and the syntax: the lexicon forms predicates and the syntax saturates them in the process of forming phrases! In fact, I believe that this very simple statement can serve as the functional definition of the syntactic component: The syntactic component saturates predicates in phrase structure. This proposition functionally delineates the lexicon and the syntax along lines that are both intuitive and empirically successful: (i) from a functional point of view, the syntax only saturates but does not create new predicates—in this sense the syntax always reacts to the lexical resources inserted into syntactic schemas and is a projection of
Causatives and Argument Structure
389
the lexicon; (ii) from a formal point of view, the syntax creates phrases and is unable to create words. The lexicon, on the other hand, is a repository of basic words, basic predicates, and patterns for the formation of derived words and predicates. But restricting all predicate formation to one grammatical component has additional conceptual advantages with directly measurable results. In a theory with heavy reliance on typed feature structures such as HPSG, locating all predicate formation in a single grammatical component makes it possible to capture generalizations across different predicates by making them subtypes of common supertypes from which they inherit shared information. This yields grammatical analyses that are able to capture linguistically significant generalizations that are left unexpressed by theories that cannot integrate all predicates into the same inheritance hierarchy, because not all predicates are formed in the same grammatical component. More concretely, given (i) the strong ties between predicates and morphology (as discussed earlier), (ii) the existence of lexical gaps and irregularities in predicate formation, and (iii) the uncontroversial independent necessity of a lexical type hierarchy, the lexicon suggests itself as the most natural component of predicate formation. Once the challenge has been met to make the theory of the lexicon compatible with the traditional conception of predicates as being expressible either synthetically or analytically, a type theory with multiple inheritance allows for a simple and intuitive capturing of the similarities and differences between various predicate types, including the causative predicates from the six languages discussed in the previous section and analyzed in more detail below. In contrast, a "mixed" theory of predicate formation (i.e., one that forms synthetically expressed predicates in the lexicon but analytically expressed predicates in the syntax) is unable to subsume these different predicates under the same inheritance hierarchy and thus misses linguistically significant generalizations that the unified theory of predicate formation can capture. One problem the mixed theory of predicate formation in HPSG faces is that it is hard to see how the mechanism of argument attraction can be adequately constrained so as to make typologically valid predictions. Nothing stands in the way, as far as I can see, of the postulation of an argument attractor that attracts the arguments of two or more zero-level verbs it combines with and that appends the reverse of the argument list of its first argument to the argument list of its additional arguments; or an argument attractor that maps the attracted argument structure into some shuffled output of the result of appending the input triply to itself; or an argument attractor that attracts a passive argument structure and in the process of
390
Gert Webelhuth
attraction reconverts the argument configuration into an active argument structure, etc. Each of these hypothetical argument attractors is easy to formulate with the vocabulary that is currently available, but no language has expressions of this kind as far as I am aware.15 Thus, for any reasonable application of argument attraction it is easy to slightly modify its formulation to generate an argument attractor that is not exemplified in any language of the world (in this sense the argument attraction theory suffers from the same conceptual problem as the theory that allows syntactic well-formedness conditions to look arbitrarily deeply into morphological derivations). In the remainder of this paper I will sketch how theories like the one defended here and in greater detail (and also greater formal precision) in Ackerman and Webelhuth (1998) can circumvent this very serious theoretical problem.16 In essence, the solution consists of two decisions in grammar design: (i) The grammar adheres to the Principle of Functional Lexicalism in (44) and consequently relegates all predicate formation to the lexical component. The syntactic component is really to be thought of as the Saturation in Phrase Structure Component, as it cannot create arbitrary new predicates in the manner suggested in the previous paragraph, (ii) The grammar imposes strong well-formedness conditions on the type inventory and type structure of the lexicon as the single grammatical component where predicate formation can occur. It does so by providing a closed inventory of (classes of) highly valued notions that can be expressed by natural language predicates and a closed inventory of (classes of) highly valued argument (= functional) structures that can realize these predicates. Once these type constraints are formulated properly (which, of course, will require much more empirical work in the properties of predicates in different languages of the world), only empirically justified predicates can exit the lexicon and no unwanted predicate types can come into existence in the syntactic component either, since predicates cannot be formed there at all. 3.2. The Index Interpretation Principle We will postulate one final constraint that seems particularly natural before we apply our overall theory of predicates to causatives and try to make good on some of the promises we made earlier. In Pollard and Sag (1987), the binding theory applies to the SUBCAT lists of signs based on the relation of relative obliqueness. Following Borsley (1987), Pollard and Sag (1994:Chap. 9) split the SUBCAT list into different valence lists for subjects, specifiers, and complements. In order to still apply the binding theory in the standard fashion, they postulate a separate
Causatives and Argument Structure
391
level of argument structure which is constrained to be the concatenation of the three valence lists. This creates a systematic relationship between the valence of a lexical sign and its argument structure. However, in later work this systematic relationship has been broken and replaced by an unconstrained relationship between valence and argument structure. For instance, Manning and Sag (1995) present lexical representations for causative predicates that allow argument structures to be arbitrarily related to the valence. Thus, in Japanese the argument structure is the concatenation of the subject and complement valences: (45) Japanese (Manning and Sag, 1995:18)
On the other hand, in Chi-Mwi:ni, the argument structure registers the causer (subject) and the causee (direct object) but arbitrarily leaves out the lower patient, even though that patient remains a subcategorized complement of the verb (in schema (46) that argument is contained in the list L which, the reader will note, is part of the argument structure of the underlying stem and of the COMPS list of the derived causative; the argument structure of the causative itself does not inherit the arguments in L of the embedded stem, though): (46) Chi-Mwi:ni (Manning and Sag, 1995:28)
392
Gert Webelhuth
For Turkish, finally, the authors are forced to postulate yet a third pattern, namely, one in which the argument structure of the derived causative again misses some of the arguments of the underlying stem by arbitrary decision. This time, it is the causee argument of the underlying stem rather than the lower patient which is left out of the derived argument structure but is carried along as a subcategorized element into the higher valence: (47) Turkish (Manning and Sag, 1995:31)
The motivation for these analyses come from binding theory. Recall from the last section, for instance, that Chi-Mwi:ni has a morphological bi-clausal causative. One of the pieces of evidence for the biclausality of the causative came from the fact that the causee but not the causer can bind an anaphor in the lower patient position. The relevant examples are repeated below: (48)
(49)
An anaphor in the lower patient position can only be anteceded by the direct object causee, as shown in (48), but crucially cannot be anteceded by the subject causer, because that causer is outside of the minimal clause containing the lower patient, given that the Chi-Mwi:ni causative is biclausal (yet realized by a synthetically expressed predicate). Since their approach does not permit them to view the Chi-Mwi:ni causative as
Causatives and Argument Structure
393
biclausal, Manning and Sag attempt to capture the binding-theoretic facts of (48)-(49) by assigning these causatives an argument structure that arbitrarily leaves the lower patient of the embedded verb off the argument structure of the derived causative and by allowing the syntactic binding theory to "look into" words and applying binding constraints not only to the highest morphological level but at any morphological level, however deeply embedded. This will account for the difference between (48) and (49): since the lower patient has been left off the highest argument structure arbitrarily, it cannot be bound by the subject causer but must be bound at the level of the argument structure of the stem. At this level the only possible binder is the subject of the verb hit which is interpreted as the direct object causee in (48) and (49). For Japanese and Turkish, similar binding-theoretic facts force the authors into making the otherwise unmotivated argument structure assumptions in (46) and (47). Manning and Sag's (1995) approach to causatives loses two further linguistically important generalizations which it is important to be aware of. First, because argument structures in their system of causatives stand in an arbitrary relationship to valence, they not only lose generalizations about the syntax of causatives but are moreover forced into a very unnatural definition of passive in terms of not argument structure alone, as one might have expected, but a combination of argument structure and COMPS. The reason for this is that in languages that can passivize causatives, all of the subcategorized arguments of the causative are inherited by the passivized causative. In Manning and Sag's system, however, the character of the argument structure does not guarantee that all the complements of the causative appear on its argument structure. Hence: Use of this noncanonical causative necessitates modification of the passive sort we gave earlier, so that it also will not lose additional complements that are not on the argument structure of the stem. A suitable reformulation . . . is:
. . . While binding evidence seems to necessitate the kind of noncanonical lexical entries that we have proposed here, the result is clearly an undesirable complication of the passive lexical entry.
394
Gert Webelhuth
This problem of course will arise not only with passives but with all other processes that might change the argument structure of the derived causative further, e.g., further causativization, middle formation, morphological reflexivization or reciprocalization, dative shift, etc. Presumably, in the end, none of the grammatical function changing operations that have traditionally been thought of as argument structure operations really have a uniform formulation in terms of argument structures. The second problem I want to mention briefly is the one the present article is concerned with mainly, namely, the similarity between synthetically and analytically expressed causatives. None of the three similar papers that we have mentioned (lida et al., 1994; Manning and Sag, 1995; Manning et al., 1996) provides an explicit comparison of the properties of synthetically and analytically expressed causatives. I believe that this is no accident. Careful consideration will show that the combination of (i) the argument structure approach to causatives favored by all three papers and (ii) the recursive lexical types approach to morphologically derived causatives cannot be extended to analytically expressed causatives in a manner that does not mention the same crucial analytical properties once in the domain of the lexicon for synthetically expressed causatives and separately again in the syntax for analytically expressed predicates. Inspection of the diagrams in (45)-(47) shows why this is the case. In each of the three structures, the valence, the content, and the ARG-S of the morphologically derived causative are created by the type definitions from information of a stem contained in the same lexical feature structure. As soon as the signal of causativization is not affixal but a free auxiliary, the data types in (45)-(47) will be useless in the derivation of the valence, the content, and the ARG-S of the syntactically formed causative predicates and completely different type constraints have to be formulated, with an obvious lack of explanation of why the constraints applying in the morphological and the syntactic system have exactly the same functional and semantic prerogatives, as shown empirically in section 2.17 I believe that the central problem that lies at the center of all these theoretical difficulties is that argument structure has in essence been made into a second syntactic level in HPSG that serves no other purpose than that of providing a level of representation to which the binding theory applies.18 Clearly, if there are no constraints on the form of argument structures then there are also no constraints on binding that the theory as a whole imposes. I will therefore propose a principle on argument structures that mutually constrains them relative to other features of a lexical sign. According to that constraint it is impossible to arbitrarily leave out valence elements simply for the purposes of binding theory. That means that a different solution has to be found to the binding problem in
Causatives and Argument Structure
395
Chi-Mwi:ni than the one that Manning and Sag propose. This problem will be readdressed later in the context of the remaining data from Chi-Mwi:ni. What is the intuition that underlies the concept of argument structure (or functional structure, for that matter)? What I want to suggest is that argument structures are a mechanism for classifying each semantic index of a predicate as one of a small number of syntactic notions for such purposes as case assignment, agreement, default topic, pro drop, binding, function changes, etc. The primary functional motivation for this seems to be the desire to code the semantic and pragmatic similarity between the properties of different argument roles, e.g., the desire to code alike the arguments of different verbs with similar entailments (e.g., the agentive hitters, eaters, kickers, etc. in the subject function of active sentences), or the desire to code the preference for particular arguments (typically the subject) to serve the function of default topic in the active voice. Given that there are many different dimensions of information in which expressions can be located (referential vs nonreferential, definite vs indefinite, specific vs nonspecific, agent vs patient, old information vs new information, topic vs focus, to list a few), it is not surprising that notions such as subject and direct object do not typically fully correlate with any of these properties. Rather, each dimension seems to specify preferred relationships to the grammatical functions (and probably also directly to one or the other pole in the other dimensions) with the result that argument structures (or functional structures) are best understood as prototypes in the sense of Rosch (1973) or cluster concepts as described for subjects in Keenan (1976); i.e., subjects are typically referential rather than nonreferential, definite rather than indefinite, old information rather than new information, etc. Given this understanding of the concept of argument structure ( = functional structure) and the grammatical functions that appear in the structure, it seems compelling to require that all signs obey the following well-formedness principle: (50) The Index Interpretation Principle For every free (i.e., non-existentially bound) index I in the CONT (= content) of a lexical predicate there is an argument in the predicate's argument structure whose index is I. For every argument in the argument structure of a lexical predicate whose index is I there is a path P in the predicate's CONT such that the value of P is I, where I is free in CONT. As can be seen, (50) requires the argument structure of a sign to stand in a certain specific relationship to the content of the same sign: the first
396
Gert Webelhuth
clause requires every index (that does not fall under the proviso) that participates in defining the content of the expression to appear in the expression's argument structure and be assigned a grammatical function there. If argument structures serve the end of grouping the semantic arguments of a predicate into syntactic classes for the purposes of a finite coding system of prototypical grammatical functions, then it is natural to require that this classification be complete in the sense that every semantic argument be so classified in every predicate (unless the semantic argument is existentially quantified at the predicate level already and not intended to ever be expressed syntactically). The second clause of (50) requires the reverse, namely, that each argument of a predicate that is syntactically classified in the argument structure must be a semantic argument of the predicate. Given the rationale for the existence of argument structure as a classification device, this constraint is natural as well.19 I continue to assume with Pollard and Sag (1994) and Ackerman and Webelhuth (1998) that besides argument-functional structure there is a separate level of categorial valence that determines how many categories a sign has to combine with in phrase structure for a categorially complete phrase to be created. It is the relation between the valence level and the argument structure that decides whether a given predicate is "pro-drop," i.e., has an understood subject that is not categorially expressed but may have semantic content, trigger agreement on the verb, bind anaphors, etc. If the argument-functional structure specifies a subject argument but the valence specifies no corresponding category, then the predicate has an understood subject. Clearly, given the propensity of natural language grammars to leave semantic arguments of predicates implicit, we cannot require that all semantic indices or grammatical functions be expressed by valence elements; this rules out a constraint for valence equivalent to the first clause of (50). Whether an equivalent to the second clause of that principle can be imposed depends on one's view of whether non-referential elements must always be assigned a grammatical function and/or whether their indexes reliably appear in the semantic content of the predicate that subcategorizes for them. We will not pursue this issue here.
4. THE ANALYSIS OF CAUSATTVES
In section 2 we presented examples of causative constructions from six languages that showed that there are at least three different recurring causative constructions in the world's languages and that each of these
Causatives and Argument Structure
397
constructions can be headed by a predicate that is expressed either synthetically or analytically. The goal of this section is to sketch an analysis of this data that optimally captures the generalizations that hold across the different causative constructions while still achieving descriptive adequacy, i.e., the theory must also be able to express in a plausible and acceptable fashion what is language-particular and unique about each causative construction. Whatever else we want to think about causatives, given what we have already said, we are committed to two assumptions: (i) many languages have such a systematic alternation between noncausative and causative predicates that their grammars must contain some mechanism that accounts for this linguistic pattern; (ii) in accordance with Functional Lexicalism, this mechanism must apply in the lexicons of these languages in such a way that it maps noncausative input entries into causative output entries. In current lexicalist theories of grammar, there are two primary candidates for the productive mapping mechanism under discussion: either (potentially recursive) lexical rules (as in Bresnan, 1982, or Pollard and Sag, 1987) or (potentially recursive) lexical data types that provide lexical entries with some attribute where they can specify which other lexical entry they are derived from (as in Krieger and Nerbonne, 1993 and Riehemann, 1993). It is not easy to find relevant differences between these two approaches, and perhaps versions of each can even be found which are both adequate for natural language and mere notional variants of each other. Be that as it may, in the present work I will use the first path and employ lexical rules to productively relate non-causative predicates to the causatives based on them.20 In what follows, I will make the following informal assumptions about signs: as in HPSG, signs are complex data structures that contain information in different dimensions of linguistic information, e.g., phonological, syntactic, semantic, etc. To simplify the exposition so that I can concentrate on the main conceptual argument I am trying to make in this article, I will be relatively vague about the definitions of the feature structures and the data types. I feel justified in doing so because a fully precise and detailed version of the theory is accessible to the interested reader in the form of Ackerman and Webelhuth (1998).21 Thus, for the top-level structure of the sign, I will assume the presence of three attributes: (51)
CSTRUC categorial structure, i.e. information about part of speech and morphological features; ISTRUC information structure, i.e., information about content ( = CONT) and argument (functional) structure (= ASTRUC);22
398
Gert Webelhuth
VAL
valence, i.e., information about the subcategorized arguments of an expression and the auxiliaries a head has to combine with to express a particular predicate.
