NLP needs to operate with the meanings of words. Where a word has more than one meaning an NLP system will need to select. If a system is to use a lexicon based on a published dictionary, the number of words with multiple meanings means that writing disambiguation strategies on a word-by-word basis will be a monumental labour. If the domain of word sense distinctions were better understood, more efficient ways of writing disambiguation strategies and operating with multiple senses could be developed. This study presents a classification scheme for distinctions between word senses, based on an empirical study of a sample of the entries in one dictionary. The dictionary used a number of devices to indicate alternative uses of a word, and these are described. Factors which determine what types of usage get treated as distinct word senses, such as frequency of occurrence and the predictability of the alternation pattern, are also discussed.
This paper is not available online