1 Bhuvana Narasimhan Vittorio Di Tomaso Cornelia M.Verspoor * Unaccusative or Unergative? Verbs of Manner of Motion # (Paper submitted) The Unaccusative Hypothesis claims that intransitive verbs fall into two subclasses-- unaccusative verbs and unergative verbs, each associated with a particular underlying syntactic structure (Perlmutter 1978). This hypothesis is based on a number of linguistic phenomena (called unaccusative diagnostics) that differentiate between the two subclasses of verbs (Levin & Rappaport-Hovav 1995:3). Thus, for instance, unaccusatives select the auxiliary verb essere in Italian, whereas unergatives select avere (Burzio 1986; Perlmutter 1989). This difference in the behaviour of the two subclasses of intransitive verbs has been linked to differences in their underlying syntactic structure (Perlmutter 1978). Within the GB approach, unaccusatives are posited to have a direct object (or direct internal argument) and no subject, while unergatives have a subject (or external argument) but no direct object at d-structure (Levin & Rappaport- Hovav 1995:3). The unaccusative-unergative dichotomy has also been linked to differences in the semantic properties of these verbs and the constructions in which they appear. The unaccusative behaviour of a subclass of intransitive verbs with respect to a number of diagnostics can be linked to factors such as agentivity or telicity, without positing a direct object at some level of syntactic representation (Van Valin 1990; Dowty 1991; Wechsler 1996). A third approach is proposed by Levin & Rappaport-Hovav (1995) who claim that, although the difference between unaccusative and unergative verbs is syntactically encoded in terms of d-structure configurations, it is also semantically determined (as originally assumed by the Unaccusativity Hypothesis as well). The class of agentive manner of motion verbs poses an interesting problem with respect to the Unaccusativity Hypothesis. This class of verbs exhibits properties of both unaccusative and unergative verbs. Any approach that posits a syntactic difference between unaccusatives and unergatives would have to assume that each verb in this class has two argument structures, one of which projects the d-structure configuration corresponding to unaccusatives while the other projects the unergative d-structure configuration. Levin & Rappaport-Hovav (1995) propose such an analysis, and formulate a lexical argument-structure changing rule which derives unaccusatives from unergative manner of motion verbs. The semantic approach, on the other hand, does not posit two argument structures since unaccusative behaviour is not linked to a particular type of syntactic encoding. Rather, the variable behaviour of this class of verbs is explained in terms of differences in the interaction of the lexical semantic properties of the verbs with the semantics of the constructions in which they occur. In this paper, we examine the arguments provided by Levin & Rappaport-Hovav in favour of a lexical rule for this class of verbs and conclude that these arguments are not well- supported empirically. While we do not attempt to elaborate an alternative semantic account (see Wechsler 1996 for such an account), our observations suggest that an account relying on semantic factors alone can provide a simpler and more adequate explanation of the unaccusative behaviour of agentive manner of motion verbs. The class of agentive verbs of manner of motion such as walk, run, hobble etc. are unergative with respect to a number of diagnostics when used in isolation (1). They occur in * Current addresses of the authors are: Bhuvana Narasimhan, Boston University, bhuvana@acs.bu.edu; Vittorio Di Tomaso, Scuola Normale Superiore, ditomaso@sns.it; Cornelia M.Verspoor, Edimburgh, kversp@cogsci.ed.ac.uk. # We are grateful to Ray Jackendoff for extensive discussions, and Joan Maling for comments and ideas offered during the seminar on Unaccusativity at Brandeis University. The views expressed here are our own, as are any errors.