In order to achieve our analytical goals, we will now employ the following strategy. We need our system of representations to make available six representations, one for each of the six causative predicates responsible for the properties of the causative constructions discussed in section 2. We find a systematic cline in the generality of the properties of these constructions: some of their properties hold for all causatives, e.g., their common meaning; some other properties, mainly their monoclausal, biclausal, or mixed behavior, hold only for a subset of the constructions; and finally some properties such as the surface expression of the causative predicates are completely language and construction specific. To be able to capture the generalizations and idiosyncrasies at all levels, we will postulate a data type called active-causative that contains the information that must be inherited by all causative predicates. For each of the three clausality types that we found in our typological survey of causatives in section 2 we will then postulate a lexical schema (= rule) which maps noncausative predicates to causative predicates in the manner consistent with the monoclausal, biclausal, or mixed profile required by the construction. This gives us the following intermediate structure of lexical schemas: (52) The Lexical Rule System of Causatives (Universal)23
While the three schemas that form the leaves of the tree in (52) determine the clausality profile and the grammatical function inventory of the causative construction they govern, they crucially do not fix the surface realization of the causative predicate or that of its categorial arguments. As we saw in section 2, one and the same clausality behavior can be expressed either synthetically or analytically and this is left to languageparticular choices. We will therefore postulate a division between the lexical schemas that the universal type system makes available for causatives and language-particular lexical rules for causative constructions. In particular, under our assumptions individual natural languages will
Causatives and Argument Structure
399
make use of the information in universally defined lexical rules to define language-particular subtypes of causative patterns. (53) shows the extension of the lexical rule graph above once the productive language-particular causative schemas from the six languages treated in this paper are adjoined to the leaves of the universal tree in (52): (53) The Lexical Rule System of Causatives (Universal and Parochial)
Our first specific task will be to account for the properties that all causatives share no matter whatever else may distinguish them. I will assume that causatives uniformly express a cause relation that holds between a CAUSER and a caused state of affairs (= CAUSED-SOA) and moreover that in the active voice the CAUSER is always expressed as the subject of the predicate.24 We can efficiently capture these invariants by incorporating this information into the definition of the data type activecausative:
The final path in this feature structure says that the sign's argument structure (contained in its information structure) contains an element which is a subject whose index is the same as the index of the CAUSER in the cause state of affairs denoted by the sign (ASTRUC = argument structure; ELT = element of; su = subject). 4.1. The Analysis of Monoclausal Causatives With this feature structure at our disposal, we are in a position to present the three universal lexical schemas for causatives. We will begin
400
Gert Webelhuth
with the lexical schema for monoclausal causatives, the schema that both German and Malayalam use as a superschema to define their own productive causative schemas. Of course this schema takes on the form of a lexical rule in the current notation: (55) The Causative Construction, Type 1, (Lexical Schema)
Let us go through the features in this schema from top to bottom. The rule takes as input on the left a predicate and returns another predicate on the right, in particular an active-causative predicate. The right-hand side of the rule thus must contain all the information in the type activecausative and as we can see it does.25 The c-structural attribute is unspecified on both sides of this schema, as are the values of the path VAL:AUX. The reason for this is that the universal schema only expresses the meaning and the argument structural properties (in a wide sense, including certain subcategorization properties) of this monoclausal causative but leaves the surface spell-out of the predicate to individual languages that make use of this schema. The information structure (= ISTRUC) is where the most relevant information is stored. As can be read off from the representation, the semantic effect of causativizing a predicate is that the content of the input expres-
Causatives and Argument Structure
401
sion becomes identified with the caused state of affairs in the meaning of the causative predicate. The argument structure likewise is transformed. The input argument structure is partitioned into the subject argument and the remainder. Then the remainder is carried over in unchanged form into the argument structure of the newly formed causative predicate. The original subject's properties and relations are carried over as well, but they are associated with a different grammatical function in the output, namely, an oblique-indirect object. This is necessary, since the original subject is not the causer of the causative predicate and according to (54) the causer must be expressed as the predicate's subject. Since every predicate can have at most one subject, the original subject's properties and relations have to be functionally expressed differently after monoclausal causativization. Finally, note that the schema also carries over the subcategorization frame of the input predicate to the output but it adds one subcategorized phrase, namely, for the additional subject argument. This allows us to move on and to inspect the different ways in which German and Malayalam each express the information contained in the monoclausal causative construction. We begin with German. It defines a language-particular causative schema which is a subschema of (55): (56) The Causative Schema, Type 1 (German Instantiation)
All that needs to be contained in the German-specific schema are the German-specific ways of expressing the causative predicate on the surface and the categorial properties of the arguments the causative predicate selects.26 The schema requires that the following formal changes are made to the input predicate in the course of creating a predicate with the meaning and argument structure of a monoclausal causative: whatever lexeme the input CSTRUC mentions, the causative output must be realized by the bare infinitive of that lexeme; second, the causativized predicate must combine with a form of the auxiliary lassen in addition to the auxiliaries that the input must already contain. The schema also must fix
402
Gert Webelhuth
the categorial spell-out of the demoted causee: it is expressed by a prepositional phrase headed by the preposition von. In the informal notation used in this paper, the following is a reasonable representation for the transitive predicate waschen 'to wash':
(57)
When this predicate is entered into the German monoclausal causative pattern that results from fully spelling out all the features that this pattern inherits from the universal monoclausal causative pattern and the type active-causative, the following causative predicate is produced:
(58)
The predicate is realized by a combination of the bare infinitive of the main verb waschen and a form of the auxiliary lexeme lassen. This auxiliary-verb combination expresses the semantic content of a caused washing. As the indexes in the argument structure show, the CAUSER is expressed as a subject, the causee (i.e., the WASHER) in a von -headed
Causatives and Argument Structure
403
oblique PP, and the lower patient (the WASHED) as a direct object. Categorially, like all NP subjects of finite verbs in German, the CAUSER appears as a nominative NP and the direct object appears as an accusative NP. Assuming obvious syntactic schemas for the cancellation of the subcategorization and auxiliary requirements of predicates, this representation supports sentences of the kind we have been referring to as "German I" causatives.27 The absence of any doubled grammatical functions (e.g., two direct objects) in (13), which is a monoclausal property, is a function of the argument structure of a typical product of (56), e.g., (58): the ASTRUC contains a subject, a direct object, and an oblique and the same will be true for any causativized transitive input. Second, the monoclausal binding facts illustrated in (17)-(18) follow from the representation above as well. Recall Reis' Generalization (in our terminology) that the anaphoric pronoun sich must be bound by the subject of the minimal argument structure containing it. Inspection of (58) shows that if the lower patient is lexicalized with sich, then it will function as the direct object and will have to be bound by the subject of the argument structure which is interpreted as the CAUSER. This is exactly the state of affairs illustrated by (17) and (18) where the CAUSER but not the oblique CAUSEE is able to antecede the reflexive. This takes care of both of the diagnostics of monoclausality as well as the surface expression of the causative predicate in the construction "German I." Let us now see how the relative similarities and differences of the other monoclausal causative construction, namely, that of Malayalam, can be captured in our approach. Like German, Malayalam defines a causative lexical schema and declares it a subtype of the universal monoclausal causative schema. Also like in German, the Malayalam construction mentions only spell-out information in its language-particular schema:
(59) The Causative Schema, Type 1 (Malayalam Instantiation)
404
Gert Webelhuth
Observe three crucial differences in the spell-out of the Malayalam monoclausal causative compared to its German analog: whereas German uses a combination of the bare infinitive of the main verb and a form of the auxiliary lassen to express the causative, (i) Malayalam does not add an auxiliary when causativizing a main verb (note that the paths VALIAUX are identical in the input and output of the Malayalam schema), (ii) Malayalam has a distinguished morphological category for the causative which it requires the main verb to appear in (cf. the path CSTRUC:VFORM), and (iii) the demoted subject is expressed as an instrumental NP. Feeding a typical transitive verb such as pinch into the Malayalam monoclausal pattern would result in the following causative predicate meaning "cause to pinch":
(60) Illustration: The Type 1 Causative of the Malayalam Verb nulli
It is easy to see that even though this predicate differs from its German counterpart seen earlier in that it is expressed by a single morphological word, the two predicates will both behave monoclausally in every respect, since monoclausality is independent of the number of words that express a predicate. In particular, the existence of three distinct grammatical functions in the causative (20) follows from the presence of a subject, a direct object, and
Causatives and Argument Structure
405
an oblique in the argument structure of (60). This was taken to be a monoclausal property. The fact that as shown in (21) under passivization the lower patient rather than the causee becomes the surface subject of the passive follows from the fact that in (60) the lower patient is the direct object and, presumably, in Malayalam as in many other languages direct objects but not obliques can promote to subject status in the passive. The passivizability of the lower patient was taken to be another indication for monoclausal causativization. Finally, the binding facts in (23) follow from the same considerations that accounted for the binding-theoretic behavior of the German anaphor sich in the German monoclausal causative. In sum, our lexical theory of causative formation captures all the generalizations across the German and Malayalam causatives that need to be captured and yet allows the two languages to express their languageparticular idiosyncrasies. The functional-semantic similarities are accounted for by postulating a universal monoclausal causative pattern which both languages define subschemas of that serve to fix the way the functional-semantic information they share is expressed on the surface in a language-particular fashion. What is crucial to the success of this account is that lexical entries of predicates can specify an auxiliary valence that allows them to commit themselves to combining with certain auxiliaries in phrase structure. Since the lexicon thus contains information about the complete spell-out of a predicate, it is possible to incorporate into one type hierarchy predicates that spell out the same information with similar argument structures even though some are expressed by single morphological words and others by verb-auxiliary combinations. In this manner generalizations can be captured that alternative theories of complex predicate formation, in particular theories involving lexically underspecified argument structures and argument attraction in the syntax, simply miss.
4.2. The Analysis of Biclausal Causatives
It is not hard to see in what ways the universal schema for purely biclausal causatives has to differ from the monoclausal schema to account for the systematic differences we find between the two kinds of causatives. The
406
Gert Webelhuth
biclausal schema has to take the following form: (61) The Causative Construction, Type 2 (Lexical Schema)
Like the monoclausal schema, this one leaves the values of CSTRUC and VALIAUX unspecified, so that this information can be fixed at the language-particular level. Note that the two schemas also have the same semantic effects. They differ, however, in their argument structure and subcategorization effects as they have to in order to account for the clausality differences we have diagnosed. Recall that the monoclausal schema took the input argument structure, demoted the old subject to an oblique and added a new subject function for the CAUSER. It thus took as input one single argument structure and returned another single argument structure as output. The biclausal causative of course has to produce two argument structure units from the single argument structure unit it takes as input. The schema above shows how this is accomplished. On the left we see that the input argument structure is partitioned into the subject and the remainder. The reason for this is that we need to keep track of the subject argument of the input, since the output will in effect be a raising to object construction and will therefore have to contain an object that is structure-shared with the subject of the embedded argument structure clause. The schema does this by embedding the whole argument structure of the input as a complement clause (= CCL) into the argument structure of the output. Moreover, it adds two grammatical functions, a subject and a direct object, to the matrix argument structure clause. The subject expresses the CAUSER as is required
Causatives and Argument Structure
407
by the type active-causative and the direct object bears the same properties as the subject of the input (and hence also the subject of the complement clause of the causative output predicate). All in all, the causative predicate produced by (61) is biclausal, with the direct object of the higher clause being structure-shared with the subject of the embedded argument structure clause. As far as subcategorization is concerned, the effect of the biclausal schema is similar to that of the monoclausal one we saw earlier. The German-specific lexical schema for the biclausal causative construction need only determine the language-specific surface expression of the predicate and its arguments. Recall that, like the monoclausal causatives, the biclausal ones are expressed by the analytic combination of a bare infinitival main verb and a form of the auxiliary lassen: (62) The Causative Construction, Type 2 (German Instantiation)
When taking the noncausative transitive main verb waschen in (57) as input, this lexical schema produces the following biclausal output predicate: (63) Illustration: The Type 2 Causative of the German Verb waschen
408
Gert Webelhuth
(63) and the biclausal representations of other transitive verbs directly account for the empirical properties found in German biclausal causatives. In particular, the presence of two accusative NP direct objects in (24) expressing the causee and the lower patient of the underlying verb follows from the presence of two argument structure clauses with one direct object each in (63). The binding-theoretic fact illustrated by (25) that in the biclausal causative the anaphor sich in the lower patient position must be anteceded by the accusative CAUSEE and cannot be bound by the subject CAUSER, follows from the biclausal argument structure of (63) as well. As the indexes show, the lower patient in the biclausal causative bears the grammatical function of direct object of the embedded argument structure clause. Recall that sich must be bound by the subject of the minimal argument structure clause containing it. As can be seen, in (63) this subject is interpreted as the CAUSEE rather than the CAUSER. Chi-Mwi:ni has a biclausal causative construction as well but expresses its predicate as a single word rather than an analytic auxiliary-verb combination. The properties of this construction laid out in section 2 can be accounted for if Chi-Mwi:ni makes use of the following languageparticular subschema of the universal biclausal causative schema:
(64) The Causative Construction, Type 2 (Chi-Mwi:ni Instantiation)
Note that this Chi-Mwi:ni biclausal pattern differs from the "German II" biclausal pattern in exactly the same way that the Malayalam monoclausal pattern differs from "German I:" Chi-Mwi:ni uses no auxiliary in its language-particular expression but has a designated morphological category to signal that an input predicate has undergone the causative schema of this language.
Causatives and Argument Structure
409
A typical causativized transitive predicate in Chi-Mwi:ni would look as follows:
(65) Illustration: The Type 2 Causative of the Chi-Mwi:ni Verb dndik
A look back at the examples in (26)-(30) should convince the reader that this representation will account for the biclausal characteristics of the Chi-Mwi:ni causative construction. The reader will be aware that our theory not only is able to capture the generalizations that this synthetically expressed causative shares with the analytically expressed causative "German II," but it does so under the restrictive conceptual assumptions of Functional and Morphological Lexicalism. In particular, our theory captures the binding-theoretic facts of these causatives without compromising the Index Interpretation Principle: all indexes that appear in the CONT of the predicate [as well as the German predicate represented in (61)] are paired with an argument in the argument structure and vice versa, as should be the case for a well-constrained argument (functional) structure.
4.3. The Analysis of Mixed Causatives This brings us to what are arguably the most interesting cases of causative constructions, those that we have called "mixed" because they behave monoclausally in some respects but biclausally in others. At this point we must finally face the fact that the diagnostics for monoclausality
410
Gert Webelhuth
and biclausality cannot be interpreted as absolute, since otherwise the mixed cases could not exist. Rather, the diagnostics are no more than pretheoretical diagnostics and at least some of them on occasion yield a "false positive," i.e., declare a construction as monoclausal or biclausal even though in effect it falls into the opposite clausality class. Recall from our prior discussion in section 2 that the mixed cases behave biclausally with respect to (some) binding theoretic properties but behave as if only one argument structure clause is present in all other respects we examined. We will take that observation as our guide to the correct analysis of these constructions: their properties can all be captured if we assume that they are monoclausal but that the binding theory treats this one causative clause as containing two binding domains for anaphors, presumably because the causative semantically contains one caused state of affairs embedded within a second, i.e., the causing, state of affairs. By allowing the binding theory to sometimes define binding domains on argument structure alone but also sometimes to take the semantic structure into account that the argument structure classifies, we have an elegant and intuitive explanation of why it is that precisely single argument structure clauses that express two semantic "clauses" can contain two binding domains.28 For the Italian construction introduced in section 2 we will therefore postulate the following causative schema: (66) The Causative Schema, Type 1 (Italian Instantiation)
Observe that there is no need for the postulation of a third universal causative schema, since we are now claiming that the mixed cases are simply monoclausal as far as their argument structure is concerned. The above schema therefore is simply a subschema of the universal monoclausal schema (55).29 The Italian subschema fixes the surface spell-out of this causative as a combination of the bare infinitive of the underlying main verb as supported by a form of the auxiliary fare. (67) contains an example of a product of this schema based on the input of the transitive
Causatives and Argument Structure
411
verb riparare 'to repair': (67) Illustration: The Type 1 Causative of the Italian Verb riparare
This representation accounts for the presence of three different grammatical functions in causativized transitive verbs documented in (31). Since the lower patient rather than the causee bears the grammatical function of direct object, this lexical structure also directly supports the observation from (32) that under passivization the lower patient of the main verb becomes the subject of the passivized causative. These are all properties that fall out immediately from a monoclausal analysis of this causative at the level of argument (functional) structure. Recall, in contrast, the data in (34)-(35). They argue for the existence of two binding domains for the anaphor se stess- in this causative. Clearly, given the other assumptions we have made, we cannot claim that these binding domains correspond to two argument structure clauses. Rather, we will assume that binding theory is not only sensitive to argument structure units but may in addition refer to the semantic units that are associated with the argument structure. Since the semantics of the causative contains two states of affairs, it is natural to suppose that the two binding domains correspond to these two states of affairs rather than simply to units in the argument structure. The Italian anaphor se stess- therefore is subject to the following binding constraint: (68) Binding-theoretic requirements of the Italian anaphor se stesso: se stess- is bound by a term grammatical function within the minimal argument structure clause containing it; it cannot be bound by a logical subject other than the logical subject of the minimal state of affairs containing its index.
412
Gert Webelhuth
(68) thus claims that a combination of argument structure notions and semantic notions must be appealed to in order to account for se stess-. This anaphor must have a non-oblique antecedent30 within the minimal argument structure clause containing it. The second clause of the condition says that if there are two potential antecedents within that minimal argument structure clause for the anaphor, then it must take the argument that serves as the logical subject of the minimal semantic unit that contains the anaphor's index. The presence of two semantic states of affairs in the causative and the sensitivity of the binding condition of se stess- to semantic states of affairs thus create two binding domains within a single argument structure clause. Since the same effect could have been achieved by the presence of two argument structure clauses, it is easy to see why our binding theoretic diagnostic yields a "false positive" precisely under these circumstances. By giving a monoclausal analysis to this Italian causative, we have captured the monoclausal properties of the construction. (68) explains the data that this monoclausal analysis could not account for: (34) is ungrammatical and (35) grammatical because the (phonologically unrealized) causer in both sentences and the indirect object causee both occur within the minimal argument structure clause containing se stess-. Under these circumstances, the semantically closer logical subject wins out which in this case is the indirect object causee; the causer is semantically one state of affairs "too far away." The analysis of the Turkish data is almost identical to that of the Italian construction we just saw. The Turkish causative under discussion also makes use of the universal monoclausal causative schema but it differs from the Italian case in requiring the causative predicate to be expressed by a causative form of the underlying main verb. This morphological signal is functionally equivalent to the presence of the auxiliary fare in Italian:
(69)
The Causative Schema, Type 1 (Turkish Instantiation)
Causatives and Argument Structure
413
An example output of the Turkish causative is found in (70). It is derived from the underlying verb "to open": (70) Illustration: The Type 1 Causative of the Turkish Verb "to open"
The internal structure of this kind of representation accounts for the presence of three different grammatical functions in causativized transitives such as (37) and also explains the passivization facts in (38)-(39) in analogy to similar Italian facts. This takes care of the monoclausal properties of this construction and leaves the binding facts in (41)-(42) to be accounted for. It is clear that the binding behavior of the Turkish anaphor kendim is similar to that of the Italian anaphor se stess- in that it must refer both to the argument structure and the semantic structure of the predicate. There is one crucial difference, however. Whereas a closer logical subject in the Italian case prevents a more distant logical subject from anteceding se stess-, the Turkish sentence is simply ambiguous, as long as both potential antecedents are functionally close enough to the anaphor. This follows from the definition in (71): (71) Binding-theoretic requirements of the Turkish anaphor kendim: kendim is bound by the subject of the minimal argument structure containing it or the logical subject of the minimal state of affairs containing its index, provided that logical subject bears a term grammatical function. This concludes our discussion of the "mixed causatives."
