complex predicates - link.springer.com978-94-009-4522-7/2/1.pdf · complex predicates 11.0. ......

12
PART II COMPLEX PREDICATES

Upload: vudien

Post on 29-May-2018

217 views

Category:

Documents


0 download

TRANSCRIPT

PART II

COMPLEX PREDICATES

PART II

COMPLEX PREDICATES

11.0. INTRODUCTION TO PART II

11.0.1. Types of Complex Predicates

In this second part, we examine what we will call COMPLEX PREDICATES, that is structures in which a verb and its infinitival complement appear to form a single unit. There are two major cases of complex predicates discussed in the literature: the 'causative' constructions of Kayne (1975), Rouveret and Vergnaud (1980), and the 'restructuring' constructions of Rizzi (1976a), (1978a). The first are associated with causative and perception verbs like make, let, see, hear, the second with various verbs, some of which are semantically 'modals' (like English want/will, must, can), some others 'aspectuals' (like begin, continue), and some verbs of motion. Perhaps the best known characteristic of these constructions is that objects of the complement verb cliticize to the main verb, as in the causative case in (la), and the restructuring case in (1 b).

(1) a. Giovanni 10 fa leggere a Piero. Giovanni it makes read to Piero

Giovanni makes Piero read it.

b. Giovanni 10 vuole leggere. Giovanni it wants to read

Giovanni wants to read it.

Much of the framework of assumptions we developed in Part I will aid us in our investigation here: the hypothesis that there are ergative verbs, the theories of inversion, of auxiliary assignment and past participle agreement, of small clause relatives, the hypothesis that there is recon­struction in LF, will all play crucial roles. In most cases, the interaction between earlier parts and the discussion of complex predicates will be mutual in that, not only will the earlier results enable us to probe more effectively into the syntax of complex predicates, but also the behavior of complex predicates will appear to confirm those results.

While the various analyses of Part I may thus be confirmed by com­plex predicates, but could nevertheless be put forth independently, our analysis of cliticization relies rather heavily on the analysis of complex predicates, and was not given in full in Part I for this reason. Yet a full discussion of complex predicates cannot be postponed since some

217

218 PART II

assumptions regarding the nature of c1iticization are needed to even begin such a discussion. For example, what identifies the structures in (1) as complex predicates is the assumption that c1iticization is always in some sense local, so that there must be a closer than usual relation between the two verbs in these cases. Therefore, in the third subsection of this introduction we outline our theory of c1iticization in full, giving some arguments, and referring forward to others that will come later.

One of the central claims of this part is that the various types of com­plex predicates result from a single syntactic process affecting different types of infinitival complements. To prepare the ground for this, in the next subsection we review the typology of infinitival complements and see how this is characterized in GB-theoretical terms.

The overall organization of this part is rather straightforward. In chapter 4 we examine causative constructions and in chapter 5 restruc­turing constructions. In chapter 6 we examine reflexive c1itics, in part considering their interaction with complex predicates, which we claim has consequences for the theory of c1itics at large, and in part elaborating on the results of 1.5.

1I.O.2. Verbs with Sentential Complements

Within the LGB framework, the class of verbs which are subcategorized for a sentential complement breaks down into four subclasses, as deter­mined by two lexical parameters: the ability to trigger S-deletion, and the ability to assign a O-role to the subject position (Os), as in (2).

(2) Class S-del. f)s

I + + (ECM) II + (Raising) III + CEqui') IV

All four subclasses are attested empirically. Thus class I is the class of Exceptional Case Marking (ECM) verbs, like English believe in (3).

(3) John believed IsBill to have left].

If believe deletes the S (maximal projection) of its complement, it will govern the embedded subject. Since it assigns subject O-role, we expect, following the discussion in 3.1 relating f)s to Case assignment (i.e. f)s .....

A), that it will assign Case to the latter embedded subject. (3) will thus be well-formed.

