Bare Resultatives

This paper provides a complex-predicate analysis of resultatives in terms of bare phrase structure. Our main claim is that the predicate types and thematic mapping rules required for simplex verbs can be used to explain core properties of resultative constructions if a strict versions of Chomsky’s inclusiveness condition is adopted. The analysis supports a view of subjects as external arguments in the sense of Williams 1980, rather than as structurally defined, as in Stowell 1981,1983, Bowers 1993, etc. 1 Subjects in bare phrase structure 1.1 On the structural definition of subject The theory of bare phrase structure proposed in Chomsky 1995a is an attempt to reduce the machinery involved in the generation of syntactic trees to a minimum. Its basic assumption, inclusiveness, states that properties of syntactic structure must ultimately derive from information stored in the lexicon. On its most restrictive interpretation, which we will adopt here, it reads as follows: (1) Inclusiveness: The syntactic properties of a non-terminal node are fully recoverable from its daughters; the syntactic properties of a terminal node are fully recoverable from its lexical entry. By implication, bar levels, coindexation and the like must be dispensed with and analyses relying on such notions must therefore be reconsidered. In this paper we will deal with one such analysis, namely that of subjects as specifiers. The most common view of subjects in government and binding theory accords them the status of specifier of a lexical head. Thus, the DP in (2) is the subject of X, if X is Ad Neeleman & Hans van de Koot 354 a noun, preposition, adjective or verb (Stowell 1981, 1983; Kuroda 1986; Koopman & Sportiche 1991). (2) [XP DP [X’ ... X ... ]] Specifiers, in this theory, are defined as the unique constituent which is the daughter of a maximal and the sister of an intermediate projection. Without bar levels, this definition of specifier and the associated view of subjects cannot be maintained. The seriousness of the problem becomes clear when we compare the underlying representations assigned to unergative and unaccusative verbs in bare phrase structure ((3a) and (3b) respectively). If subjects are generated in the projection of V, these representations are identical (abstracting away from linear order): (3) a. [V DP V] b. [V V DP] There is a way in a theory of bare phrase structure to maintain the view that subjects are specifiers of lexical heads. Chomsky (1995b) defines a head’s complement as the first phrase merged with it and its specifier as the phrase merged subsequently. On the assumption that unergative verbs are hidden transitives with an incorporated object, as argued by Hale & Keyser (1993), the structure in (3a) must be revised as in (4). Since the subject is now merged after merger of a nominal category, it counts as a specifier. (4) [V DP [V N-V tN]] As will be obvious, this analysis stands or falls with the claim that unergatives are hidden transitives. On the face of it, there is some evidence for deriving unergatives through noun-incorporation: many unergative verbs have a nominal counterpart, witness pairs like a laugh and to laugh. Such pairs would be related by incorporation of the noun into an empty verbal head: (5) [V they [V laughN-0V tN]] Despite its initial appeal, this proposal must be rejected on the basis of empirical evidence, discussed by Don (1993), which suggests that in some cases the noun is derived from the verb, rather than vice versa. This means that at least some unergatives cannot have the underlying structure in (5). Bare resultatives 355 Don analyzes analogous noun-verb pairs in Dutch. The Dutch data are more informative than the English ones, since Dutch nouns either take het or de as their definite determiner. As in English, Dutch verbs have either regular or irregular verbal inflection. When we look at existing noun-verb pairs in Dutch, all combinations are attested, except one: there are no nouns which take het as a definite determiner, while the related verb is inflected irregularly. (6) Dutch noun-verb pairs Nde Nhet Vregular de fiets the bicycle fietsen fietste gefietst cycle cycled cycled het werk the work werken werkte gewerkt work worked -worked Virregular de slaap the sleep slapen sliep geslapen sleep slept slept Don argues that this pattern can be accounted for by assuming two morphological processes, whose existence is in fact supported by a wealth of further evidence. The process in (7a) derives nouns that take de from verbs, while the process in (7b) derives regular verbs from nouns. The analysis thus attributes the gap in the paradigm in (6) to the fact that there is no process that can connect a het-taking noun with an irregular verb. (7) a. V 6 Nde b. N 6 Vregular One consequence of Don’s proposal is that irregular unergative verbs cannot be derived from the corresponding noun. On the contrary, the noun slaap in (6) must be derived from the homophonic verb. But this means that Dutch irregular unergatives cannot be analyzed as hidden transitives. Therefore, the problem in (3) seems unavoidable for this class of verb: their