Traditionally, vector-based semantic space models use word co-occurrence counts from large corpora to represent lexical meaning. In this article we present a novel framework for constructing semantic spaces that take syntactic relations into account. We introduce a formalization for this class of models which allows linguistic knowledge to guide the construction process. We evaluate our framework on a range of tasks relevant for cognitive science and natural language processing: semantic priming, synonymy detection and word sense disambiguation. In all cases, our framework obtains results that are comparable or superior to the state of the art. 1.

Abstract not found

Levin’s (1993) study of verb classes is a widely used resource for lexical semantics. In her framework, some verbs, such as give, exhibit no class ambiguity. But other verbs, such as write, have several alternative classes. We extend Levin’s inventory to a simple statistical model of verb class ambiguity. Using this model we are able to generate preferences for ambiguous verbs without the use of a disambiguated corpus. We additionally show that these preferences are useful as priors for a verb sense disambiguator.

In this paper, we propose an account of linking patterns that does away with intermediary mechanisms such as thematic or actor/undergoer hierarchies. Instead, constraints on word classes, defined by both syntactic and semantic criteria, encode generalizations between semantic roles and syntactic arguments. We show that the generalizations a linking theory needs to capture can be modeled via the same mechanisms as other lexical generalizations, using conditions specified within the hierarchy of word classes. Each condition provides a partial specification of the mapping between semantic roles and syntactic arguments. We argue that this constraint-based, verb-class-based view of linking offers several empirical advantages: partial regularities and exceptions are easily accomodated, fine-grained semantic distinctions relevant to linking are countenanced, and cross-cutting similarities between semantic and syntactic verb classes are economically captured.

Abstract not found

We challenge the predominant view of the English dative alternation, which takes all alternating verbs to have two meanings: a caused possession meaning realized by the double object variant and a caused motion meaning realized by the to variant. Instead, we argue that verbs like give and sell only have a caused possession meaning, while verbs like throw and send have both caused motion and caused possession meanings. We show that the caused possession meaning may be realized by both variants. Concomitantly, we argue that verbs like give, even in the to variant, lack a conceptual path constituent, and instead have a caused possession meaning which can be understood as the bringing about of a ‘have ’ relation. We reassess evidence for alternative approaches adduced from inference patterns and verb–argument combinations and demonstrate how our verb-sensitive analysis, when combined with an account of variant choice, provides a more insightful explanation of this data, while having wider coverage. Our investigation affirms proposals that a verb’s own meaning plays a key role in determining its argument realization options. To conclude, we consider the crosslinguistic implications of our study, attempting to explain why so

lexical rule example.

The present study addresses the gradience of the dative alternation. It is shown that central evidential paradigms that have been used to support semantic explanations for the choice of dative constructions are not well founded empirically. Some widely repeated reports of intuitive contrasts in grammaticality appear to rest instead on judgments of pragmatic probabilities. An informational theory of the dative alternation is supported by the results of a corpus study on the distribution of person across dative NP and PP recipients in spoken English, and a formal model of the theory is given within the framework of stochastic Optimality Theory.

In this work, we develop and evaluate a wide range of feature spaces for deriving Levinstyle verb classifications (Levin, 1993). We perform the classification experiments using

can be used to describe events of transfer, show two options for expressing their arguments, jointly referred to as the dative alternation, illustrated in (1)-(3) with English data. (1) a. Terry gave Sam an apple b. Terry gave an apple to Sam (2) a. Martha sent Myrna a package b. Martha sent a package to Myrna (3) a. Leigh threw Lane the ball b. Leigh threw the ball to Lane In a recent paper, M. Rappaport Hovav and B. Levin (2008) challenge the predominant view of the English dative alternation, which takes all alternating verbs to have two meanings and, concomitantly, associates each meaning with a particular syntactic realization (e.g. S. Beck and K. Johnson 2004, G. Green 1974,