Class II is the class of Raising verbs, like seem in (4).

(4) Johniseems [stito have left]

COMPLEX PREDICATES 219

If seem triggers S-deletion, (4) will be well-formed since the trace will be governed, as prescribed by the ECP. On the other hand, cases like * It seems [PRO to talk about ourselves] will be ungrammatical since PRO is governed, while it must not be (cf. 1.0.3). Since seem does not assign sub­ject O-role, we also predict the ungrammaticality of * It seems [John to have left] (for lack of Case on John) as we discussed in 3.1.

Class III is the traditional class of 'Equi' verbs, i.e. of Control verbs like hope in (5). (As in Part I, we ignore the distinction between S and S in the diagrams. Thus, there is an S in (5) since this is an instance of class III).

(5) John hoped [sPRO to leave].

Since government of the embedded subject by the main verb does not obtain here, PRO is allowed, while lexical subjects are not, as illustrated by *John hoped [Bill to leave], where lack of government prevents Case assignment. (Since hope assigns Os we assume that it would otherwise assign Case.)

In English, members of Class IV are not very numerous. We find remain, as in It remains [PRO to talk about ourselves], although this verb is presumably also in class II, given This remains to be seen, which must involve Raising. Another case is suffice, as in It suffices [PRO to meet the requirements].! But in Italian there are several such verbs. One is bisognare of (6a). Others are listed in (6b). (Some of these verbs take indirect objects controlling the embedded subject PRO).

(6) a. [e] bisogna [sPRO parlargli] (it) is necessary to talk-to-him

b. merita, conviene, basta, (it) is worthwhile (it) is advantageous (it) suffices

capita, piace, sembra, giova. (it) happens (it) pleases (it) seems (it) is helpful

This characterization of infinitival complements accounts for important similarities in the distribution of Raising and ECM phenomena and for the essentially complementary distribution of Raising-ECM and Control. In particular it accounts for the fact that the verbs that allow ECM as in (3) are (to a large degree) the same that allow passivization with subject­to-subject raising as in (7).2

(7) Bill j was believed [s tj to have left].

The correspondence follows from our assumption that that S-deletion is involved, since only if a verb triggers S deletion will it govern the embedded subject so as to either assign Case as in (3), or satisfy the ECP as in (7). Thus, verbs which do not trigger S-deletion, like hope in (5) will not allow the type of passivization in (7), as shown in (8).

220 PART II

(8) *Bill j was hoped [s tj to leave].

The correspondence between ECM and the type of passive in (7) is further confirmed by the fact that in Romance, where, as we will show, ECM cases like (3) are lacking, cases like (7) are also rather generally lacking. Another similarity in the distribution of ECM and subject-to­subject raising is the lack of both in the presence of a filled complemen­tizer, as illustrated by (9).

(9) a. *John knows [[what] Bill to do].

b. *Billjwas known [[what] tj to do].

This can also be naturally captured by our system, provided that we assume that the intervening complementizer prevents government of the embedded subject by the main verb. We may simply say that S-deleltion is possible only if no lexical material separates the S from the S boundary. Then Case assignment will fail in (9a), and the ECP will be violated in (9b), whence the parallel ungrammaticality. The same consid­erations that rule out (9b) will then account for the fact that we find no Raising verbs that take indirect questions (Cf. * John seems what to read).

While the system thus described, with the two parameters plus-or­minus S-deletion and plus-or-minus Os will account for the relative distribution of the various types of infinitival complements, we assume that in the unmarked case subcategorization for a sentential complement is neutral between the tensed and the infinitival options. Thus, alongside of (3) (ECM) we find John believed [that Bill had left], with (4) (Raising) we find It seems [that John has left], with (5) (Equi) we find John hoped [that he could leave], and with (6a) (of the fourth class) we find Bisogna [che tu gli parli] '(It) is necessary that you talk to him'.