subject, if generated internally to the verbal projection, is structurally indistinguishable from a complement (and hence from the subject of an unaccusative). The problem in (3) only arises if we combine inclusiveness (which underlies bare phrase structure) with the assumption that subjects are generated internally to the Ad Neeleman & Hans van de Koot 356 projection of the predicative head. Whereas inclusiveness seems unavoidable in a minimalist theory, the same cannot be said about the second assumption. In fact, Williams 1980 and subsequent work argues that subjects are external arguments; that is, generated outside the maximal projection of the predicative head. On this view, it becomes trivial to distinguish the underlying subject of an unaccusative from the subject of an unergative. The former is generated internally to the verbal projection, while the latter occupies a position in the projection of a higher head, represented as " in (8). (8) a. [" DP [" " V]] b. [" " [V V DP]] It seems therefore that bare phrase structure necessitates a view of subjects as VPexternal. 1.2 The status of the mediating head The debate about the position of subjects is directly relevant to the question of how syntactic structures are mapped onto semantic representations. Stowell’s uniform definition of subjects as specifiers presupposes that the semantic relation of predication corresponds to exactly one syntactic configuration. By contrast, Williams’ theory allows subjects to be realized in a variety of positions, which means that the same semantics can be encoded by more than one structure. Although bare phrase structure favours an analysis of subjects as external arguments, it does not answer the question whether semantic relations are mapped onto unique syntactic configurations. Indeed, it is fair to say that both Stowell’s and Williams’ conception of the mapping between syntax and semantics can be incorporated into a minimalist syntax. A one-to-one mapping is guaranteed if the relation between a predicate and its subject is mediated by a unique functional head. This is the line taken in Hornstein & Lightfoot 1987, Bowers 1993, Den Dikken 1995 and other work. Henceforth we will refer to this head as H* and to the analysis assuming it as the single head hypothesis: (9) Single Head Hypothesis [" DPsubject [" " X]], where " = H* On Williams’ approach, the head whose projection hosts subject and predicate does not have a mediating function. Consequently, predication is possible in the projection of any head: Bare resultatives 357 1 On this theory, a DP may be related to multiple predicates. See for discussion Chomsky 1986 and Williams 1983, 1994. (10) Arbitrary Head Hypothesis [" DPsubject [" " X]], where " is any head Consider the effects of the two analyses for secondary predication. If we assume that a predicate and its subject can be hosted by any projection, the minimal structure for sentences containing an object and a subject depictive are as in (11a) and (11b), respectively. In (11a) the head allowing external realization of the subject is V, while v has this function in (11b).1 (11) a. [v John [v met-v [V Mary [V tV drunkA ]]]] b. [v John [v [v met-v [V tV Mary]] drunkA ]] By contrast, if predication involves a unique mediating head, the secondary predicate cannot enter into a direct relation with its semantic subject. Rather, it must be the complement of a H* head, whose specifier is PRO, as shown in (12). The correct interpretation then results from control by one of the arguments of the verb. (Note that v must be identified with H* if (9) is correct). (12) a. [H* John [H* met-H* [V Mary [V tV [H* PRO [H* H* drunkA ]]]]]] b. [H* John [H* [H* met-H* [V tV Mary]] [H* PRO [H* H* drunkA ]]]] There is a trade-off between the complexity of syntactic structures and the complexity of the syntax-semantics mapping. The structures in (11) are maximally simple, but require that predication can be established in more than one syntactic configuration. The structures in (12), on the other hand, allow a maximally simple mapping, but necessitate complications of the syntax. Not only is the structure larger, but in addition a relation of control must be assumed. It will be clear, then, that considerations of simplicity are not sufficient to choose between (9) and (10). The choice is an empirical one. On the basis of depictives, no convincing empirical argument can be constructed, since PRO is inaudible and the properties of obligatory control and predication are largely identical (although problems concerning the distribution of PRO and the non-availability of optional control remain to be addressed). However, as we will argue in this paper, resultative predicates provide evidence in favour of the view that there is no unique Ad Neeleman & Hans van de Koot 358 2 There is no evidence for overt movement of the verb to v in the Germanic OV languages. The movement could be covert or, alternatively, VP-shell formation may be restricted to VO languages (Haider 1997; Neeleman & Weerman 1999). externalizing head, as expected if semantic relations can be encoded in more than one syntactic configuration.