Context sensitive reasoning with lexical and world knowledge

1. Introduction The question of polysemy and the representation of different lexical word-senses has recently attracted an increased interest in the linguistics and computational linguistics literature. has proposed to enrich the format of lexical representation so that a number of 'novel' word senses can be generated. In this paper, we discuss GLT and show that, while being able to derive novel word-senses, there are, however, problems with this approach. In particular, we argue that GLT lacks the expressivity necessary for supporting context sensitive reasoning, and show this by considering relevant examples involving underspecification inherent in verbs and the process of establishing semantic predicate-argument structures in utterance interpretation. The theoretical background of our work is the notion of utterance interpretation as where it is argued that verbs are lexically underspecified as to their semantic type, so that a verb's eventual arity will only be established in the utterance context. From this perspective, the interplay between this structural underspecification, lexical polysemy and context sensitive interpretation leads to an alternative view on knowledge representation employed in utterance interpretation and parsing, namely that hearers employ world knowledge, rather than 'lexical' knowledge, in the task to derive the proposition expressed, and that enrichment of meaning involves a level of conceptual representational structure. In this paper, we are not discussing the theoretical aspects of this view in detail (cf. but rather we propose a formalization of enrichment of natural language expressions as used in parsing by means of default logic statements (Reiter 1980). We set out by discussing GLT in some detail, and then sketch our formalization of the basic architecture for relating parsed natural language structures and world knowledge.