Our claim in the chapters that follow will be that the process which forms complex predicates is intrinsically neutral, not formulated to apply to one or the other class in (2).3 In particular, we will take causative constructions to arise from application of that process to verbs of class I and their infinitival complements, and restructuring constructions to result from its application to verbs of classes II and III and their comple­ments. No complex predicate seems to arise from structures of class IV: a fact which will call for an explanation.

It is obvious, at least for classes II and III (restructuring), that only some members of those classes give rise to complex predicates, so that some sort of constraint must be assumed to govern the process in ques­tion. Rizzi (1976a), (1978a), observes that the distinction between restructuring and non-restructuring verbs seems to run along semantic lines, with semantically more impoverished verbs allowing restructuring more readily. A typical minimal pair is the one in (10), where 'want' allows restructuring, while 'wish' does not.

COMPLEX PREDICATES

(10) a. Lo voglio leggere.

b. *

(I) it want to read

I want to read it.

(I) Lo desidero leggere. it wish to read

221

With Rizzi, we will assume that only semantically weak verbs can combine with others to form complex predicates, although we will not attempt to express this idea formallly. (For a discussion of this issue, see Napoli (1981 ).) As we expect on the basis of our claim that the same process is involved, we find that an analogous constraint, based on the ~emantics of the verb, is operative within the class of causative­perception verbs, as shown in (11 ).

(11 )a. Gliel'ho visto prendere. (I) to-him-it have seen take

I ~aw him take it.

b.?* Gliel'ho osservato prendere. (I) to-him-it have observed take

11.0.3. Ihe Syntax ofClities

There are two major questions that arise in any attempt to characterize the syntax of clitic~. One concerns the nature of the locality conditions holding between the clitic and the position which receives the O-role. The other question is whether clitics are moved or base-generated. Since the two can be regarded as independent, we will consider them separately, in that order.4

In the literature, we find two important pieces of evidence that rela­tions between a clitic and the position to which the relevant O-role is assigned (the relevant O-position) fall under the binding conditions, ~pecifically the Specified Subject Condition (SSC), which we may use a~ a descriptive category here. The first piece of evidence has to do specifically with complex predicates. Thus, the literature on causative constructions, in particular Kayne (1975), and analogously the literature on restructuring constructions, in particular Rizzi (197oa), (197Ha) shows us that the cases in which an object clitic can appear on a verb higher than the one with which it would normally be associated, are just the cases for which one has independent evidence that no subject is present between the higher and the lower verb. We will see this in detail below.

The second piece of evidence is provided by the discussion of extrac­tion from NP in Italian in Cinque (19HO). Cinque observes first that the only constituents that can be 'extracted' from NPs by cliticization are

222 PART II

phrases of the form di NP 'of NP' which cliticize as ne, while cliticization of other phrases, for example of a NP'to NP', is always impossible.5 He then notes that not all di NP phrases can be extracted, and argues that it is only those that represent the subject of the head noun that can, which explains why a NP phrases, always being objects, are excluded. We thus have a perfect SSC effect.

If Cinque is correct, the facts he discusses provide a stronger argument for the view that cliticization obeys the SSC than complex predicates do. For, although both Kayne and Rizzi assume the SSC, their discussion does not rule out certain alternatives. One of these consists of regarding cliticization as sensitive to clause boundaries. Another, adopted in fact in the context of a discussion of complex predicates in Radford (1977), has cliticization sensitive to VP boundaries. A third, elaborated in Borer (1981), and essentially also compatible with an analysis of complex pred­icates, consists of assuming that a clitic must govern the O-position. How­ever, neither S boundaries nor VP boundaries can be invoked to account for the facts Cinque discusses (nor do NP boundaries seem to play any role), so that the first two alternatives are discounted. As for the third, it also fails to extend to Cinque'S facts. Consider (12).