414
Gert Webelhuth
5. SUMMARY
This paper has presented the data from six causative constructions in the world's languages. It is not difficult to analyze each of these causatives by itself without reference to the others. However, it was shown that there are linguistically significant generalizations that hold across them that a linguistic theory can and hence should capture. On the other hand, there are also generalizations that hold only across a subset of the causatives and each construction bears properties that are idiosyncratic to it as well. The analytical challenge to the linguist consists in whether he or she can develop a linguistic formalism that is able to express all the generalizations at the required levels of generality from universal, to construction specific, to idiosyncratic. It would be disappointing if the best we can do is to assign a set of properties to each individual causative construction without any reflection or recognition that some properties of causatives are found in causative construction after causative construction while others are completely language-particular or even confined to one single causative construction. The approach to causatives (and predicates in general) in this paper is meant to take a first step in the desired direction. It makes an analytical distinction between universal and language-particular patterns and views each causative construction as typically being composed of a combination of both. In particular, the universal schemas contribute the meaning and the argument structure properties of causatives and leave it to the language-particular and in some cases construction-particular causative schemas to fill in the surface realization of the particular causative. In order to reach the goal of capturing the semantic and argument structure generalizations across the different causatives, however, it is necessary to abandon the wide-spread assumption that the lexicon is subject to the constraint that it cannot contain entries for analytically expressed predicates. From a primarily information-driven perspective, this is a curious constraint in any event in that it makes morphology more important than semantics or argument structure in defining the function and the limits of the lexicon. It has the consequence that two predicates which merely differ in syntheticity but otherwise have exactly the same meaning or argument structure must nevertheless be generated in different grammatical modules. Not only does it strike me as curious that the two German past tense forms lachte "laughed" and gelacht hat "laughed" must be generated in different grammatical modules even though they can be used with exactly the same meaning, in addition this goes against the intuition of traditional grammarians that both of these expressions are members of the paradigm of the predicate lachen. In any event, the
Causatives and Argument Structure
415
modern generative literature has not only moved away from this traditional insight but seems to maintain this form-based distinction more as a matter of inherited ritual than because it is superior conceptually or empirically to the traditional information-driven alternative. The present paper has returned to the traditional insight and treats synthetic and analytical forms of the same predicate in the same grammatical component. If this is desirable, then the strong affinity between the lexicon and morphology and the wide-spread existence of idiosyncrasies in predicate formation argues for the lexicon as the single locus of all predicate formation. Consequently, in order to handle both synthetically and analytically expressed predicates, I follow Ackerman and Webelhuth (1998) in postulating a valence attribute AUX that allows predicates to consist of auxiliary-main verb combinations. The combination of (i) a distinction between universally available and language-particular predicate schemas and (ii) a formalism that can describe synthetically and analytically expressed predicates in the same grammatical component and in the same inheritance hierarchy makes it possible to capture the generalizations concerning causatives at all the levels of generality shown to exist in the present article. The properties that all causatives share are expressed in the type active-causative that all causative predicates belong to. The monoclausal and biclausal causatives inherit their specific properties from universal causative schemas and the properties idiosyncratic to each causative construction are captured at the level of language-particular causative schemas that inherit from the universal causative type and one of the two universal schemas. It remains to be seen whether other assumptions are able to reach this amount of explanation in the area of causatives. At this time, it seems to me that mixed theories of predicate formation, i.e., those that allow predicates to be formed both in the lexicon but also in the syntax through argument merger, are principally incapable of reaching the same level of linguistic explanation as the uniform and information-driven theory of predicate formation in the lexicon that was proposed in the present paper and is fleshed out in greater detail in Ackerman and Webelhuth (1998).
ACKNOWLEDGMENTS Versions of this paper were presented at the Germanic Linguistics Annual Conference in Ann Arbor, the HPSG Workshop in Tubingen, the HPSG Workshop in Pittsburgh, the Ohio State University, the University of South Carolina, the University of Southern California, Stanford University, and the 1996 Holland
416
Gert Webelhuth
Institute of Linguistics Summer School. I thank the audiences at these presentations for their helpful input and friendly suggestions. In particular, I acknowledge the help of Mirjam Fried, Bob Kasper, Chris Manning, Carl Pollard, and Ivan Sag. Farrell Ackerman, of course, deserves special thanks for his great influence on my thinking over the last few years. That influence can be felt on every page of this paper. Two anonymous reviewers read an earlier draft of the paper very carefully and gave me very helpful and constructive advice, so much in fact that I had to leave some of it for future research. All remaining errors are of course my own.
NOTES 1 Ich widme diese Arbeit meinen Geschwistern Ellen Gabriel und Wolfgang Webelhuth fur deren liebe Freundschaft und Unterstiitzung wahrend meiner intellektuellen Wegfindung. An anonymous reviewer asks about the necessity of pretheoretical diagnostics, given that "these properties are defined on ARG-S in HPSG (or perhaps VALENCE lists)." As the reviewer's wording suggests, the nature of ARG-S and VALENCE and their relative delimitation are still under healthy discussion in HPSG. As with the nature of syntheticity vs analyticity I felt that it was worth trying to approach causatives with a relatively empty slate of theoretical commitments in order to see which theoretical commitments are necessary for an account that best captures their apparent linguistically significant generalizations. I may of course be mistaken in what generalizations are linguistically significant. 3 See Kiparsky (1987), Bresnan and Kanerva (1989), and the use of obliqueness hierarchies in Pollard and Sag (1987, 1994). 4 Reis notes some exceptions to the domain restriction (but not the antecedent restriction) when sich is embedded within a prepositional phrase or occurs in certain word orders in subject-to-object raising constructions. Both of these circumstances will be carefully avoided here; all our examples are therefore expected to fall under Reis' Generalization. A reviewer asks me to address "an obvious alternative analysis of German I . . . [that] involves morphologically unrealized passivization of the downstairs verb, with the patient becoming the subject of the downstairs verb and the object of the upstairs verb 'cause.'" One problem with that analysis is that not all passivizable verbs can appear in German I. The verb sehen 'to see,' for instance, can be passivized but cannot be causativized with German I: *Sie liefl das Buck von dem Jungen sehen 'She made the boy see the book.' The binding-theory also argues strongly against a biclausal construal of German I: in Siet lie ft en die Bucher aneinander i schicken. 'They had the books sent to each other.' the surface subject can bind into the prepositional phrase argument of the lower verb. In the biclausal analysis the antecedent and the reciprocal would have to be in two different clauses but this reciprocal can never be bound across clauses elsewhere, as far as I am aware.
Causatives and Argument Structure
417
6 The Malayalam data is cited from Marantz (1984, pages 276, 279, 282), where it is attributed to K. P. Mohanan (1982a, b, 1983). 7 The Chi-Mwi:ni data is cited from Marantz (1984:267, 270, 271), where it is attributed to Abasheikh (1979). 8 Following Marantz (1984), these two identically marked NPSbear two identical grammatical functions, implicating two separate clauses. However, a reviewer makes the point that the constrast between (27) and (28) might be made to follow from a monoclausal analysis where the second object of ditransitive verbs cannot be passivized. As the reviewer further suggests, this issue should be decidable through inspection of the behavior under passivization of the second object of a lexically ditransitive verb like give. 9 The Italian data is cited from Burzio (1986:225, 254, 264) and Rosen (1990:107). 10 Since it is well known that in languages with more than one anaphor these do not all need to have the same binding domains (cf. Dalrymple, (1993)), we cannot exclude that other anaphors in Italian will behave differently. This is immaterial to our major theoretical point, however, that this Italian causative in at least some respects behaves monoclausally and in at least some other respects behaves biclausally. n The Turkish data is cited from Aissen (1974:10, 11, 14, 15, 131, 132). 12 Note also the proposal in Baker (1988) that grammatical function changes are frequently the consequence of incorporation, a traditionally morphological concept. 13 See lida et al. (1994) for an approach to Japanese causatives that allows the syntactic binding theory to look into morphologically formed words at arbitrary depths. Manning et al. (1996) both provide a different interpretation of the binding data from Japanese causatives than the earlier paper and eliminate all syntactic references to the properties of the stems of morphologically derived items. It is clear that their original proposal would have opened the door to the formulation of an indefinite number of typologically unmotivated syntactic constraints on morphological derivations. The Principle of Morphological Lexicalism nails that door shut again. 14 LFG's Principle of Direct Syntactic Encoding outlawed changes in grammatical functions. (44) takes this one step further and additionally forbids that the initial function assignment is accomplished in the syntactic component. 15 Also conspicuously absent from the predicate formation toolbox are all kinds of alphanumeric orderings, even though those are omnipresent in cognition in general. Why are there no argument attractors that order the arguments they attract in alphabetic order using part of speech as a key? And no predicate formation processes have been reported in any language, as far as I am aware, that would order arguments by reference to the properties even and odd number. (Phonological processes involving stress do seem to be sensitive to these properties, though, so one cannot argue that they are never used in grammars—they are simply not used in predicate formation!) In my view, the systematic absence of these otherwise cognitively natural notions in predicate formation casts strong suspicion on the view that predicate formation is the product of such algebraically generalized relations as argument attraction in
418
Gert Webelhuth
combination with other freely available cognitive notions, including properties such as even and odd that are demonstrably present elsewhere in the linguistic system. Even if—contrary to my expectation—such a process as argument attraction will be retained in the linguist's toolkit, I am convinced that its contribution will eventually be seen as insignificant in comparison to the importance of the restrictions (no reverse, alphabetize, triple append, even, odd, etc.) that have to be imposed on it to exclude the massive amount of structures it would otherwise wrongly allow for in the world's languages. 16 Note that rather than being a new problem, this is the classical problem of post-Aspects syntactic theorizing in Generative Grammar. (See, for instance, my comments in connection with Government and Binding Theory in Webelhuth, 1995:9, "Introduction.") Exactly the same kinds of problems have arisen over the years with respect to the proliferation of Fillmore's Case roles, Aspects style phrase structure rules, transformational rules, functional heads like AGR-O, TNS, NEG, etc. One of the reviewers proposes that commonalities between analytic and synthetic causatives might be captured by virtue of the common result of their formation and asks why I apparently accept "only arguments that capture commonalities in analytic and synthetic causatives by positing commonalities in the process by which they are created." The reason I am biased toward the latter view is simply that I know that it can capture the linguistically significant generalizations that must be dealt with [see section 4 below and chapter 9 of Ackerman and Webelhuth, 1998]. A similar demonstration has not been given for the alternative solution and I am not convinced that one can be given. If further research shows otherwise, then I would be very interested in weighing the relative advantages and disadvantages of the two approaches and might well end up favoring the solution the reviewer proposes. 18 In Manning and Sag's (1995:4) words, the architecture which allows arbitrary dissociations between valence and argument structure "takes HPSG a certain distance from the monolevel, monostratal roots of GPSG and HPSG2." 19 At least as far as the literal use of predicates is concerned. Perhaps idiomatic predicates should be defined as those which violate the second half of (50); i.e., they have grammatical arguments that do not correspond to any of their semantic arguments. Likewise, for expletives if they do not contribute any meaning to CONT. As stated, raising verbs as well violate the first half of (50), since the verb embedded under them may contribute indexes to the raising verb's CONT for which there is no corresponding argument in the raising verb's argument structure. To cover this case, predicates must be allowed to delegate the argument structure "identification" of some of the semantic indexes they govern to subordinate predicates. I will not attempt a proper formulation here that takes this into account. (50) allows one to characterize the class of the syntactically most productive verbs as those that satisfy both parts of (50). These are the verbs that can typically appear in most syntactic constructions such as actives, passives, questions, relative clauses, etc. Interestingly, by making reference to the parts of (50) which they
Causatives and Argument Structure
419
violate, one can also characterize the standard classes of verbs that have a much less regular syntactic distribution. For instance, raising verbs, verbs governing expletive subjects, and idioms frequently cannot be passivized and are syntactically exceptional in other respects as well. I believe that the conjunction of the two parts of (50) is a central part of the definition of "unmarked predicate" in Universal Grammar. But UG also allows for one of the conjuncts to be slightly weaker; as far as I can tell, though, it seems to require every predicate to satisfy at least one conjunct fully. Should that claim hold up, this would be a significant predictive constraint on natural language predicates that the lexical type hierarchy of HPSG can express (i.e., as a relational constraint). 20 To cover all bases, however, Ackerman and Webelhuth (1998) take the second path. I am somewhat more sympathetic to the data type analysis as compared to the lexical rule approach, since it seems to be easier to integrate with a sign-based approach based on typed feature structures and multiple inheritance in a type hierarchy. But this may be no more than a personal preference. 21 That theory has been implemented in the PAGE grammar development environment developed by the Deutsche Forschungsgesellschaft fur Kimstliche Intelligenz. I am deeply indebted to the Center for the Study of Language and Information at Stanford University, especially Ivan Sag and Dan Flickinger, for allowing me to use their facilities, and to Dan Flickinger, Rob Malouf, Peter Skadhauge, and Stephan Oepen for their generous and much-needed help with the PAGE system. 22I use the terms "argument structure" and "functional structure" interchangeably. This structure simply associates each argument of a predicate with a grammatical function from traditional grammar, e.g., "subject" or "object." I believe that cross-linguistic generalizations across such recurrent relations as passives, middles, causatives, etc. cannot be captured without grammatical functions, since languages are too diverse categorially for grammatical categories to serve this purpose (for instance, some languages express direct objects as prepositional phrases rather than as noun phrases). 23 Actually, we will soon integrate the mixed causatives with the monoclausal causatives. (52) and (53) are thus only preliminary. 24 T. Mohanan (1990), Alsina (1993), and Butt (1995) assume a three-place analysis of causative predicates instead. Alsina (1993) claims that the semantic three-place analysis in combination with an argument merger parameter is capable of deriving the grammatical function inventory of different causative predicates. His arguments leave me unconvinced, especially because almost all the logically possible function inventories exist for some causative or other even though Alsina's theory predicts the existence of at most two patterns (for a list of languages with different functional patterns in causatives, see Ackerman and Webelhuth, 1998: Chap. 9). Moreover, as Alsina points out the functional inventory of perception verbs in many languages is similar to that of causatives, even though causatives involve a causer affecting a patient and perception verbs under reasonable assumptions do not (as Alsina himself concludes). This strongly suggests that the function
420
Gert Webelhuth
inventories of predicates cannot be predicted from their semantic properties or their argument structure alone. In the same vein, many of the semantic effects that Alsina claims can only be derived from a three-place analysis of causatives arise with uncontroversially two-place predicates as well. For instance, the contrast between "The president made the generals clean the toilets" and "The president made the toilets be cleaned by the generals" with the underlined expression being perceived as more affected by the causer in many languages that have this alternation, reappears with the English raising-to-object verb believe: "The president believed the generals to have cleaned the toilets" vs "The president believed the toilets to have been cleaned by the generals." Every English informant I have presented this pair of sentences to agrees that in each case the president's belief is more about the underlined expression than the second object. Under Alsina's analysis this is only explicable if the underlined expression in both of these sentences is a semantic argument of the verb believe. This, however, is inconsistent with the empirical fact that this verb takes nonreferential objects, e.g., the existential there ("Kim believed there to have been a riot") or parts of idioms ("Kim believed advantage to have been taken of Sandy"). I therefore see no strong reason to adopt the three-place analysis of causatives and I would dispute the claim that the function inventories of causatives have been shown to follow from their meanings (or argument structures, in Alsina's sense which makes it different from functional structures—I am using the terms "argument structure" and "functional structure" interchangeably). 25 The notation [1:2] for members of the argument structure is interpreted to mean that the grammatical function GF bears the semantic index GO and is realized by the subcategorized element bearing the index [1]. 26 It also needs to be stated whether the demoted agent is realized as an oblique or an indirect object. I suppress this information throughout to simplify. 27 Note that the syntactic schemas must be formulated to ensure that auxiliaries are the categorial heads of their predicates even though they are not the functional-semantic heads. For instance, the finiteness of a phrase correlates with the finiteness of the auxiliary. This is easy to accomplish; see Ackerman and Webelhuth (1998). 28 The idea that the binding theory can refer to the semantics (perhaps in the form of semantically construed argument structures) has previously been employed in Jackendoff (1972), Kiparsky (1987), T. Mohanan (1990), and Alsina (1993), among others. 29 The nodes labeled "Monoclausal Causative" and "Mixed Causative" should therefore be collapsed in (52) and (53). 30 Terms are nonoblique grammatical functions, i.e., subjects, direct and indirect objects. This condition is present to rule out a demoted agent in a passive by-phrase as a possible binder of se stess-. In other languages, this grammatical function can bind anaphors (see, for instance, Marathi as documented in Dalrymple, (1993:13), adding to the evidence that logical subjecthood needs to be referred to in binding theory on occasion and that relative obliqueness of surface grammatical functions does not suffice.
