Sense and preference

AbstractSemantic networks have shown considerable utility as a knowledge representation for Natural Language Processing (NLP). This paper describes a system for automatically deriving network structures from machine-readable dictionary text. This strategy helps to solve the problem of vocabulary acquisition for large-scale parsing systems, but also introduces an extra level of difficulty in terms of word-sense ambiguity. A Preference Semantics parsing system that operates over this network is discussed, in particular as regards its mechanism for using the network for lexical selection