(12) a. Apprezziamo [Ia generosita [di Giovanni]] (we) appreciate the generosity of Giovanni

b. Nei apprezziamo [la generosita lie]] (we) of-him appreciate the generosity

While it is entirely plausible to assume that Ie] in (12b) is governed by the head noun fa gene rosita, thus satisfying the ECP, it is less plausible to assume that it is governed (across the head noun) by the clitic." Besides, the government alternative would shed little light on the fact that some di NP phrases can be cliticized, while others, and other PPs cannot.

Although various questions still arise, as I will briefly discuss below, it seems to me that at the present stage the binding theory remains the best candidate to account for the distribution of cliticization. I will therefore assume that the latter is indeed what is relevant, and that ec's related to clitics are anaphors, like NP-traces.7

We now consider the question of movement versus base-generation. Kayne (1975) assumes that clitics are moved. In the context of his discussion and given his theoretical framework, there are strong reasons to assume a movement analysis. For example, the effects of his assump­tion that reflexive cliticization is cyclic while non-reflexive cliticization is post-cyclic, cannot be readily duplicated within a base-generated system (as we will see in the course of the following chapters). While Kayne's analysis has been very influential, more recent studies, in particular Rivas (1977), Jaeggli (1980), Borer (1981) have claimed that Romance object

COMPLEX PREDICATES 223

clitics are base-generated. The typical argument that these studies have adduced in favor of the base-generation hypothesis, has to do with clitic­doubling, as in the Spanish example in (13).

(13) Le entregue la carta a el. (I) to-him delivered the letter to him

Under a movement analysis, the clitic in (13) would have no source since the O-position is occupied by a lexical phrase (a il). On the other hand a base-generation analysis will face no problem of principle.

As we mentioned in 1.4 above, we are also claiming that object clitics are base-generated. However, without rejecting the clitic-doubling argu­ment, our argument will be of a rather different type. We will argue that the base-generation and movement hypotheses actually predict different configurations of data with respect to complex predicates, the base­generation hypothesis making the right predictions. In essence, we will argue that, whereas a movement analysis of cliticization requires clitic-ec relations to be well-formed only at S-structure, a base-generation analysis requires such relations to be well-formed at D-structure as well, whence the difference in empirical predictions. As we discussed in 1.4, the requirement that clitic-ec pairs hold well-formed relations at all levels at which they exist follows from the projection principle, since the latter rel­ations are necessary for transmission of the O-role to the clitic by the ec. It is entirely reasonable to suppose that the well-formedness conditions that hold at D-structure are the same as those that hold at S-structure. Since we assume the binding theory for S-structure, we will assume the binding theory for D-structure as well. This raises a certain question con­cerning the element PRO.

It is clear that the distribution of PRO is determined exclusively by S-structure. Thus, while in S-structure PRO can only be in subject, and not in object position as shown by (14), its D-structure position is immaterial, as shown by (lSa) which must be well-formed to produce (ISb).

(14) a. [PRO to invite Bill] would be a good idea.

b. *[for Bill to invite PRO] would be a good idea.

(15) a. [[e] to be invited PRO] would be nice.

b. [PRO; to be invited til would be nice.

In the LOB framework, the distribution of PRO and the contrast between (14a, b) is determined by the requirement that PRO be un­governed at S-structure, which in turn derives from the binding theory (d. 1.0.3 above). The lack of a requirement that PRO be ungoverned at D-structure as well, illustrated by the well-formedness of (15), would

224 PART II

follow from the assumption that the binding theory does not apply in D-structure. Our claim that it does, may thus seem to rule out (15) incorrectly.

However, our claim does not imply that the binding theory must apply to PRO in D-structure. In fact the reasons why we assume that it must apply to clitic-ec relations is that those relations must exist, so as to allow transmission of O-role to the clitic. But this reasoning does not carry over to PRO, or for that matter to other elements, like each other in (16).

