This research focuses on the interface between syntactic and semantic components of unaccusativity. Its starting point is the observation that the so-called unaccusative mismatches (i.e. the lack of correspondence between the semantic components denoted by a verb and the syntactic behavior that might be predicted on the basis of those components) are problematic only to the extent one expects unaccusative and unergative verbs to represent syntactically AND semantically homogeneous classes. Since most of the diagnostics of unaccusativity/unergativity (such as auxiliary selection in French, impersonal passives in Dutch, resultative constructions in English, etc.) tend to identify semantically coherent subsets of verbs, the challenge for a theory of unaccusativity is to single out the syntactically relevant components of meaning in different languages.
In common with others (e.g. Levin & Rappaport 1995), I assume that some semantic components are more relevant than others in determining the syntactic status of intransitive verbs, and that what must be explained is the systematic linking of a multi-categorial lexical-semantic level to a necessarily binary and discrete syntactic level. At the same time, I depart from conventional assumptions in maintaining that the level of lexical-semantic representation is a multidimensional space arising out of the intersection of a number of gradient dimensions such as DYNAMIC/STATIC, TELIC/ATELIC, and CONCRETE/ABSTRACT. These (potentially universal) gradient continua - "hierarchies" - distinguish core unaccusative/unergative verbs from progressively more peripheral verbs; the notion of "dynamic change" is at the core of unaccusativity and that of "atelic non-dynamic activity" is at the core of unergativity. The closer to the core a verb type is, the stronger the link between its single argument and the position of internal or external argument at argument structure, and the more determinate its syntactic status as either unaccusative or unergative. Languages choose different 'minimal triggers' of unaccusativity along these hierarchies: for example, "existence of state" in Italian is sufficient to guarantee unaccusative status, whereas in French - which has a narrower class of syntactically unaccusative verbs - the crucial component is "change of location". On this view, linking rules governing the mapping between lexical semantics and syntax are language-specific and may be regarded as the main locus of crosslinguistic variation. Fig. 1 represents the Unaccusative Hierarchy for monadic verbs and the particular minimal triggers of unaccusativity (indicated in all caps) selected by four languages.
Fig. 1: Unaccusative hierarchy for monadic verbs ("A >> B" = "A is more central than B")
French: CHANGE OF LOCATION >> change of state >> continuation of state >>existence of state
English: Change of location >> CHANGE OF STATE >> continuation of state >> existence of state
Dutch: Change of location >> change of state >> CONTINUATION OF STATE >> existence of state
Italian:Change of location >> change of state >> continuation of state>> EXISTENCE OF STATE
This representation of unaccusativity is similar to recent work on markedness hierarchies carried out in Optimality Theory (Bakovic 1995; Legendre et al. 1995), where it is assumed that the grammar of a particular language dichotomizes a hierarchy into two parts that are treated differently by the syntax. In contrast with this work, however, my theory predicts that (a) the greater the distance between the minimal trigger and the core, the larger the class of syntactic unaccusatives, and the more degrees of variation a language displays, and (b) verb categories adjacent to the minimal trigger exhibit a higher degree of syntactic variation. This characterization thus addresses one of the controversial issues of linguistic theory: the formal incorporation of gradience and systematic variation within the grammar (Sorace, forthcoming). In this respect, my proposal represents a refinement of Dowty's (1991) idea that the unaccusative/unergative distinction corresponds semantically to the two concept clusters of Proto-Patient and Proto-Agent, which are inherently non-discrete.
This view has been supported by an increasing amount of evidence and underlies several ongoing projects.
At the synchronic level, there is ample evidence that unaccusative and unergative hierarchies exist in the intuitions of native Italian speakers and guide their grammaticality judgments on syntactic properties such as auxiliary selection and NE-cliticization (Sorace 1992; the same design was successfully replicated in Bard, Robertson and Sorace1996). Incorrect auxiliary choice (AVERE instead of ESSERE with unaccusatives, and ESSERE instead of AVERE with unergatives), for example, is less unacceptable for peripheral than for core verbs. Furthermore, descriptive studies of Italian (e.g. Berruto 1987; Rohlfs 1969) indicate that there is more variation in auxiliary usage for peripheral verbs than for core verbs.
The existence of unaccusative/unergative hierarchies is also supported by the findings from a study of auxiliary selection and impersonal passives in Dutch (Sorace and Vonk, forthcoming).The data show orderly gradience in the judgments of native Dutch speakers on zijn and hebben largely corresponding to the intransitive hierarchies identified for Italian. In addition, they show that the acceptability of impersonal passives (a construction traditionally regarded as a diagnostic of unergativity) is also affected by semantic factors, particularly 'internal control', which cut across the unaccusative-unergative distinction.
At the diachronic level, there is evidence from studies on the historical development of auxiliaries in Romance (e.g. Benzing 1931, Tuttle 1986) showing that core verb types tend to be the last to be affected by the replacement of ESSE-reflexes by HABERE-reflexes, whereas peripheral verb types are the most vulnerable to the change. A recent study by Cennamo (1995) suggests that the development of reflexives se/sibi in Late Latin as markers of split intransitivity followed a path largely consistent with the unaccusative/unergative hierarchies. Sorace and Cennamo (University of Naples) carried out a collaborative research project on unaccusativity in Italian dialects. The focus of this project is auxiliary selection and other typical manifestations of unaccusativity, such as impersonal passives, participial absolutes, ne-cliticization, and pleonastic reflexives. The data from the first study, on Paduan, provide support for the theory (see Sorace and Cennamo, forthcoming).
At the developmental level, data from the acquisition of Italian as a non-native language show that syntactic properties such as auxiliary selection and NE-cliticization are acquired earlier with core verbs and then gradually extended to more peripheral verb types (Sorace 1993a, 1995a). Moreover, Italian learners of French find it more difficult to acquire AVOIR as the auxiliary for verbs closer to the core than for peripheral verbs (Sorace 1993b, 1995b). These developmental regularities can be explained by assuming that the acquisition of the syntax of unaccusatives crucially depends on the internalization of the specific linking rules instantiated by the target language. Sorace's results are consistent with studies on the acquisition of auxiliary selection in Dutch and German as native languages (van Hout et al. 1993) also suggest that linking rules may be language-specific, and that the crucial semantic determinant of unaccusativity in Dutch is telicity, whereas it is transition in German.