Causatives and Argument Structure
421
REFERENCES Abasheikh, M. I. (1979). The Grammar of Chimwi:ni Causatives. Doctoral dissertation, University of Illinois, Urbana. (Distributed by University Microfilms.) Ackerman, F. and G. Webelhuth (1998). A Theory of Predicates. CSLI Publications, Stanford: CA. Aissen, J. (1974). The Syntax of Causative Constructions. Ph.D. dissertation, Harvard University. [Published (1979), Garland, New York.] Alsina, A. (1993). Predicate Composition: A Theory of Syntactic Function Alternations. Ph.D. thesis, Stanford University. Anderson, S. (1992). A-Morphous Morphology. Cambridge University Press, Cambridge, UK. Baker, M. (1988). Incorporation: A Theory of Grammatical Function Changing. University of Chicago Press, Chicago. Bloch, B. (1947). English verb inflection. Language 23, 399-418. Borsley, R. (1987). Subjects and Complements in HPSG. Technical Report CSLI107-87. Center for the Study of Language and Information. Stanford, CA. Bresnan, J. (1982). The passive in lexical theory. In J. Bresnan (ed.), The Mental Representation of Grammatical Relations. MIT Press, Cambridge, MA. Bresnan, J. and J. M. Kanerva (1989). Locative inversion in Chichewa: A case study of factorization in grammar. Linguistic Inquiry 20, 1-50. Burzio, L. (1986). Italian Syntax: A Government-Binding Approach. Reidel, Dordrecht. Butt, M. (1995). The Structure of Complex Predicates in Urdu. CSLI Publications, Stanford, CA. Copestake, A. (1992). The Representation of Lexical Semantic Information. Cognitive Science Research Papers 280, University of Sussex. Dalrymple, M. (1993). The Syntax of Anaphoric Binding. CSLI Publications, Stanford, CA. Heinz, W. and J. Matiasek (1994). Argument structure and case assignment in German. In J. Nerbonne et al. (eds.), German in Head-Driven Phrase Structure Grammar, 199-236. CSLI Publications, Stanford, CA. Hinrichs, E., and T. Nakazawa (1989). Flipped Out: AUX in German. In Papers from the 25th Meeting, 193-102. Chicago Linguistic Society. Hinrichs, E. and T. Nakazawa (1994). Linearizing AUXs in German verbal complexes. In J. Nerbonne et al. (eds.), German in Head-Driven Phrase Structure Grammar, 11-37. CSLI Publications, Stanford, CA. lida, M., C. Manning, P. O'Neill, and I. Sag (1994). The Lexical Integrity of Japanese Causatives. Presented at the LSA 1994 Annual Meeting. Jackendoff, R. (1972). Semantic Interpretation in Generative Grammar. MIT Press, Cambridge, MA. Kathol, A. (1994). Passives without lexical rules. In J. Nerbonne et al. (eds.), German in Head-Driven Phrase Structure Grammar. 237-272. CSLI Publications, Stanford, CA. Kathol, A. (1995). Linearization-Based German Syntax. Ph.D. dissertation, Ohio State University.
422
Gert Webelhuth
Keenan, E. (1976). Toward a universal definition of 'subject.' In C. Li, (ed.), Subject and Topic. Academic Press, New York. Kiparsky, P. (1987). Morphology and Grammatical Relations. Unpublished manuscript, Stanford University. Krieger, H.-U. and J. Nerbonne (1993). Feature-based inheritance networks for computational lexicons. In T. Briscoe et al. (eds.), Default Inheritance within Unification-Based Approaches to the Lexicon. Cambridge University Press, Cambridge, UK. Manning, C. and I. Sag (1995). Dissociations between Argument Structure and Grammatical Relations. Unpublished manuscript, July 1995, Carnegie Mellon University and Stanford University. Manning, C., I. Sag, and M. lida (1996). The lexical integrity of Japanese causatives. In G. Green and R. Levine (eds.), Readings in HPSG. Cambridge University Press, Cambridge, UK. (To appear.) Marantz, A. (1984). On the Nature of Grammatical Relations. MIT Press, Cambridge, MA. Mohanan, K. P. (1982a). Grammatical relations and anaphora in Malayalam. In A. Marantz and T. Stowell (eds.), MIT Working Papers in Linguistics, Vol. 4, 163-190. MIT Press, Cambridge, MA. Mohanan, K. P. (1982b). Grammatical relations and clause structure in Malayalam. In J. Bresnan (ed.), The Mental Representation of Grammatical Relations. MIT Press, Cambridge, MA. Mohanan, K. P. (1983). Move NP or lexical rules? Evidence from Malayalam causativization. In L. Levin et al. (eds.), Papers in Lexical-Functional Grammar. Indiana University Linguistics Club, Bloomington. Mohanan, T. (1990). Argument Structure in Hindi. Ph.D. thesis, Stanford University. (Published by the Center for the Study of Language and Information, Stanford, CA.) Nino, M.-E. (1995). Split Inflection and the Architecture of Grammar. The Case of Finnish Verbal Morphology. Unpublished manuscript, Stanford University. Nordlinger, R. (1995). Split Tense and Imperative Mood Inflection in Wambaya. Unpublished manuscript, Stanford University. Pollard, C. and I. Sag (1987). Information-Based Syntax and Semantics, Vol. 1. Fundamentals. CSLI Lecture Notes No. 13. CSLI Publications, Stanford, CA. Pollard, C. and I. Sag (1994). Head-Driven Phrase Structure Grammar. CSLI Publications, Stanford, University of Chicago Press, Chicago. Reis, M. (1976). Reflexivierung in Deutschen A.c.I.-Konstruktionen. Ein transformationsgrammatisches Dilemma. Papiere zur Linguistik 9, 5-82. Riehemann, S. (1993). Word Formation in Lexical Type Hierarchies: A Case Study of Bar-Adjectives in German. M.A. thesis, University of Tubingen. Rosch, E. (1973). Natural categories. Cognitive Psychology 4, 328-350. Rosen, S. (1990). Argument Structure and Complex Predicates. Ph.D. dissertation, Brandeis University. Webelhuth, G. (ed.) (1995). Government and Binding Theory and the Minimalist Program: Principles and Parameters in Syntactic Theory. Blackwell, Oxford.
INDEX
Page references followed by n indicate notes.
A
Abasheikh, M. I., 417n, 421 Abbott, B., 361n, 366 Abeille, A., xii, xiii, xvii, 5, 14, 17, 19, 25, 34n, 35n, 36n, 37n, 39, 310n, 311, 319, 320, 323, 341, 361n, 366 Abraham, W., 266n, 267 Abstraction, functional, 274 Accessibility relation, 272, 274, 277-281 Accusatives /za-causative construction, 187, 188, 193-194 with infinitive, 267n Acl verbs, 144 Ackerman, F., 256, 257, 258, 259, 267«, 270, 370, 371, 390, 396, 397, 415, 418n, 419n, 420n, 421 ACLR, see Argument Composition Lexical Rule a-composition, 4, 23 Active-causative types, 398, 415 Causative Construction, Type 1, (Lexical Schema), 400 definition of, 399 Actives, 418n Adjuncts, 361n
in Korean, 217n LP constraints on, 204 multiple, 217n true, 204 Adjunct scrambling, 205 in German, 197 in Korean, 196-205 Adverb phrases, 204, 205 Adversity passive, 213n ADVP, see Adverb phrase Affected agents, 85, 87, 88 Affectedness, xiv, 89, 94-102, 104-105, 108n for Urdu causatives, 96-97 Affixes, 370 inflectional, 340, 352 invisible, 256 pronominal, 6, 7, 34, 303-304 pseudo-affixes, 262n Aff(ixal) synsems, 7, 303 Afterthought expression, 162, 180 Agreement, 9 in French, 25 long-distance, 108n participle, 25 subject-verb, 252-253 in Urdu, 80
423
424 Aissen, J., 3, 5, 19, 39, 39n, 73, 75, 76, 109, 313, 366, 417n, 421 Aktionsart, 95 Albanian, te element, 364n Alphanumeric orderings, 417n Alsina, A., 24, 34n, 35n, 37n, 39, 74, 75, 77, 78, 79, 80, 84, 85, 87, 88, 90, 91, 92, 94, 95, 98, 101, 103, 105, 107n, 108n, 109, 387, 419n, 420n, 421 Alternation, case, 257 auxiliaries in, 257-258 of dative argument, 256 with (de)emotion auxiliaries, 165, 166, 174-176, 211 in Korean, 175-177 Ambiguity, spurious, 48 A'-movement, xiv A-movement, 211 am preposition (German), 232-233 Analytic predicates, 371, 375, 397, 414, 415 in German, 379-380, 407, 409 in Italian, 383 similarity with synthetic predicates, 394 in syntax, 389 Anaphors, 375 binding domains, 410 binding-theoretic requirements, 411-412, 413 in Chi-Mwi:ni, 381-382, 392-393 in Italian, 417n kendim (Turkish), 384-385, 413 in Malayalam, 378 in Marathi, 420n se stesso (Italian), 383, 411-412, 420n sich (German), 377, 380, 381, 405, 408 Anderson, S., 421 dndik verb (Chi-Mwi:ni), 409 Application, functional, 274 Applicatives, 75, 93, 94 Architecture, multimodal, 276 ARG-S, see Argument structure ARG-ST, see Argument structure Argument attraction, 388 Argument composition, xii-xvi, 79, 93, 217n a-composition, 4, 23 applicability range of, 212-213 and auxiliary verb construction, 160-177 and branching structure, 223-225 c-composition, 4, 23 clitic climbing as, 305-307
Index
concept of, 222 in Dutch, 320 extension of, 257 in French, 3, 320 in German, 121, 123, 126, 129, 144, 146, 167-168 in German third construction, 131-135, 150-151 in German verb clusters, 260, 320 in Italian, 313, 320, 321-323, 347 in Korean, 159-220 lexical rules, 200-202 in long-distance scrambling, 177-205 in Polish, 320 in restructuring verbs, 321-323 in Romance, 93 in third construction, 131-135, 150-151 in verb clusters, 260, 320 Argument Composition Lexical Rule, 205, 321, 325-327, 331, 357, 359 Complement Cliticization Lexical Rule interactions, 325 in Italian tough constructions, 341 Middle si Lexical Rule interactions, 335, 336-340 Argument fusion, 74, 78-79, 90-94, 103, 104, 106, 107n analog to functional control, 92 analog to syntactic control, 74 characterization, 92 in Dutch resultatives, 93 in Italian, 343 parameters, 87-89 properties, 92 rationale for, 84-85 restriction, 92 Argument merger, 75-77, 79 constraining through aspect, 73-113 formal considerations, 89-94 in French, 77 full, 77 in Italian, 77 in Japanese, 77 partial, 77 in Spanish, 77 Argument raising, 92-93 in Dutch, 247-248 in German, 122, 131-135, 147 in third construction, 131-135
Index
Arguments, 76 case marking on, 216n clitic, 304 dative, 256 external, 108n internal, 99-100 licensing of, 94 proper, 223 in third construction, 263n Argument structure, xiii, xiv, 56, 83, 98, 101, 107n, 394 aspect and, 104 Causative Construction, Type 1, (Lexical Schema), 400-401 causatives and, 369-422 combinatory possibilities for, 92-93 complex, 82 concept of, 395, 419n constraint on, 394-395 embedded, 92-93 in French, 4 in German, 51, 56 in Japanese, 391, 393 mapping to f-structure, 90 matrix, 92-93 in monoclausal causatives, 403 passivization and, 105 of Permissive, 91 semantically construed, 420n term, 420n in Turkish, 391-392, 393 valence and, 418n Argument-to-Function Uniqueness, 78 Argument transfer, 93 Asher, N., 40 Aspect, 94-102, 370 and argument structure, 104 clausal, 94 compositional, 94, 95-96, 99 constraining argument merger through, 73-113 in internal arguments, 99-100 lexical, 95-96 roles, 99 tiers, 94 Aspectual inertia, 99 Aspectual light verbs, 85
425 Aspectual verbs, in Italian, 314 Associativity, 274, 276, 279 Mixed Associativity principle, 281, 282, 289, 301 modally controlled, 310n multimodal, 281 structural postulate, 274 a(rgument)-structure, see Argument structure Attribute-Value Matrix, 90 Auger, J., 6, 39 AUX feature, 343 Auxiliary flip construction, 153n in German, xii, 44-49, 58-59, 117-120, 130, 150, 229 Auxiliary-main verb combination, 415 in German, 376 Auxiliary verbs, 144, 370, 386, 394 argument composition and, 160-177 in case alternation, 257-258 as categorial heads, 420n in Chi-Mwi:ni, 408 cz-passives (Korean), 164, 165, 172, 213n, 214n classification of, 253 construction, 191 emotion, 165, 174, 175-177 (de)emotion, 165, 166, 174-176, 211 in French, xii, 361n in German, 377, 379, 401-404, 407 in Italian, 319, 363n, 364n, 383, 410 in Korean, 161-163, 171 in Malayalam, 404 passive, 257-258 passivizable vs nonpassivizable, 213n perfect, 253 selection, 15, 26, 37, 364n in Italian, 313, 315, 342-344, 364n tense, 253, 261n in Dutch, 239-240 in French, 2, 3, 14-17 inversion of, 239-240 in Turkish, 412 valence of, 171, 398, 405, 415 AVM, see Attribute-Value Matrix Axiomatic presentations, 309n
426 B
Baker, K., 49, 63, 71, 225, 267 Baker, M., 74, 77, 93, 109, 386, 417n, 421 Bantu, 101 morphological causatives, 74 unspecified object deletion, 105 Barry, G., 312 Basic verbs, 5, 11, 13, 29 Bayer, J., 125, 154, 154n, 211, 218 Bech, G., xiv, xvii, 116, 120, 133, 150, 152n, 154, 154n, 215n, 218, 223, 237, 265n, 267 Biclausal causatives, 379-382, 405-409, 416n in Chi-Mwi:ni, 380-382, 392-393, 394-395, 408-409 in German, 379-380, 407-408, 409 Lexical Rule System of Causatives (Universal), 398 Lexical Rule System of Causatives (Universal and Parochial), 399 Biclausality, 377, 383, 385, 410, 415 pretheoretical diagnostics of, 372, 373-375 property of, 375 Bierwisch, M., 133, 154, 262n, 267 Binding, 369, 373-378, 380, 382-385, 390 Binding theory, 392-394, 405, 416n, 418n, 420n in biclausal causatives, 408 kendim anaphor (Turkish) requirements, 413 principles, 25 of German reflexives, 381 Principle A (for French), 26 se stesso anaphor (Italian) requirements, 411-412 Bloch, B., 421 Borsley, R., 132, 152n, 154, 179, 218, 263n, 267, 390, 421 Bottom up deduction, 309n Bouma, G., xiii, 6, 35n, 39, 44, 50, 56, 63, 64, 65, 71n, 72, 205, 218, 225, 234, 239, 240, 248, 261n, 265n, 266n, 267, 270, 320, 368 Branching argument composition and, 223-225 binary, 225
Index
flat, 225 left, 224, 225-230, 261n, 262n nonbinary analysis, 48 right, 224, 262n for verbal complex, 147 Bratt, E. O., 36n, 39, 216n, 218 Bresnan, J., xii, xviii, 39, 77, 83, 90, 91, 92, 101, 110, 111, 263n, 268, 397, 416n, 421 Brindisi province, Italy, Salentino of, 356-360 Briscoe, E., 40 Burzio, L., 37n, 40, 322, 364n, 366, 417n, 421 Butt, M., xii, xiv, xvii, 75, 77, 79, 84, 86, 91, 93, 104, 108n, 110, 343, 366, 387, 419n, 421
C Calabrese, A., 347, 349, 350, 353, 356, 364n, 366 Calcagno, M., 362n, 366 Canonical verb clusters, in German, 242-243 Canon(ical) synsems, 7, 10, 13 Cardinaletti, A., 363n, 366 Casadio, C., 305, 307, 311 Case alternation, 257 auxiliaries in, 257-258 of dative argument, 256 with (de)emotion auxiliaries, 165, 166, 174-176, 211 in Korean, 175-177 assignor, 211 lexical, 174, 176 marking on arguments, 216n in Urdu, 104 principle, 174 roles, 418n structural, 174, 176 Case-driven theory, 211 Case Principle (for Korean), 174, 175 Catalan causative constructions, 34n clitics, 304-305 reflexives, 107n
Index