(16) TheYiexpected [each otherito be invited tJ

In (16), each other is within a certain local domain of they in derived structure, but it is not in D-structure, showing that the D-structure posi­tion of each other, like that of PRO, is irrelevant to well-formedness. As we pointed out in 3.3, our approach here consists of assuming that only some relations must obtain at all levels, as determined by the requirement that O-structure be represented at all levels, that is by the projection principle. The relations which will be required to obtain at all levels will then be those that involve O-role transmission. Others will typically be required to obtain only at LF. We therefore take the binding theory to be available at all levels, but to apply at any particular level only to those relations which must obtain at that level.

We finally consider some doubts and difficulties which remain under the proposed account of cliticization. One set of problems is represented by the fact that clitics can be 'extracted', with various degrees of success. out of small clauses, as in (17), (18). Rather similar facts are brought up as problematic for the sse account of cliticization in Kayne (1975, 4.6).

(17) a. Quellibroi g~i fu [se ti dato ti [~) da Giovanni)

that book to-him was given by Giovanni

b. Giovanni; ~e pareva [set; deluso [f))

Giovanni of-it seemed disappointed

Giovanni seemed disappointed about it.

(18)a. ? Ne considero [se Giovanni deluso [ell I

(I) of-it consider Giovanni disappointed

I consider Giovanni disappointed about it.

b. ?? Gli consideravo [se Maria fedele [~])

(I) to-him considered Maria faithful

I considered Maria faithful to him.

COMPLEX PREDICATES 225

Note that in general small clauses behave just like other clauses with respect to the binding conditions, as shown for example by the behavior of the pronoun in (19a) and of the reflexive in (19b).

(19) a. Mariaj considerava [sc Giovanni orgoglioso di leiil Maria considered Giovanni proud oJ her

b. *Mariaj considerava [sc Giovanni orgoglioso di se stessa iI Maria considered Giovanni proud oj herself

The exceptionality of (17), (18) is therefore peculiar to cliticization. One might attempt to account for (17), (18), by su~gesting that locality

conditions on cliticization are relaxed when the local domain does not contain an element that takes clitics (i.e. a verb). Results change in fact whenever the sc is replaced by an infinitival complement (which does contain a verb), as shown by the contrast between (17b) and (20a), whose grammatical counterpart is (20b).

(20) a. *Giovanni j ne pareva [Sti essere deluso [ell I I

Giovanni oj-it seemed to be disappointed

b. Giovanni j pareva [s ti esserne deluso [ell ..... '-----',

Giovanni seemed to be-oj-it disappointed

However, alone, this consideration would incorrectly lead us to expect a relaxation of the conditions in the case of extraction from NP, too. (That is, cliticization out of NPs should be possible not just in the cases noted by Cinque, but generally.) We may then appeal to the fact that a verb taking a sc complement and the predicate in the sc bear a special relation to each other, as evidenced by the existence of selectional dependencies between the two (a matter discussed in Stowell, (1983». This second consideration (which has no counterpart in the case of extraction from NP), might make it natural to regard the main verb and the predicate in the sc as some sort of discontinuous predicate, and to account in this way for the apparent exceptionality of (17), (18). The first consideration (i.e. the absence of a target for cliticization within the sc) would still be required however, to distinguish the exceptional behavior of clitics from the non exceptional behavior of other elements (cf. (19». Matters are complicated further by the difference in the level of acceptability between (17) and (18). This difference seems to stem at least in part from the dif­ferent positions of the lexically realized subject of the sc (within the sc in (18), outside the sc in (17». For cases like (18) improve if the subject is extracted, as shown by the contrast between (18b) and (21).

226 PART II

(21) ?Mariai gli era cons ide rata [ti fedele [ell !~----------------------------~!

