Using Context for Sense Preference

Abstract

A system for preferring word senses according to context is described. The system uses a machine-readable dictionary (the Longman Dictionary of Contemporary English), that contains special subject coding fields, and keeps track of these codes as a lexicon is constructed for a text. By restructuring these codes into a pragmatic hierarchy, an effective structure for word sense preference is produced. This resulting structure has value for other systems of language analysis, as a source of global text knowledge, and (possibly) also as a reduced representation of text topic.