Categorial Grammar, xii base logic, 272-276 clitic climbing accounts, 305 division, 320 division type-shift, 305 Dutch verb raising example, 282-285 French object clitics analysis, 303-307 logics, 272-285 multimodal, xv-xvi, 280-281 product operators logical, 275 structural, 275 structural control, 276-281 Categorial structure, 397 Causation Hierarchy, 99 Causatives, xvi, 94, 227, 287 active-causative type, 415 affected agent, 85 analysis of, 396-413, 418n alternative, 103-105 prerequisites, 385-396 analytically expressed, 394 argument structure, 103, 369-422 biclausal, 369-370, 379-382 analysis of, 405-409, 416n in Chi-Mwi:ni, 380-382, 392-393, 394-395, 408-409 in German, 379-380, 407-408, 409 pretheoretical diagnostics of, 372, 373-375 case marking alternation, 85 in Chichewa, 88 clausality of, 371-375 in composition FAIRE, 19-22, 22-23 construction, xiii, xvi, xvii in composition FAIRE, 19-22, 22-23 in English, 374 in French, 1, 3, 362n with ha (Korean), 184-188, 188-191, 193-194 in Italian, 34n Romance, 1, 34n, 73 subject-to-object raising, 374 Type 1 (Lexical Schema), 400 Type 2 (Chi-Mwi:ni Instantiation), 408 Type 2 (German Instantiation), 407 Type 2 (Lexical Schema), 406 data from, 371-385 in English, 374, 420n
427 in German, 414, 416n Hindi/Urdu morphemes, 85-86 in Italian, 417n in Japanese, 417n Lexical Rule System of Causatives (Universal and Parochial), 399 Lexical Rule System of Causatives (Universal), 398 mixed, 369-370, 382-385, 420n analysis of, 409-413 in Italian, 382-383, 410-412 in Turkish, 383-385, 412-413 monoclausal, 369-370, 376-379, 420n analysis of, 399-405 in German, 376-377, 401-403, 405 in Malayalam, 378-379, 400, 401, 403-405 pretheoretical diagnostics of, 372-373 morphologically derived, 394 noncanonical, 393 passivization of, 109n Romance, 77, 109n in Spanish, 75 synthetically expressed, 394 terminology, 371-375 three-place, 78, 88, 103, 104, 419n two-place, 78, 88, 103, 109n type 1 of nulti (Malayalam), 404 of riparare (Italian), 411 of "to open" (Turkish), 413 type 2 of dndik (Chi-Mwi:ni), 409 of waschen (German), 407 unithood of, 369 in Urdu, 74, 85-89, 94-102, 106, 109n Causative Schema, Type 1 (German Instantiation), 401 Causative Schema, Type 1 (Italian Instantiation), 410 Causative Schema, Type 1 (Malayalam Instantiation), 403 Causative Schema, Type 1 (Turkish Instantiation), 412 Causativized form, 108n Cause a-structure, 103, 104 relation, 399 Cause clause, 379
428 Caused-SOA, 399-400, 402, 404, 406-407, 409, 411, 413 Caused clause, 379 Causee, 85, 103, 372, 403 affected, 89 in French, 14, 19 oblique, 104 Causer, 372, 399, 403, 406 CCLR, see Complement Cliticization Lexical Rule c-composition, 4, 23 CELR, see Complement Extraction Lexical Rule Centineo, G., 364n, 366 che element (Italian), 351, 365n Chichewa, causatives, 88 Chi-Mwi:ni, 383, 417n dndik verb, 409 argument structure, 391 causatives, 370 biclausal, 380-383, 392-393, 394-395, 408-409 construction, type 2, 408 Lexical Rule System of Causatives (Universal and Parochial), 399 Chinese, 77 resultatives, 94 Cho, J.-H., 209, 218 Cho, J. O., 169, 218 Chung, C, xiv, 184, 215n, 217n, 218, 218n, 231, 268 Cinque, G., 363n, 366 ci-passives (Korean), 164, 165, 172, 213n, 214n, 215n Classifications default role, 90 intrinsic role, 90 Clausal aspect, 94 Clausality; see also Biclausality; Monoclausality pretheoretical diagnostics for, 371-375 Clause-external scrambling, 160 Clause-final, 140 Clause-initial, 117, 138, 140 Clause-internal scrambling, 160, 211 in Hindi, 210 in Korean, 209 Clause Reduction Analysis, 313 Clauses finite, 355-360 relative, 418n
Index
Clause union, 73, 74, 107, 301 definition of, 301 distribution rule Ku for, 301 in Italian, 326 original RG analysis, 75-76 in Spanish, 75 Clefting, 317 in Spanish, 319 Clitic arguments, 304 Clitic attachment, 289 Clitic climbing in French, 2, 14, 19, 21, 34n, 300-302, 364n as argument composition, 305-307 in Italian, 305, 311n, 322, 327-328, 337-338 as argument composition, 305-307 lexical analysis, 357 and long NP-movement, 334 low, 328-332, 330 restructuring verbs and, 323-328, 330 optional, 35n in Salentino, 314, 347-352, 361 and finite clauses, 355-360 in Spanish, 75, 305, 320 in Standard Italian, 314, 348 Clitic climbing trigger verbs in Italian, 322, 328, 334 in Salentino, 348 Clitic clusters, in Italian, 324 Clitic host, 304 Cliticization, 38n Complement Cliticization Lexical Rule, 323, 324, 325, 326, 327, 332-333, 339, 348, 362n downstairs in composition FAIRE, 22-23 in French, 36n, 299-300 upstairs, in composition FAIRE, 19-22 diticized-word word types, 7, 10, 13, 16, 33 n(on)s(ubject), 7, 11 su(bject), 7, 11, 12 Clitic order in French, 292-296, 304-305 in Italian, 324, 329 Clitics, 1, 34n; see also Object clitics; Subject clitics bisyllabic, 328-332, 329, 360, 363n in Catalan, 304-305
Index
enclitic, 315 in French, 5-14, 6, 296-302, 303-304 intrinsic, 5, 12-14 loro (Italian), 328-332, 360, 363n middle si (Italian), 335-336 Lexical Rule, 346 monosyllabic, 329, 360 object in French, 271-312 in French causative constructions, 363n proclitic, 315 as pronominal affixes, 6, 303-304 reflexive, in French, 23-33, 308, 362n Romance, 34n split, 328, 332, 360, 363 clitic sort, 362n Clitic trapping, 5 in composition FAIRE, 33 CLTS {} condition, 321 cl-wd, see cliticized-word Coherence field, 133 Coherent constructions, 146-147, 223 in German, 215n, 231, 250 infinitive, 120 Commutativity, 274, 276, 279 Mixed Commutativity principle, 281, 282 modally controlled version, 310n multimodal, 281 structural postulate, 274 Complement Cliticization Lexical Rule, 323, 324, 326, 327, 332-333, 339, 348, 362n Argument Composition Lexical Rule interactions, 325 Complement Extraction Lexical Rule, 56, 61, 230, 362n Complement inheritance, 43, 54 in German verb clusters, 53-55 Complement inheritance verbs, 53 Complement inheritors, 45 Complementizer field, 56 Complementizers che (Italian) as, 351, 365n ka (Salentino) as, 351, 354, 365n ku (Salentino) as, 350, 351, 352, 356 Romance, 356 Complements finite, xiv Linear Precedence constraints on, 321 nonfinite, xiv direct objects in, 254-255
429 indirect objects in, 255-260 subjects in, 252-254 sentential, 177, 194 valence attributes for, 263n Complement verbs, 205 Complex clauses, in Korean, 188-205 Complex event-bundling, 94 Complex NP shift, 317 Complex predicates, xii-xvii, 167-177, 230, 231, 249, 260, 343, 378-388, 405; see also Predicates advantages of, 172 analysis, 81-83 approaches to, 74-79 defining properties of, 81-82 dynamic, 91 extension of, 159-220 formation, 80-85, 314 mixed theory of, 389-390 in syntax, 91 in French, 1-41 structure, 163 with three arguments, 76 Complex projection base, 262n Complex verbs, in Italian, 314, 316, 320, 342, 343, 358 Complex-word sort, 170 Complex-word-structure sort, 170 Composition, 79 FAIRE, 6, 17-23, 33 in French, 1-41 multimodal forms of, 280-281 phrasal, 282 with tense auxiliaries, 14-17 Compositional aspect, 94, 95-96, 99 COMPS feature, xiii, 230, 231 in Dutch, 64 in German, 56, 61, 63, 70n, 71n, 132 specification of, 151 Conative verbs, in Italian, 322 Configurationality, 159 CONS function, 215n Consider constructions, in Dutch and German, 84, 93 Constituency, xv, 225-233 discontinuous, 241 in Dutch verb clusters, 263n in German VPS, 44-50
430 Constituency (Continued) in Italian, 316-317, 319 requirements for, 262n in Spanish, 319 Constraints; see also specific types on argument structure, 394-395 competing, 70n Constructions auxiliary, 191 auxiliary flip, 153n causative, xiii, xvi, xvii, 1, 3 in French, 362n in Italian, 34n Romance, 34n, 73 coherent, 146-147, 223 in German, 215n, 231, 250 coherent infinitive, 120 control, 181-182, 187, 188, 191, 193 in German, 205 /ia-causative, 188-191 in Korean, 181-182, 215n in Tamil, 181 copular, 34 faire-par, 35 genuine unbounded dependency, 218n /za-causative, 184-188 accusative, 187, 188, 193-194 dative, 187, 188, 191 infinitive, in German coherent, 120 incoherent, 116-118, 231 modal verb, 205 optionally coherent, in German, 215n participle inversion, 65 in Dutch, 67-69 passive, xiv permissive, xiv raising, 183-184, 188, 205 raising-to-object, 182-184, 191-193, 252 in Korean, 216n raising-to-subject, in Korean, 215n s-complement, 194-196 separable prefixes, 65 in Dutch, 65-67 serial verb, 365n simple clitic, 288-292 subject-control, 190
Index
subject-to-object raising causative, in English, 374 subject-to-subject raising, 221 third, xiv, 115-157, 263n as coherence phenomenon, 146-147 in Dutch, 50 as focus raising, 154n in German, 116 as special case of vp extraposition, 122-129 tough in English, 340-341, 363n, 364n in French, 341 in Italian, 316, 340-341, 346, 363n unbounded dependency, 218n, 345 vp-complement, 188-194 in Korean, 177, 178-188 V-Projection Raising, 238 CONTENT value, 200, 202 simplification of, 217n Control, 14, 18, 80, 84, 92, 353 Control constructions, 19, 35n, 37n, 191, 193 in German, 205 /la-causative, 188-191 in Korean, 181-182, 2l5n scrambling in, 181-182, 187, 188 in Tamil, 181 Control operators n-ary, 277 unary, 277-279, 310n Control raising, syntactic, 92-94 Control verbs, 14, 255 coherent, 251 in German, 117, 118, 120, 127, 129, 130, 132, 133, 134, 135, 136, 137, 139, 140, 144, 145, 147, 150, 151, 153n, 223, 255 object-control verbs, 125, 126, 129 subject-control verbs, 115, 116, 124-126, 142-145 incoherent, 251 in Korean, 162, 163, 177, 178, 180-193 Coordination, 291-292, 319, 350 in Italian, 319 in Spanish, 319 Copestake, A., 40, 421 Copular constructions, 34 CSTRUC, see Categorial structure c-structure, 83 Curry-Howard correspondence, 274
431
Index
D
Daliymple, M., 25, 35n, 40, 41, 102, 110, 375, 417n, 420n, 421 Dative argument, 256 Dative controller, 125, 126 Dative ha -causative constructions, 187, 188-191 Dative NP, 302 Dative passives, 256, 258, 267n Dative reflexives, 28 Dative shift, 394 Davies, W., 73, 110 Davis, A., 8, 40 Deemotion auxiliary, 166, 175 Defaults, 35 role classifications, 90 De Hoop, H, 99, 110 De Kuthy-GroBkopf, K., 152n, 155 Deletion, unspecified object, in Bantu, 105 Den Besten, H., 115, 116, 145, 146, 155, 240, 268 De Rooij, J., 71 Descles, J.-P., 25, 38n, 40 Deutsche Forschungsgesellschaft fur Kunstliche Intelligenz, 419n Diacritics, 256, 257 Diesing, M., xi, xvii DIR, 56, 58, 68 Directionality, 51, 52 Direct Syntactic Encoding, Principle of, 417n Distant passives, 254 Distributivity rule K, 281, 283, 293-296 rule Ku, 301 rule Kl, 281, 283, 293 rule K2, 281, 283, 293 Division, 305 in Categorial Grammar, 320 DLR approach, 362n Dobrovie-Sorin, C, 40 Domain, 296-300, 302, 304 locality, 279 Domain formation, 251 Domain-inserted objects, 241 Domain-objects, 265n
Domain trees, 250 Domain union, 242 Dominance, 274, 276 Dorre, J., 310n, 311 Downstairs cliticization in composition FAIRE, 22-23 in French, 22-23, 299-300 Dowty, D., 94, 95, 97, 108n, 110, 213n, 219, 298, 311 Drach, E., 243, 268 DTRS value, 200 Durie, M., 94, 110 Dutch, 37, 52, 77, 225, 266n argument composition, 320 argument raisers, 247-248 consider constructions, 84, 93 cross-serial dependencies, xii floated prefixes, 266n GOV feature, 231 infinite verbs, 71n inversion, 239-240 main verbs, 235 nonfinite modals, 248 obliqueness constraint for, 65 participle inversion, 67-69 resultatives, 93 separable prefixes, 65-67, 240 subordinate clauses, 50, 63 third construction, 50, 145 verbal complexes, 121, 244-245, 247-248 verb clusters, 43, 121, 233, 261n constituent structure, 263n word order, 63-69, 234 word order constraints on, 43-72 verb raising, xii, 279, 309n categorial logic example, 282-285 Dynamic complex predicates, 91
E Edmondson, J., 240, 268 Elimination rule, 274 Embedded clause construction, in Korean, 177 Emonds, J., 3, 40, 322, 361n, 366 Emotion auxiliaries, 165, 166, 174-176, 175-177, 211 Encoding, direct syntactic, 417n
432
Engel, U., 243, 268 English, 183, 355, 361n, 365n aspect, affectedness, and licensing of arguments, 94 causatives, 374, 420n domain tree, 250 parallelism with German, 250-251 passivization, 211 raising, 221 subject-to-object, 374 that element, 365n tough constructions, 340-341, 363n, 364n VP extraposition, 153n Equi, 84 Equi verbs, 125, 126, 223 Ergatives, 30 Ergativity, split, 108n Ersatzinfinitiv, 46, 59, 264n, 265n Essential reflexives, 37n Evers, A., 107n, 110, 121, 155, 241, 250, 268 Exceptional Case Marking, 353 Expletives, 418n, 419n Extended X-bar Theory, 387 External arguments, 108n Extraction, 288 Complement Extraction Lexical Rule, 362n Extraposition, 231 EXTRAP feature, 119, 130, 132, 134-137, 140, 150 partial, verbs, 52 verb phrase, 118-122 in English, 153n plus long-distance scrambling, 145 third construction as special case of, 122-129 third construction interactions, 115-157
F
FAIRE composition, 6, 17-23 clitic trapping with, 33 composition causatives and downstairs cliticization in, 22-23 composition causatives and upstairs cliticization in, 19-22
Index
intransitive complement, 19-20 transitive complement, 19-20 faire-par construction, 35 Finite clauses in Salentino, 355-360 in Welsh, 179 Finite verbs in Italian, 315 placement of, 153n Flat structure, 17, 55, 361n Flat verb phrases, 44, 69 Flemish, 265n Flemish dialects, 240 FLIP feature, 46, 47, 57, 60, 235 Flip, 46, 47, 59 auxiliary construction, 153n in German, xii, 44-49, 58-59, 117-120, 130, 150, 229 Floated prefixes, in Dutch, 266n Focus peripheral, 266n raising, 154n Fokusrandprinzip ('principle of peripheral focus'), 266n Fontana, J., 362n, 366 FOOT features, 364n Frame condition, 274, 280 Franconian, 265n Franco-Provencal, 363n Frank, A., 108n, 110, 152n, 153n, 155 Free pronominals, ku element (Salentino) as, 353 French, 1-41, 323, 362n, 363n, 364n argument composition, 320 argument merger, partial, 77 auxiliaries, xii auxiliary verbs, 361n Binding Principle A for, 26 causative constructions, 34n, 362n, 363n clitic climbing, 2, 14, 19, 21, 34n, 300-302, 364n as argument composition, 305-307 ditidzed-word word types, 7, 10 n(on)s(ubject), 7, 11 su(bject), 7, 11, 12 clitic order, 292-296, 304-305
433
Index
clitics intrinsic, 12-14 lexical analysis of, 5-14 object, 271-312 as pronominal affixes, 6, 303-304 subject, 308 word hierarchy, 7 word types, 6-12 composition FAIRE, 6, 17-23 clitic trapping with, 33 composition with tense auxiliaries, 14-17 downstairs cliticization, 36n, 299-300 inflectional word types, 7, 8 LEXEME word types, 7, 8-9 Non-Constituent Coordination, 319 object clitics, 296-302 analyses, 303-307 data, 285-288 deductive account of, 271-312 order, 287-288, 292-296 as pronominal affixes, 303-304 properties of, 285-288 simple clitic constructions, 288-292 participle agreement, 25 plain-word word types, 7 pronouns strong form, 285 weak form, 285 (CLITIC-)REALIZATION word types, 7, 9 reflexives, 308 reflexive verbs, 23-33 subject clitics, 308 tough constructions, 341 f-structure, see Functional structure Functional abstraction, 274 Functional application, 274 Functional control, 92 Functional heads, 418n Functional Lexicalism, 397, 409 Principle of, 387, 388, 390 Functional structure, xii, 82, 83, 101, 108n; see also Argument structure complex, 82 concept of, 419n mapping from a-structure to, 90 OBJg at, 102 term, 420n
