Computational processing of verbal polysemy with conceptual structures

Abstract

Our work takes place in the general framework of lexicosemantic knowledge representation to be used by a Natural Language Understanding system. More specifically, we were interested in an adequate modelling of verb descriptions allowing to interpret semantic incoherence due to verbal polysemy. The main goal is to realise a module which is able to detect and to deal with figurative meanings. Therefore, we first propose a lexico-semantic knowledge base; then we present the processes allowing to determine the different meanings which may be associated to a given predicate, and to discriminate these meanings for a given sentence. Each verb is defined by a basic action (its supertype) specified by the case relations allowing to specify it (its definition graph), that means the object, mean, manner, goal and/or result relations which distinguish the described verb meaning from the specified basic action. This description is recursive: the basic actions are in turn defined using more general actions. To help interpreting the different meanings conveyed by a verb and its hyponyms, we have determined three major types of heuristics consisting in searching the type lattice, and/or examining the associated definitions.