An Integrative, Layered Approach to Lexical Semantics and its Application to Machine Translation

Representation and Acquisition of Lexical Knowledge: Polysemy, Ambiguity, and Generativity

Volume

Representation and Acquisition of Lexical Knowledge: Polysemy, Ambiguity, and Generativity

Contents

One of the practical problems which stands in the way of the goal of developing a common lexicon for a language is the variation in granularities of meaning representation. These varying granularities have developed out of ecological needs. For example, in the field of machine translation (MT), researchers working on problems of translation divergences at the semantic level [Dorr 93a] [Dorr 93b], [Dorr and Voss 93] have tended to favor fairly abstract representations of meaning, such as Jackendoffian Conceptual Structures (CS) [Jackendoff 83], [Jackendoff 90]. Others (whether working in an "interlinguar' paradigm, e.g., [Carbonell and Tomita 87], [Carlson and Nirenburg 90], [Meyer et al. 90], [Barnett et al. 94], or semantic transfer paradigm, e.g., ['Nagao 87]) have depended on more concrete specifications of meaning, for example, based on microtheories of different domains, e.g., [Levin and Nirenburg 94]. The question then arises as to (1) whether these different granularities meaning representation can be related to each other. In addition, there is (2) the problem framework compatibility; in particular, between meaning representations more or less based on logic, with its emphasis on the semantics of reference and quantification, and the more conceptual structure-oriented approaches Oike Jackendoff-inspired approaches) which focus mainly on a rich ontology of primitives and structures. In general, the logic-based representations appear to be especially useful in application to "functional" (or closed-class) categories associated with modality, mood, tense, aspect, conjunctions, determiners, etc., while the Jackendoffian approaches are particularly useful in connection with "lexicar' (or open-class) categories.