Function-Argument Biuniqueness, 77-78, 106 Principle of, 77 Function composition, xii
G
Gap, 7, 10, 35 Gavarro, A., 277, 304, 312 GB, see Government-Binding Theory Geach, P., xii, xvii Geerts, G., 66, 67, 71 GeilfuB, J., 116, 148, 155 Gentzen sequent presentations, 309n Gerdemann, D., 167, 219 German, 35n, 77, 197, 212, 215 accusative with infinitive, 267n adjunct scrambling, 197 am preposition, 232-233 argument composition, 167-168 auxiliary flip, xii, 44-49, 58-59, 117-120, 130, 150, 229 canonical verb clusters, 242-243 causatives, 370, 386, 414, 416n biclausal, 379-380, 399, 407-408, 409 Construction, Type 2, 407 monoclausal, 376-377, 399, 400, 401-403, 405 Schema, Type 1, 401 coherent constructions, 215n, 223, 231, 250 consider constructions, 84, 93 control verbs, 153n, 205 domain tree, 250 Ersatzinfinitiv, 46, 59, 264n, 265n finite verb placement, 153n incoherent infinitive constructions, 116-118, 231 long-distance passivization, 206 Mittelfeld, 117, 121, 152n branching, 225 NP complements in, I54n scrambling in, 130-131, 145-146 verb clusters, 225, 261n modal verbs, 205, 253
434 German (Continued) nonfinite complements, 251-252 Oberfeldumstellung, xii, 44, 229, 234, 236-238, 264n, 265n order in, 263n optionally coherent constructions, 215n parallelism with English, 250-251 partial vp fronting, 61-63 passives, 144 ACC-NOM, 258, 259 raising as S-embedding, 250 Raising-to-Object type constructions, 252 raising verb constructions, 205 raising verbs, 222-223 reflexives, 381 third construction, xiv, xv, 115-157 third construction triggering verbs, 153n topicalization, 152n verbal complexes, 93, 225, 229-230 constituency, 227 formation, 232 linearization of, 243-244, 245-247 order variation, 228-229 verb clusters, xii, 79, 121, 233, 261n, 388 argument composition, 260, 320 order domain approach to, 260 verbal elements, 262n word order, 51-63, 69, 234 word order constraints on, 43-72 verb phrase extraposition, 115-157 verb phrases, 44-50 VFORM feature, 152n Zwischenstellung ('intermediate placement'), 50, 239 Germanic languages noun incorporation, 109n verbal complexes, 212 verb clusters, 205 Geuder, W., 108n, 110 Gibson, J., 73, 110 Girard, J.-Y., 271, 311 Godard, D., xii, xiii, xvii, 5, 14, 25, 34n, 35n, 38n, 39, 40, 41, 311, 319, 320, 361n, 366 Governee, 57, 170 Government, 51, 228 direction of, 55 in German verb clusters, 55-60 status, 262n
Index Government-Binding Theory, xiv, 386, 418n A-movement approach, 211 in Romance causatives, 77 Governor, 57, 60 Governor constraints, 57, 65 GOV feature, 170, 215n for Dutch, 231 for Korean, 231 GPSG, 309n, 386, 418n Grammar logic, 271, 284, 305 Grammatical function, 373, 375, 404-406, 408, 411, 413 change, 385, 386, 395 features, 90 classification of, 90 function of, 91 in German, 376, 380 in Malayalam, 378 in Turkish, 384 Greek, na element, 364n, 365n Grevisse, M., 23, 40 Grimshaw, J., 23, 25, 37n, 40, 74, 75, 77, 79, 92, 94, 98, 99, 108n, 110, 111, 363n, 366 Grover, C, 363n, 366 Guentcheva, Z., 40 G(O)V(ERN)OR feature, 235, 243, 246, 247, 264n in Dutch, 63, 66, 68 in German, 57, 58, 59, 60
H
ha -causative constructions (Korean), 184-188 accusative, 186-187, 188, 193-194 control, 188-191 dative, 187, 188-191 Haegeman, L., 212, 219 Haeseryn, W., 71 Haider, H., xii, xvii, 116, 152n, 155, 262n, 268 Hale, K., 83, 111, 159, 168, 219 Halvorsen, P.-K., 102, 111 HAUPTVERBCOMPLEX 'main verb complex,' 133 Head adjunction, 282, 283, 284, 289, 309 Head-adjunct-struc, 200 Head clusters, xi
435
Index
Head-Complement ID Schema, 132, 133 Head-complement schema in Dutch, 43 in German, 43-45, 48, 50-54, 60, 61 Head-Driven Phrase Structure Grammar, xii, xiv, 214n, 221, 311n, 313, 386, 387-388, 418n Binding Principles, 25 in Dutch, 43 in French, 4, 10, 303-307 in German, 43, 115-157, 167-168 immediate precedence, 266n impersonal passives, 267n ku element analysis, 352-355 mixed theory of predicate formation, 389-390 order domains, 241 Raising-to-Object constructions, 252 structural case, 174 subject-verb agreement, 252-253 word order theories, 153n Head Feature Principle, 70n, 309n Head-filler structures, 63 Head-finality in Korean, 192 LP constraints on, 218n Heads, 387 functional, 418n Head-Subject-Complement-Adjunct Schema, 197, 200 Head-Subject-Complement Schema, 168, 178 Head-Subject Schema, 178 Head-to-head movement, 74 HEAD values, 266n HEADWORD feature, 170 Heinz, W., 173, 219, 257, 268, 388, 421 Hepple, M, 277, 279, 311, 312 HFP, see Head Feature Principle Higginbotham, J., 90, 111 Higher order type, 288, 303-307 Hindi, 77, 107n causative morphemes, 85-86 long-distance scrambling, 210, 218n N-V complex predicates, 80 Hinrichs, E., xii, xiv, xv, xviii, 35n, 40, 43, 44, 49, 50, 53, 55, 56, 57, 59, 60, 70n, 71, 79, 93, 111, 120, 121, 122, 132, 133, 138,
152n, 153n, 155, 160, 167, 168, 205, 214n, 219, 222, 223, 231, 237, 238, 242, 26ln, 262n, 263n, 264n, 265n, 268, 320, 367, 387, 388, 421 Hohle, T. N., 140, 153n, 155, 243, 254, 268, 362n, 367 Hong, K-S., 216n, 219 HPSG, see Head-Driven Phrase Structure Grammar Hyman, L. M., 3, 17, 40 Hypothetical reasoning, 288, 303, 306
I Idiomatic predicates, 418n Idioms, 372, 419n ID schemas, 153n features sensitive to, 131-132 Head-Complement ID Schema, 132, 133 Partial VP Expansion ID Schema, 148, 149 in third construction, 131-135 Verbal-Complex ID Schema, 132, 133, 139, 231 lida, M., 72, 227, 268, 394, 417n, 421, 422 Immediate precedence, 57, 266n Impersonal passives, 267n long, 142-143 short-distance, 141-142, 143 Impersonal Subject Lexical Rule, 38n Impersonal verbs, 26, 37 Inchoative reflexives, 30 Inclusion principle, 280-281, 289 Incoherent infinitive construction in German, 116-118, 120, 231 Incompleteness, 272 Incorporation, 387, 417n Index Interpretation Principle, 390-396, 409 Infinitives accusative with, 267n constructions coherent, xiv, 120 in German, xiv, 116-118, 120, 231 incoherent, xiv, 116-118, 120 in Dutch, 71n morphological, 262n Inflection, 7
436 Inflectional affixes, ku element (Salentino) as, 351, 352 Inflectional types, 7, 8 Informational identity, 221 Information structure, 397 Causative Construction, Type 1, (Lexical Schema), 400-401 Ingestives, 104-105 Inherent reflexives, 37n INHERJOC feature, 345, 346 Inner zone, see I-zone Insertion, domain, 241 Interaction principle, 281, 289 Internal arguments, 99-100 Interpretive Principle, 203, 217n Intrinsic clitics, 5, 12-14, 22 Intrinsic reflexives, 26-27, 37n Intrinsic roles, classification of, 90, 101 revised, 101 Introduction rule, 274 Inuit, causatives, 109n [INV +], 138 Inversion, 234 in Dutch, 239-240 Invisible affixes, 256 ISTRUC, see Information structure Italian, 37, 417n anaphors, 417n argument merger, full, 77 aspectual verbs, 314 auxiliary selection, 342-344, 364n causatives, 370, 417n constructions, 34n Lexical Rule System of Causatives (Universal and Parochial), 399 mixed, 382-383, 410-412 Schema, Type 1 (Italian Instantiation), 410 che element, 365n clause union effects, 326 clitic climbing, 305, 311n, 314, 322-328, 330, 334, 337-338, 357 clitic climbing trigger verbs, 322, 328, 334 clitic clusters, 324 clitic order, 324, 329 clitics bisyllabic low, 328-332, 360, 361n middle si, 335-336 object, 308 complementizers, 356
Index conative verbs, 322 coordination, 319 long NP-movement, 313, 315, 345-346 analysis of, 336-340, 360 restructuring verbs and, 333-340 loro, 363n middle si, 335-336, 360 modal verbs, 314 motion verbs, 314 Non-Constituent Coordination, 319-320 object clitics, 308 restructuring verbs, 313-368 basic facts, 314-316 classes of, 314 lexical analysis of, 320-344 syntactic structure, 316-320 riparare verb, 411 se stesso anaphor, 383, 411-412, 420n tough constructions, 316, 340-341, 346, 363n verb structures, 35n I-zone, 51, 60, 61 complements, 71n in German verb clusters, 53-55
J
Jaberg, K., 365n, 367 Jackendoff, R., 75, 77, 108n, 111, 420n, 421 Japanese, 218n adversity passive, 213n amelioration of weak crossover effects, 211 argument merger, 77 argument structure, 391, 393 causatives, 109n, 417n passive construction, 213n long-distance, 206 scrambling, 211 long-distance, 211-212 sum, 77, 93 verbal complexes, 227 Jespersen, O., 84, 111 Johnson, M., xii, xviii, 266n, 268 Joshi, S., 74, 85, 87, 88, 92, 94, 95, 103, 109 Jud, J., 365n, 367 Jurafsky, D., 32, 40
437
Index
K
Kachru, Y., 85, 111 ka element (Salentino), 351, 352, 354, 365n Kanerva, J., 90, 101, 110 Kanerva, J. M., 416n, 421 Kang, B. M., 184, 185, 219 Kaplan, R., xii, xviii, 92, 95, 111, 268 Kasper, R., 197, 217n, 219 Kasper, R. T., 264n, 269 Kathol, A., xv, xviii, 55, 57, 60, 63, 70n, 72, 144, 153n, 156, 172, 219, 241, 243, 252, 256, 257, 262n, 263n, 265n, 266n, 267n, 269, 304, 311, 388, 421 Kayne, R., 3, 23, 39, 39n, 40, 363n, 367 Keenan, E., 395, 422 kendim anaphor (Turkish), 413 Kernel sentence, xi Keyser, S. J., 83, 111 Kiparsky, P., 75, 90, 98, 111, 416n, 420n, 422 Kiss, T., 35n, 40, 72, 146-147, 152n, 153n, 154n, 156, 167, 205, 206, 219, 222, 223, 224, 231, 252, 254, 257, 267n, 269 Koenig, J.-P., 19, 25, 32, 35n, 36n, 37n, 38n, 40 KOHAERENZFELD 'coherence field,' 133 Komplexe Projectionsbasis ('complex projection base'), 262n Korean, 159 adjuncts, 217n adjunct scrambling, 196-205 argument composition, 159-220 auxiliary verbs, 161-163, 171 case alternation phenomenon, 175-177 Case Principle for, 174, 175 d-passive, 164, 165, 172, 213n, 214n, 215n complex clauses, 188-205 control construction, 181-182 control verbs, 189, 215n embedded clause construction, 177 GOV feature, 231 ha -causative constructions, 184-188 accusative, 186-187, 188, 193-194 control, 188-191 dative, 187, 188-191, 191 head-final property, 192 Head-Subject-Complement-Adjunct Schema, 197 lexical rules, 195-196
linear precedence constraints, 192-193 nominative case, 216n object-control verbs, 215n passive construction, 213n passivization, 164 long-distance, 206-209 raising construction, 180-181 raising-to-object construction, 216n raising-to-subject construction, 215n raising verbs, 193 s-complement constructions, 194-196 scrambling, 211 clause-internal, 209 long-distance, 159-220 verbal complexes, 159, 160, 170, 211 constituency of, 227 properties of, 215n vp-complement constructions, 177, 178-188, 188-194 VP constituents, 178-181 weak crossover effects, 218n amelioration of, 209-210, 211-212 Kornfilt, J., 125, 154, 154n, 211, 218 Kraak, E., xv, 364n, 367 Krieger, H.-U., 397, 422 Krifka, M., 74, 89, 94, 95, 96, 111 Kripke frame, 272, 277 Kroch, A. S., 116, 157, 232, 265n, 269 ku element (Salentino), 349-352, 364n, 365n in Brindisi and Taranto, 356, 357, 359 equivalents in Balkan languages, 365n equivalents in other languages, 364n HPSG analysis of, 352-355 properties of, 349, 350, 351, 352 Kuh, H., 179, 181, 219 Kupsc, A., 320, 368 Kurtonina, N., 277, 279, 311 Kvam, S., 148, 156
L Lambda calculus, 274 Lambek, J., xii, xviii, 271, 311 Lambek systems, xv, 276 extension of, 280 nonassociative, 274 Lambek-van Benthem calculus, 274 Lamping, J., 110 Lascarides, A., 35n, 40
438 Latin, 349 Lee, Y-S., 211, 219 Legendre, G., 19, 40 Leslie, N., 312 Levine, D., 269 LEX-DTRS attribute, 170 Lexeme, 401, 402 Lexeme selection, 226 LEXEME word type, 7, 8-9, 28 LEX feature, 226, 227, 229-230, 262n in German, 48, 61, 133 in Italian, 321 Lexical analysis of French clitics, 5-14 of Italian clitic climbing, 357 of Italian long NP-movement, 336-340, 360 of Italian restructuring verbs, 320-344 alternative, 344-347 Lexical aspect, 95-96 Lexical case, 174, 176 Lexical entries, 362n, 397 Lexical-Functional Grammar, xii, 73, 95, 101, 221, 263, 386, 387 in French, 35n in Italian, 343 Linking Theory, 98 Principle of Function-Argument Biuniqueness, 77 Lexicalism Functional Lexicalism, 397, 409 Principle of, 387, 388, 390 Morphological Lexicalism, 409 Principle of, 386-387, 417n strong form of, 386-390 Lexical motivation, 257 Lexical rules, 6, 28 Argument Composition Lexical Rule for Italian, 321, 325-327, 331, 335-341, 357, 359 for Korean, 200-202, 205 Causative Construction, Type 1 schema, 400-401 Complement Cliticization Lexical Rule, 323-327, 332-333, 339, 348, 362n Complement Extraction Lexical Rule, 56, 61, 230, 362n
Index
features sensitive to, 131-132 formalization of, 361n for Korean, 195-196 for long-distance scrambling, 216n Long Passive Lexical Rule, 143-144 Medio-Passive Lexical Rule, 31, 32 Middle si Lexical Rule, 335-336, 336-340, 346 for object clitics, 345, 435 for passives, 207-209, 255, 257 System of Causatives (Universal and Parochial), 399 System of Causatives (Universal), 398 Third Construction Lexical Rule, 126, 130, 135-138, 148, 151 preliminary version, 123-124 Lexical semantics, 77-78 Lexical types, recursive, 394 Lexical verbal complements, 332 Lexicon, 388-389 LFG, see Lexical-Functional Grammar Li, C, 213n, 219 Li, Y., 77, 94, 95, 111 Liberation metarule, 216n Liberation principle, 2\6n Light verbs, 83-85 aspectual, 85 incomplete predicate, 84 Linear assignments, 266n Linearization, 233-248 direct, 249 in Dutch, 244-245, 247-248 in German, 129, 243-244, 245-247 Linearization-based theory, 248-260 Linear Logic, xv, 271, 278 Linear order, xv Linear precedence, xiii, 51, 119, 234 constraints, 264n, 266n on adjuncts, 204 on complements, 321, 330, 331, 332 in German, 46, 51, 54 governor, 57, 243 on head-finality, 218n in Korean, 168, 192-193 zone, 54
439
Index
immediate, 246 principles, 51 rules, 138-140, 153n Linguistic composition, multimodal, 280-281 Linking, 89-92, 97-102, 100-102 principles of, 75 and semantics, 106 thematic hierarchy for, 90 theory of, 75, 78, 105, 107, 108 LISP, 215n List concatenation, 35n Locality domains, 279 Locality Principle, 166, 167, 175 definition of, 214n Logic categorial, 272-285 feature, 308 grammar, 271, 284, 305 linear, xv, 271, 278 modal, 272, 278 resource-conscious, xv, 271 rules of, 308 Logical operators, 275, 279, 293 Long-distance agreement, 108n Long-distance passives, 154n, 160, 206 Long-distance passivization in German, 206 in Japanese, 206 in Korean, 206-209 Long-distance scrambling, 154n, 177-205 in Hindi, 210, 218n in Japanese, 211-212 in Korean, 159-220 lexical rules for, 216n theoretical predictions, 206-210 vp extraposition plus, 145 Long NT-movement, in Italian, 313, 315, 345-346 analysis of, 336-340, 360 restructuring verbs and, 333-340 Long Passive Lexical Rule, 143-144 Long passives, 140-144, 254 personal and impersonal, 142-143 low (Italian), 328-332, 360, 361n, 363n Linear Precedence Constraint 1, 330 Linear Precedence Constraint 2, 330 LP, see Linear precedence ban, see LEXEME