Maria to-him was considered faithful

While it remains unclear how the different role of lexical and non-lexical subjects is to be captured by our grammar, we may note that this pheno­menpon stresses the relevance of the notion of subject to the distribution of cliticization.8

Another question relates to the evidence in Cinque (1980). Although it achieves a significant degree of explanatory power, Cinque's account of cliticization from NP is in one important respect problematic, as he notes. Cinque shows that the possibility of extracting from NP by clitici­zation corresponds rather closely to the possibility of extracting by Wh­movement. This suggests of course that analogous factors are at work in the two types of extraction. Yet,.as has been argued in recent years (cf. in particular Rizzi (1978b), Chomsky (1980), (LGB), Freidin and Lasnik (1981 », the view that Wh-movement is subject to the binding conditions (specifically, that Wh-traces are anaphors) does not seem tenable (cf. 1.0.3, fn. 16, ch. 1).

In spite of the various reservations that the above discussion may suggest, we will continue to assume that cliticization is governed by principle A of the binding theory, i.e., that ec's related to clitics are anaphors. We further assume, pending discussion of the relevant evidence, that object clitics are base-generated and related to their ec's at all levels, under the same locality conditions (i.e. the binding theory) at each level.

NOTES

I As for 'be+adjective' cases, like It is illegal [PRO to park here], the considerations of Note 13, ch. 1, Note 63, 73, ch. 2, suggesting that there are no 'ergative' adjectives, would lead us to analyze the sentential argument as having subject rather than object O-role, so that these cases would not instantiate the subcategorization __ So On the other hand, infinitivals occurring with Raising adjectives, like the one in John; is likely [t; to win] can­not be readily analyzed as having subject O-role. More plausibly, they have object O-role. The issue raised in the Notes cited above is thus further complicated by these cases. 2 There are a few exceptions to' this generalization, noted and discussed in Marantz (1981, 3.1.1). These are cases like feel, reason, say, as in Elmer was felt to have overstepped his boundaries/*We felt Elmer to have overstepped his boundaries.

Notice also that, while one finds cases of Raising across indirect objects, as in John; seemed to us [t; to have left], there are no cases of ECM across indirect objects (cf. *1 proved to him [the problem to be unsolvable]). This difference follows from the assumption of Stowell (1981), that Case-marking (at least in English) requires linear adjacency (cf. also Note 4, ch. 4). 3 While the classification in the text does not consider cases in which there is a NP or a :':>P object beside the sentential complement, we will see that restructuring applies to some such cases as well (cf. ergative verbs andare, venire). 4 The two questions are not entirely independent however, since - as we will argue -base-generation entails stronger locality effects than movement.

COMPLEX PREDICATES 227

5 The term 'extract' is used here for expository convenience, not to imply cliticization by movement. 6 Following Belletti and Rizzi (1981), we assume however that the ec related to ne is go­verned by the verb (and hence, presumably, by ne) in the case of quantified nominals, as in (i) (cf. Note 23, ch. 1).

(i) Ne ho visti [due [e]] (I) of-them saw two

Government by V in (i) versus lack of government by V in (12b) is supported by a differ­ence in past participle agreement, as we noted in Note 49, chapter 1. 7 The text discussion, and especially our reference to relations between c1itics and empty categories, is partly misleading since we do not presume that the syntax of clitics is substantially different in cases of clitic-doubling, which of course do not involve empty categories. H Contrasts between cases in which the cliticization path crosses a lexical subject and cases in which it only crosses its trace are even stronger with the 'pro-predicate' 10 (cf. Note 11, ch. 3), as illustrated by the following.

(i) a. Ritengo [Giovanni deluso[ (I) believe Giovanni disappointed

b. * Lo ritengo [Giovanni Ie]] !

(I) it believe Giovanni

c. ?Giovannii 10 era ritenuto [tile]] ,

Giovanni it was believed

(ii) a. Giovannii e [ti ammalato[ Giovanni is sick

b. Giovannii I~ e [ti[~]]

Giovanni it is

c. C'e [Giovanni ammalato[ there is Giovanni sick

d. *Ce 10 e [Giovanni [ell !

there it is Giovanni

We have no explanation for the greater sharpness of the above contrasts compared with those of the text.