Work in lexicography provides good insight into lexical semantics, in particular regarding word-sense distinctions. Kilgarriff [66] calls to task the notion of there being a distinct set of word senses independent of the application, as commonly assumed by word-sense disambiguation (WSD) researchers. Several lines of support are provided for this view. First, the WSD position is considered counter to much current work in linguistics (and theoretical computational linguistics) that address polysemy by exploiting systematic relationships among the senses. Second, there is no clear notion of what a word sense is, both in linguistics and in lexicography. What appears as discrete senses in dictionaries are as much the result of the exigencies of commercial publishing as the actual distinctness of the senses. Finally, there are likely to be many nonpredictable usages that don't make it into dictionaries due to relative infrequency compared to the other usages. So a given set of distinctions will not in general be suitable for the needs of particular applications.
Interest in Computational Lexicography, a formalization of lexicon-building inspired by standard lexicography, arose during the last decade, partly to provide a better foundation for the use of MRD's but also since it was felt that MRD analysis would never be sufficient to meet the needs of applications, as discussed by Atkins [6]. She presents several reasons for this point of view. The main gist is that dictionaries are inadequate because they were designed for human use and represent much filtering of the body of evidence about the language, such as the citations from which the lexicographers construct the entries. She feels that professional lexicographers must play an integral part in the development of future lexical databases, because dictionary compilation takes time to master. However, MRD's can still be used to populate the information stored there, provided that there is subsequent manual validation.