M
Mahajan, A., 210, 219 Main verbs, 257-258 in German, 235 nonfinite classification of, 235 in Dutch, 235 vs passive verbs, 257-258 Malayalam, 78, 376, 417n causatives, 370, 382, 386, 408 Lexical Rule System of Causatives (Universal and Parochial), 399 monoclausal, 378-379, 383, 400-405 Schema, Type 1, 403 causativized unergatives, 88 nulli verb, 404 Malouf, R., 39 Manandhar, S., 310n, 311 Manning, C, 26, 40, 56, 72, 75, 79, 80, 111, 268, 391, 392, 393, 394, 417n, 418n, 421, 422 Manzini, M. R., 363n, 367, 368 Mapping principles, 75 MA principle, see Mixed Associativity principle Marandin, J.-M., 37n, 40 Marantz, A., 74, 111, 417n, 422 Marathi anaphors, 420n morphological causatives, 74 reflexives, 375 Markers, 352, 354, 359 Marking Principle, 354 Masica, C., 85, 111 Matiasek, J., 173, 219, 257, 268, 388, 421 Matrix verbs, 205 Matsumoto, Y., 93, 112 Mchombo, S., 90, 101, 105, 109 MC principle, see Mixed Commutativity principle Medio-Passive Lexical Rule, 31, 32 modification, 39 Medio-passives, 23, 30-32 Melillo, M., 350, 367 Melis, L., 37n, 40 Mester, A., 77, 79, 93, 111 Meurers, W. D., 50, 72, 239, 262n, 263n, 269, 362n, 367 Middle-se, 30
440 Middle si Lexical Rule, 335-336, 346 Argument Composition Lexical Rule interactions, 335, 336-340 Middle verb forms, 363n Middle voice, 370, 394 MIDSI-LR, see Middle si Lexical Rule Miller, P. H., xvi, xviii, 5, 6, 10, 18, 19, 24, 25, 33, 35n, 39, 40, 41, 303, 307, 310n, 311, 312, 320, 362n, 364n, 366, 367 Milner, J.-C, 35n, 41 Minimalist Program, 386 Mirror Principle, 74 Mittelfeld, 56, 70n, 117, 121, 152n branching structure of, 225 NP complements in, 154n scrambling in, 130-131, 145-146 verb clusters, 225, 261n Mixed Associativity principle, 282, 289, 301 Mixed causatives, 382-385, 409-413, 420n false positive, 410 in Italian, 382-384, 410-412 Lexical Rule System of Causatives (Universal), 398 Lexical Rule System of Causatives (Universal and Parochial), 399 in Turkish, 383-385, 412-413 Mixed Commutativity principle, 281, 282 Mixed predicates, 389 Mixed theory of predicate formation, 389-390 Miyagawa, S., 211, 220 Modals nonfinite, 248 passives, 255 Modal verbs constructions in German, 205 in German, 253 in Italian, 314 MOD value, 200, 217n Mohanan, K. P., 84, 112, 417n, 422 Mohanan, T., 75, 77, 78, 80, 83, 84, 108n, 112, 419n, 420n, 422 Monachesi, P., xii, xvi, xviii, 5, 6, 34n, 35n, 41, 79, 93, 112, 307, 311n, 312, 320, 323, 330, 335, 344, 355, 361n, 362n, 363n, 364n, 365n, 367 Monoclausal causatives, 376-379, 399-405, 420n in German, 376-377, 400, 401-403, 405
Index
Lexical Rule System of Causatives (Universal), 398 Lexical Rule System of Causatives (Universal and Parochial), 399 in Malayalam, 378-379, 400, 401, 403-405 Monoclausaliry, 371, 382-385, 398, 410, 412, 415 pretheoretical diagnostics of, 372-373 property of, 375 Moore, J., 34n, 35n, 41, 319, 361n, 367 Moortgat, M., xii, xv, xviii, 271, 277, 279, 282, 309n, 311, 312, 320, 367 Moot, R., 291, 312 Morin, J.-Y., 19, 41 Morphological causatives, 394 in Bantu, 74 in Marathi, 74 Morphological changes, 386 Morphological infinitive, 262n Morphological Lexicalism, 409 Principle of, 386-387, 417n Morphological units, 386 Morphology, 303, 386 perfective, 108n template, 324 Morrill, G., 277, 279, 304, 309n, 310n, 312 Moshi, L., 90, 101, 110 Motion verbs, in Italian, 314 Move Alpha, 385 Miiller, S., 71n, 72, 262n, 269
N na element (Greek), 364n, 365n Nagai, N., 206, 220 Nakazawa, T., xii, xiv, xv, xviii, 35n, 40, 43, 44, 49, 50, 53, 55, 56, 57, 59, 60, 70n, 71, 79, 93, 111, 121, 122, 132, 133, 138, 152n, 153n, 155, 160, 167, 168, 205, 214n, 219, 222, 223, 231, 237, 238, 242, 261n, 262n, 263n, 264n, 265n, 268, 320, 367, 387, 388, 421 Napoli, D., 319, 322, 361n, 367 Natural deduction format, 309n Neeleman; A., 75, 77, 84, 92, 93, 112 Negation, 351 Nerbonne, J., 44, 48, 61, 72, 152n, 156, 230, 262n, 269, 397, 422
441
Index
Nespor, M., 363n, 368 Netter, K., 71ra, 72, 152n, 153n, 156 Nino, M.-E., 387, 422 Nishida, C, xvi, xviii, 277, 305, 307, 312, 320, 368 NL, see Nonassociative Lambek calculus NLp, see Nonassociative Lambek Permutation No, Y., 163, 169, 172, 184, 214n, 220 Nominatives passives, 267n in Urdu, 104, 108n Nonassociative Lambek calculus, 274 Noncanonical causatives, 393 Noncan(onical) synsems, 1 Non-Constituent Coordination in French, 319 in Italian, 319-320 Nonfinite clauses, in Welsh, 179 Nonfinite complements direct objects in, 254-255 in German, 251-252 indirect objects in, 255-260 subjects in, 252-254 Nonlocal Feature Principle, 70n, 313, 320, 344, 345 Nonlocal features, 320, 347, 360-361, 364n Nonphrase sort, 170 complex-word subsort, 170 word subsort, 170 Nordlinger, R., 387, 422 Normalization of proofs, 306 Noun incorporation, 75 in Germanic and West Greenlandic, 109n Noun phrase complex shift, 317 dative, 302 long movement analysis of, 336-340, 360 in Italian, 313, 315, 333-340, 345-346, 360 restructuring verbs and, 333-340 vpi-internal, 302 Noun phrase complements, in Mittelfeld, 154n Noun-verb complex predicates, 80 NP, see Noun phrase NPCOMP feature, 133, 134, 139, 148, 226 in German, 45, 47, 48, 50, 60, 61, 122, 132 nulli verb (Malayalan), 404
O
Oberfeld, 238 Oberfeldumstellung, xii, 44, 229, 234, 236-238, 265n Ersatzinfinitiu and, 264n order in, 263n OBJ, 90, 101, 102, 105, 106 OBJo, 90, 101, 102, 105, 106 O B J b e n e f i c i a c y , 101 OBJgo, 101
Object causee, 103 Object clitics; see also Clitics in French, 296-302, 363n analyses, 303-307 data, 285-288 deductive account of, 271-312 order, 287-288, 292-296 as pronominal affixes, 303-304 properties of, 285-288 simple constructions, 288-292 in French causative constructions, 363n in Italian, 308 Lexical Rule for Object Clitics, 345 Romance, 308 Object-control verbs, 125, 126 in Korean, 215n Object position, 37n Objects direct, 254-255 indirect, 255-260 raising, 182-184 in Korean, 216n scrambling in, 191-193 raising-to-object constructions, 182-184 in Korean, 216n scrambling in, 191-193 unspecified, 105-106 deletion for Bantu, 105 Object sharing, 93 OBL0, 90
Oblique causee, 104 Obliqueness, 44, 52, 64, 70n, 416n constraint for Dutch, 65 o-command, 218n Oehrle, R., xii, xviii, 277, 279, 282, 309n, 312 O'Neill, P., 268, 421 Optionally coherent construction, in German, 215n
442 Order domains, 241 in German, 260 insertion, 241 union, 242 Outer zone, see O-zone Overgeneration, 281, 282, 298, 305, 311 O-zone, 51, 60, 63 complements, 64, 71n
P PAGE grammar, 419n Partial extraposition, 50, 52 Partial verb phrases, 44 derivation of, 52 fronting of, xiii, 231, 262n in German, 44, 48, 52, 61-63, 69 Partial VP Expansion ID Schema, 148, 149 Participle agreement, in French, 25 Participle inversion, 65 in Dutch, 67-69 Passives, 34n, 93, 94, 105-106, 258, 259, 373, 418n adversity, 213n auxiliaries, 257-258 d-auxiliary (Korean), 164, 165, 172, 213n, 214n construction of, xiv, 213n dative, 256, 258 DAT-NOM, 267n distant, 254 in German, 144 impersonal, 232, 267n long, 142-143 short-distance, 141-142, 143 in Japanese, 213n in Korean, 213n lexical rules for, 207-209, 255, 257 long, 140-144, 254 Lexical Rule, 143-144 long-distance, 154n, 160, 206 modal, 255 personal long, 142-143 short-distance, 141-142, 143 recipient, 256 remote, 154n short-distance, 144 verbs, 257-258
Index
Passivizable auxiliaries, 213n Passivization, 105, 109n in Chi-Mwi:ni, 381 in English, 211 in German, 206 in Japanese, 206 in Korean, 164, 206-209 long-distance, 206-209 Patient, 372 Patientlike roles, 101 Perception verbs, 287, 302 Perfective morphology, 108n Peripheral focus, principle of, 266n Perlmutter, D., 3, 5, 19, 37n, 39, 39n, 41, 73, 75, 76, 109, 313, 366 Permissives, 91 construction, xiv in Urdu, 74, 75, 79, 80-82, 84 Permutation, 305 Personal passives long, 142-143 short-distance, 141-142, 143 Peters, S., 268 Phrasal category, 70n Phrasal composition, 282 normal mode, 289 stronger mode, 289 Phrasal selection, 226 Phrase sort, 170 Phrase Structure Grammar, 206-207 Pied-piping, 316-317 Plain-word word types, 7, 9, 13, 16 Polish, argument composition, 320 Pollard, C, xii, xviii, 4, 25, 41, 52, 70n, 72, 144, 148, 152n, 153n, 154n, 156, 166, 173, 183, 188, 200, 214n, 217n, 218n, 220, 221, 225, 232, 252, 255, 262n, 263n, 267n, 269, 313, 344, 352, 354, 362n, 368, 390, 396, 397, 416n, 422 PP complement, 339 Precedence, 274, 276; see also Linear precedence immediate, 57, 266n in verb cluster, 244 PRED feature, 82
443
Index
Predicates; see also Complex predicates active-causative type, 415 analytic, 371, 375, 397, 414, 415 in German, 379-380, 407, 409 in Italian, 383 similarity with synthetic predicates, 394 in syntax, 389 idiomatic, 418n impersonal, 222 mixed, 389 modification, 99 raising, 261n synthetic, 371, 375, 397, 415 in Chi-Mwi:ni, 380-382, 409 in lexicon, 389 similarity with analytic predicates, 394 unmarked, 419n Prefixes in Dutch, 65-67, 240, 266n floated, 266n separable, 65-67, 69, 232, 240 Prefix + verb combinations, in West Germanic, 240 Principle M, 74 Pro-drop, 56, 396 Progressives, 232 Projection Principle, 77 Pronominal affixes, 7, 34 clitics as, 303-304 in French, 6, 303-304 Pronominals, free, ku element (Salentino) as, 353 Pronouns, in French strong form, 285 weak form, 285 Proofs, 274 normalization of, 306, 307 Proto-roles, 97-98 Prototype, 395 Przepiorkowski, A., 320, 368 Pseudo-affixes, 262n Pseudo-raising, 211, 212 Pullum, G., 172, 216n, 220, 303, 312
Q Questions, 418n
R Raising, 84, 92, 93, 106, 205, 211, 212, 261n constructions in English, 221 in Korean, 180-181 scrambling in, 183-184, 188 in English, 211 pseudo-raising, 211, 212 right node, 317, 319, 361n subject-to-subject, 221 verb, in Dutch, 279, 282-285 verb projection, 212 V-projection, 238, 265n Raising-to-object constructions, 373, 374, 379, 380, 381, 383, 406, 416n in German, 252 in Korean, 182-188, 191-193, 216n scrambling in, 191-193 Raising-to-subject construction, in Korean, 215n Raising verbs, 418n, 419n constructions in German, 205 in French, 14 in German, 125, 205, 222-223, 250, 251-253 in Korean, 193 in nonfinite complements, 252 Rambow, O., 93, 112, 116, 152n, 156 Ramchand, G., 95, 99, 105, 109n, 112 Raposo, E., 73, 110 (CLITIC-)REALIZATION words, 6, 7, 9 Reanalysis, 212, 250 Reape, M., xv, xviii, 56, 72, 153n, 156, 241, 242, 248, 252, 257, 269 Recipient passives, 256 Recursive lexical types, 394 Recursive x0 structures, 262n Reduced thematic hierarchy, 98 Reduced verbs, 5, 11, 13, 29 Reflexives, 377-379, 394, 403 in Catalan, 107n classes of, 23 dative, 28 in French, 5, 22, 308 in French causative constructions, 362n
Index
444
Reflexives (Continued) in German, 381 intrinsic, 23 in Marathi, 375 medio-passives, 23, 30-32 true, 23 Reflexive verbs essential, 37n in French, 23-33 inherent, 37n intrinsic, 26-27, 37n true, 27-30, 37n Reflexivity, 273, 279, 298 Reis, M., 265n, 270, 376, 422 Reis' Generalization, 377, 403, 416n Relational Grammar, xi, 73, 107n Clause Union, 74, 75-76 Relative clauses, 418n Remnant topicalization, 262n Remote passive, 154n Rentier, G. M., xii, xviii, 227, 231, 234, 261n, 270, 320, 368 Residuation, 273, 274, 275, 277, 279, 280 Resource-conscious logic, xv, 271 Resource management, 280, 288, 307 rules, 309 Restructuring, 316 Restructuring Rule, 313 Restructuring verbs auxiliary selection of, 364n in Italian, 313-368 alternative analysis of, 344-347 basic facts, 314-316 classes of, 314 lexical analysis of, 320-344 structures, 318 syntactic structure, 316-320 Romance, 93 in Salentino, 347-360, 361 Resultatives, in Chinese, 94 RG, see Relational Grammar Richter, F., 153n, 157 Riehemann, S., 397, 422 Right branching, 224, 227, 262n Right node raising, 317, 361n in Spanish, 319 riparare verb (Italian), 411 Ritter, E., 94, 95, 112 Rizzi, L., 35n, 41, 313, 314, 316, 322, 340, 360, 361n, 364n, 368
RNR, see Right node raising Roles patientlike, 101 proto-roles, 97-98 thematic, 97-98, 108n classification of, 101 hierarchy, 90, 99 0-role, 74 association, 212-213 types, 95 Romance languages causatives, 74, 109n constructions, 34n, 73 Government-Binding approach to, 77 three-place, 78 two-place, 78 clitics, 34n, 308 complementizers, 356 complex predicates, 79, 80, 91 object clitics, 308 restructuring verbs, 79, 93 Rosch, E., 395, 422 Rosch, S., 422 Rosen, C., 73, 110 Rosen, S., 75, 77, 94, 95, 112, 417n, 422 Rouveret, A., 3, 36n, 39, 39n, 41 Rule K, 293-296 Rule Ku, 301 Rule Kl, 281, 283, 293 Rule K2, 281, 283, 293 Rumanian, sd element, 364n Rutten, J., 115, 116, 145, 146, 155 Ruwet, N., 37n, 41 Ryu, B-R., 167, 220
S sd (Rumanian), 364n Sag, I., xii, xiii, xvi, xviii, 4, 5, 6, 10, 24, 25, 33, 34n, 35n, 38n, 39, 40, 41, 52, 56, 70n, 72, 75, 79, 111, 153n, 154n, 156, 166, 183, 188, 200, 214n, 217n, 218n, 220, 221, 252, 255, 263n, 268, 269, 303, 307, 310n, 311, 312, 313, 320, 344, 352, 354, 362n, 365n, 366, 367, 368, 390, 391, 392, 393, 394, 396, 397, 416n, 418n, 421, 422 Sailer, M., 153n, 157, 253, 270 Saito, M., 198, 204, 218n, 220 Saksena, A., 85, 86, 88, 94, 108n, 112
445
Index
Salentino, 365n of Brindisi, 356-360 clitic climbing, 314, 347-352, 361 clitic climbing and finite clauses, 355-360 clitic climbing trigger verbs, 348 ka element, 351, 352, 354, 365n ku element, 349-352 HPSG analysis of, 352-355 properties of, 349, 350, 351, 352 restructuring verbs, 314, 347-360, 361 of Taranto, 356-360 Salvi, G., 368 Santorini, B., 116, 157, 232, 265n, 269 Saraswat, V., 110 Saturation in Phrase Structure Component, 390 Scalise, S., 7, 41 s-complements, see Sentential complements Scottish Gaelic aspect and argument structure, 95, 100 linking, 100 Scrambling, 172, 188, 223 in accusative ha -causative constructions, 187, 188, 193-194 adjunct, 196-205 of adverb phrases, 204, 205 amelioration of weak crossover effects by in Hindi, 210 in Korean, 209-210 case marking on arguments and, 216n clause-external, 160 clause-internal, 160, 211 in Hindi, 210 in Korean, 209 in control constructions, 181-182, 187 in dative /tfl-causative constructions, 187, 188, 191 in Japanese, 211 in Korean, 211 long-distance, 145-146, 154n, 177-205 in Hindi, 210, 218n in Japanese, 211-212 in Korean, 159-220 lexical rules for, 216n theoretical predictions, 206-210 vp extraposition plus, 145 in Mittelfeld, 130-131, 145-146 in raising constructions, 183-184
in in in in
raising-to-object constructions, 191-193 s-complement constructions, 194-196 subject-control constructions, 190 vp-complement constructions, 188-194, 205 Secondary patientlike roles, 101 Sells, P., 92, 112, 169, 216/1, 220 Semantically construed argument structures, 420n Semantic dependency, 217n Semantics lexical, 77-78 linking and, 106 of middle verb forms, 363n S-embedding, 250, 253-254 SEM value, 217n Sentences, kernel, xi Sentential complements, 177 analysis of, 166-167 constructions, 163 adjuncts in, 198-199, 202-203 analyses of, 194-196 scrambling in, 194-196 Separable prefixes, 65 in Dutch, 65-67, 240 placement of, 240 Sequence union, 55, 242 Serial verbs, 93, 365n se stesso anaphor (Italian), 411-412, 420n Shaumyan, S., 40 Short-distance passives, 141-142, 143 Shuffle, 35n, 242 si (Italian), middle, 335-336, 360 Lexical Rule, 335-336, 336-340, 346 Sign, 7, 272, 275, 280 sign sort, 170, 241, 263n, 265n nonphrase subsort, 170 phrase subsort, 170 Simple-word sort, 170 Simpson, J., 324, 368 Skytte, G., 362n, 364n, 368 SLASH feature, 48, 61, 62, 122, 263n Sort clitic, 362n Sort hierarchy, 170 Southern German dialects, 265n
446 Spanish argument merger, partial, 77 causative constructions, 34n causativization, 75 clefting, 319 clitic climbing, 305, 320 coordination, 319 right node raising, 319 verb structures, 35n Specificity, 86, 104 Split ergativity, 108n SPR feature, 56 Spurious ambiguity, 48 Sranan, complex event-bundling, 94 S(emantic)-structure, 102, 106 Stabler, E. P., 277, 278, 312 Starke, M., 363n, 366 Status, 264n (quasi-)lexical, 227 Status government, 262n Stem, 9 Stiebels, B., 232, 270 Strong distribution rule K, 293-296 Strong Lexicalism, 7 Hypothesis, 386 Structural case, 174, 176 Structural control, 276-281 Structural operators, 279, 283, 297 unary, 277-279, 310n Structures, 271 flat, 17, 55, 361n in Italian, 318 left-branching, 225-230 right-branching, 224, 227 syntactic, 318 Structure sharing, 320 Subcategorization frame, 6 principle, 52 verbal complement selection via, 232 SUBCAT feature, 168, 230, 231, 390-391 Subject clitics; see also Clitics in French, 308 Subject-control construction, in Korean, 190 Subject-control verbs, 115, 124-126, 142-144 Subjecthood, 221
Index
Subjects expletive, 419n in Korean, 215n in nonfinite complements, 252-254 raising-to-subject construction, 215n understood, 222 valence attributes for, 263n Subject-to-object raising causative construction, in English, 374 Subject-to-subject raising constructions, 221 Subject-verb agreement, 252-253 SUBJ feature, 56, 90, 132, 230 Subordinate clauses, in Dutch, 63 Syntactic control, 74 Syntactic control-raising, 92-94 Syntactic encoding, direct, 417n Syntactico-semantic complexes, synsems, 4, 6, 263n, 321 aff(ixal), 7 canon(ical), 7, 10, 13 noncan(onical), 7 of sort clitic, 362n Syntactics, 303, 418n Syntactic structure, in Italian, 316-320 Synthetic predicates, 371, 375, 392, 397, 415 in Chi-Mwi:ni, 380-382 in German, 379 in Italian, 383 in lexicon, 389 similarity with analytic predicates, 394
T
Tamil control construction, 181 word order, 179 Taranto province, Italy, Salentino of, 356-360 Tasmowski, L., 18, 39, 39n, 41 te (Albanian), 364n Template morphology, 324 Tenny, C, 74, 89, 94, 112 Tense, 370
447
Index Tense auxiliaries, 253, 261n in Dutch, 239-240 in French, 2, 3, 14-17 inversion of, 239-240 Terzi, A., 356, 364n, 368 that element (English), 365n Thematic roles, 97-98, 108n classification of, 101 hierarchy, 90, 99 reduced, 98 Thematic tiers, 94 Theme suppression, 105 Theta Criterion, 77-78, 106-107 O-identification, 90 O-role, 74 association, 212-213 Third construction, xiv, 52, 115-116 argument composition in, 131-135, 150-151 as coherence phenomenon, 146-147 in Dutch, 50 as focus raising, 154n forcing argument raising in, 131-135 in German, 116 linearization approach for, 129 ID schemas in, 131-135 as special case of VP extraposition, 122-129 verbal arguments, 263n VP extraposition interactions in Dutch, 145 in German, 115-157 linearization approach for, 129 Third Construction Lexical Rule, 126, 130, 135-138, 148, 151 preliminary version, 123-124 Third construction triggering verbs, 125, 151, 153n dual properties of, 129-131 Thompson, S., 213n, 219 Three-place causatives, 78, 88, 103, 104 Tiers aspect, 94 thematic, 94 TO-BIND|OC feature, 345
Topicalization, 117, 152n, 228 in German, 152n remnant, 262n TOPO attribute, 266n Topological fields, 51, 243, 304 Topological indices, 266n Topology, 51 Tough constructions, xvi in English, 340-341, 363n, 364n in French, 341 in Italian, 313, 316, 340-341, 346, 363n Trace, in object position, 37n Transformation, 385 Transformational Grammar, xi Transformational rules, 418n Transitivity, 8, 13, 19, 20, 27, 36, 273, 279 Tree modifying operations, 385 Turkish, 417n argument structure, 391-392, 393 causatives, 370 Lexical Rule System of Causatives (Universal and Parochial), 399 mixed, 383-385, 412-413 Schema, Type 1 (Turkish Instantiation), 412 kendim anaphor, 413 TVP, 214n Two-place causatives, 78, 88, 103, 109n
U
UG, see Universal Grammar Unaccusative Hypothesis, 364n Unary control operators, 277-279, 310n Unbounded Dependency Constructions, 313, 320, 344, 345 Uniformity of Theta Assignment Hypothesis, 74, 77 UNIONED attribute, 251 Universal Grammar, 419n Unmarked predicates, 419n Unspecified object deletion, 105 Unspecified objects, 105-106 Unterfeld, 238, 239 Upstairs cliticization, in composition FAIRE, 19-22 .
448 Urdu, 77, 107n, 108n case marking, 104, 108n causatives, 74, 79, 85-89, 94-102, 109n affectedness, 96-97 data, 85-87 lexical and compositional aspect, 95-96 morphemes, 85-86 causativization pattern, 86 data, 85-86 dynamic complex predicate formation in syntax, 91 nominatives, 104, 108n permission, 93 permissives, 74, 75, 79, 84, 94 analysis, 82 formation, 80-81 transitives, 87 Uszkoreit, H., 70n, 72, 116, 146, 153n, 154n, 157 UTAH, see Uniformity of Theta Assignment Hypothesis
V Valence, 217n, 390, 391, 394, 396, 416n, 418n auxiliary, 398, 405, 415 in Korean, 171 Principle, 52, 212, 337 for subjects vs complements, 263n Van den Toorn, M. C, 71 Van Geenhoven, V., 75, 109n, 112 Van Noord, G., xiii, 44, 50, 56, 63, 64, 65, 71n, 72, 205, 218, 225, 234, 239, 240, 248, 261n, 266n, 267, 270, 320, 368 Van Riemsdijk, H., 152n, 157, 212, 213, 219 VCOMP feature, 263n VCOMPL feature, 70n, 230-233, 231 verbal complement selection via, 232 verbal complex formation via, 232 Veenstra, T., 365n, 368 Vendler, Z., 95, 113 Verbal arguments, 263n Verbal complement-complex, 231 Verbal complements, 70n selection via VCOMPL vs SUBCAT, 232
Index
Verbal complexes, 43, 44, 70n, 221-270, 261n branching for, 147 constituency of, 225-233 discontinuous, 238-248 discontinuous (partial), 241 in Dutch, 121, 244-245, 247-248 formation via VCOMPL, 232 in French, 296-302 in German, 49, 121, 227-230, 243-247 Germanic, 212 HAUPTVERBKOMPLEX 'main verb complex,' 133 in Japanese, 227 in Korean, 159, 160, 170, 211, 215n, 227 linearization, 233-248 linearization-based theory, 248-260 as recursive X° structures, 262n split, 239 West Germanic, 261 Verbal-Complex ID Schema, 132, 133, 139, 231 Verbal spine, 263n Verb clusters, xiii, 70n, 261n canonical, 242-243 complement inheritance, 53-55 constituent structure, 263n in Dutch, 43-72, 63-69, 121, 233, 234, 263n fronting, 229, 262n in German, 43-72, 79, 121, 233, 242-243, 262n, 388 Germanic, 205 I-zone, 53-55 in Mittelfeld, 225 partial, fronting of, 229 topological field, 56 verbal elements, 262n West Germanic, 247 word order, 63-69, 234 word order constraints on, 43-72 Verb constructions auxiliary, 160-177 control, 205 in German, 205 modal, 205 raising, 205 serial, 365n
Index
Verb movement, 243 Verb phrase, xi complements, 122 adjunct scrambling, 198-199 analysis, 163-166 understood subject of, 222 complement structure, 18, 163 constituents, 178-181 flat, 44, 60, 69 in German, 44-50 in Korean, 178-181 nonextraposed, 150 partial, 44 derivation of, 52 fronting of, xiii, 44, 48, 52, 61-63, 69, 231, 262n Partial VP Expansion ID Schema, 148, 149 relatives, 117, 118, 123 topicalization of, 152n vp,-internal NP, 302 Verb phrase-complement constructions adjuncts in, 202-203 analyses of, 188-194 in Korean, 177, 178-188, 188-194 scrambling in, 205 in Korean, 188-194 Verb phrase extraposition, 118-122 double, 119, 149-150 in English, 153n plus long-distance scrambling, 145 third construction as special case of, 122-129 third construction interactions, 129 in Dutch, 145 in German, 115-157 Verb projection raising, 212 Verb raising, 120, 121 in Dutch, 279, 309n categorial logic example, 282-285 Verb-raising verbs, in Dutch, 63 Verbs, 5 Acl, 144 dndik (Chi-Mwi:ni), 409 argument structure specification, 4 aspectual, in Italian, 314 aspectual light, 85 auxiliary, 144 in French, 361n
449 basic, 5, 11, 13, 29 causative, 287 classification of, Aktionsart, 95 clitic, intrinsic, 5 clitic climbing trigger in Italian, 322, 328, 334 in Salentino, 348 complement, 205 complement inheritance, 43, 53 complex, in Italian, 314, 316, 320, 342, 343, 358 conative, in Italian, 322 control, 14, 255, see also Control Verb coherent, 251 in German, 117, 118, 120, 127, 129, 130, 132, 133, 134, 135, 136, 137, 139, 140, 144, 145, 147, 150, 151, 153n, 223, 255 incoherent, 251 in Korean, 189 (de)emotion, 174 equi, object-control, 125, 126 EQUi-control, 223 finite in Italian, 315 placement of, 153n governing expletive subjects, 419n impersonal, 26, 37 light, 83-85 main, 257-258 in German vs Dutch, 235 nonfinite, 235 vs passive verbs, 257-258 matrix, 205 medio-passive, 30 middle forms, 363n modal, 296-300 in German, 253 in Italian, 314 motion, in Italian, 314 non-finite, in Italian, 315 nulll (Malayalam), 404 object-control, in Korean, 215n partial extraposition, 52
450 Verbs (Continued) passive, 257-258 perception, 287, 296-300, 302 placement in German, 243-244 raising, 14, 125, 251, 418n, 419n characterization of, 253 in German, 125, 205, 222-223, 250, 251-253 in Korean, 193 in nonfinite complements, 252 reduced, 5, 11, 13, 29 reflexive essential, 37n in French, 23-33 inherent, 37n intrinsic, 26-28, 37n true, 27-30, 37n restructuring, xvi auxiliary selection of, 364n in Italian, 313-368 Romance, 93 Salentino, 347-360 riparare (Italian), 411 sequence of, 238 serial, 93 STATUS Of, 152n
subject-verb agreement, 252-253 third construction, 52 third construction triggering, 125, 151, 153n dual properties of, 129-131 types embedding nonfinite complements in German, 251-252 verb-raising, 63 Verb-second placement, 138, 153n Verb structures in Italian, 35n in Spanish, 35n Vergnaud, J.-R., 3, 36n, 39, 39n, 41 Verkuyl, H., 94, 96, 113 Versmissen, J. A. G., 277, 279, 310n, 312 VFORM feature, 59, 264n for German, 152n Visser, F. Th., 157 VP, see Verb Phrase VPR, see Verb projection raising
Index
V-projection raising, 265n constructions, 238
W Wanner, D., 363n, 368 Warlpiri, 159 WCO effect, see Weak crossover effect Weak crossover effect, xiv, 160, 206 amelioration of in Japanese, 211 in Korean, 209-210, 211-212 Webelhuth, G., xvi, 256, 257, 258, 259, 267n, 270, 370, 371, 390, 396, 397, 415, 418n, 419n, 420n, 421, 422 Wechsler, S., 8, 41 Wedekind, J., 95, 111 WEIGHT feature, 16 Welsh, 132 word order, 179 Wesche, B., 152n, 153n, 156 West Greenlandic languages, noun incorporation, 109n wh-movement, in Italian, 318 Williams, E., 92, 113 Withgott, M, 324, 368 Wollstein-Leisten, A., 116, 148, 157 Word formation, 386 Word-likeness, 262n Word objects, 362n Word order constraints on verb clusters in German and Dutch, 43-72 domains, 55 flat, 56 in Dutch verb clusters, 63-69, 234 flipped, 46, 47, 59 in German verb clusters, 51-63, 69, 234 government of, 55-60 head-complement schema, 51-53 HPSG theories, 153n in Korean complex clauses, 188-205 normal, 47 in Tamil, 179 in Welsh, 179 Word-phrase status, 232
451
Index
Words (CLITIC-REALIZATION, 6, 7, 9 types of, 6-12 Word sort, 170 Wunderlich, D., 232, 270
X XCOMP feature, 82 x° structures, recursive, 262n
Y Yatabe, S., 206, 211, 220 Yoo, E. J., 173-174, 183, 188, 192, 215n, 220
Yoon, J. H. S., 163, 220 Yoshimura, N., 211, 218n, 220
Z Zaenen, A., 35n, 37n, 41, 90, 91, 98, 108n, 110, 113, 268 Zimmer, K. E., 3, 17, 40 ZONE, 53-54, 67, 71n inner, see I-zone outer, see O-zone Zone LP constraints, 54 Zubizarreta, M. L., 3, 41, 363n, 368 Zwicky, A., 216n, 220, 262n, 270, 303, 312 Zwischenstellung ('intermediate placement'), 50, 59, 239
This page intentionally left blank
SYNTAX AND SEMANTICS
Volume 1 edited by John P. Kimball Volume 2 edited by John P. Kimball Volume 3: Speech Acts edited by Peter Cole and Jerry L. Morgan Volume 4 edited by John P. Kimball Volume 5: Japanese Generative Grammar edited by Masayoshi Shibatani Volume 6: The Grammar of Causative Constructions edited by Masayoshi Shibatani Volume 7: Notes from the Linguistic Underground edited by James D. McCawley Volume 8: Grammatical Relations edited by Peter Cole and Jerrold M. Sadock Volume 9: Pragmatics edited by Peter Cole Volume 10: Selections from the Third Groningen Round Table edited by Frank Heny and Helmut S. Schnelle Volume 11: Presupposition edited by Choon-Kyu Oh and David S. Dinneen Volume 12: Discourse and Syntax edited by Talmy Givon Volume 13: Current Approaches to Syntax edited by Edith A. Moravcsik and Jessica R. Wirth
Volume 14: Tense and Aspect edited by Philip J. Tedeschi and Annie Zaenen Volume 15: Studies in Transitivity edited by Paul J. Hopper and Sandra A. Thompson Volume 16: The Syntax of Native American Languages edited by Eung-Do Cook and Donna B. Gerdts Volume 17: Composite Predicates in English Ray Cattell Volume 18: Diachronic Syntax: The Kartvelian Case Alice C. Harris Volume 19: The Syntax of Pronominal Clitics edited by Hagit Borer Volume 20: Discontinuous Constituency edited by Geoffrey J. Huck and Almerindo E. Ojeda Volume 21: Thematic Relations edited by Wendy Wilkins Volume 22: Structure and Case Marking in Japanese Shigeru Miyagawa Volume 23: The Syntax of the Modern Celtic Languages edited by Randall Hendrick Volume 24: Modern Icelandic Syntax edited by Joan Maling and Annie Zaenen Volume 25: Perspectives on Phrase Structure: Heads and Licensing edited by Susan D. Rothstein Volume 26: Syntax and the Lexicon edited by Tim Stowell and Eric Wehrli Volume 27: The Syntactic Structure of Hungarian edited by Ferenc Kiefer and Katalin E. Kiss Volume 28: Small Clauses edited by Anna Cardinaletti and Maria Teresa Guasti Volume 29: The Limits of Syntax edited by Peter Culicover and Louise McNally Volume 30: Complex Predicates in Nonderivational Syntax edited by Erhard Hinrichs, Andreas Kathol, and Tsuneko Nakazawa