Statistical Sense Disambiguation with Relatively Small Corpora Using Dictionary Definitions

Statistical Sense Disambiguation with Relatively Small Corpora Using Dictionary Definitions Microsoft Institute North Ryde, NSW 2113, Australia t-alphal@microsoft.com Alpha K. Luk Department of Computing Macquarie University NSW 2109, Australia Abstract Corpus-based sense disambiguation methods, like most other statistical NLP approaches, suffer from the problem of data sparseness. In this paper, we describe an approach which overcomes this problem using dictionary definitions. Using the definition- based conceptual co-occurrence data collected from the relatively small Brown corpus, our sense disambiguation system achieves an average accuracy comparable to human performance given the same contextual information. 1 Introduction Previous corpus-based sense disambiguation methods require substantial amounts of sense-tagged training data (Kelly and Stone, 1975; Black, 1988 and Hearst, 1991) or aligned bilingual corpora (Brown et al., 1991; Dagan, 1991 and Gale et al. 1992). Yarowsky (1992) introduces a thesaurus-based approach to statistical sense disambiguation which works on monolingual corpora without the need for sense-tagged training data. By collecting statistical data of word occurrences in the context of different thesaurus categories from a relatively large corpus (10 million words), the system can identify salient words for each category. Using these salient words, the system is able to disambiguate polysemous words with respect to thesaurus categories. Statistical approaches like these generally suffer from the problem of data sparseness. To estimate the salience of a word with reasonable accuracy, the system needs the word to have a significant number of occurrences in the corpus. Having large corpora will help but some words are simply too infrequent to make a significant statistical contribution even in a rather large corpus. Moreover, huge corpora are not generally available in all domains and storage and processing of very huge corpora can be problematic in some cases.Z In this paper, we describe an approach which attacks the problem of. data sparseness in automatic statistical sense disambiguation. Using definitions from LDOCE (Longman Dictionary of Contemporary English; Procter, 1978), co- occurrence data of concepts, rather than words, is collected from a relatively small corpus, the one million word Brown corpus. Since all the definitions in LDOCE are written using words from the 2000 word controlled vocabulary (or in our terminology, defining concepts), even our small corpus is found to be capable of providing statistically significant co- occurrence data at the level of the defining concepts. This data is then used in a sense disambiguation system. The system is tested on twelve words previously discussed in the sense disambiguation literature. The results are found to be comparable to human performance given the same contextual information. 2 Statistical Sense Disambiguation Using Dictionary Definitions It is well known that some words tend to co-occur with some words more often than with others. Similarly, looking at the meaning of the words, one should find that some concepts co-occur more often with some concepts than with others. For example, the concept crime is found to co-occur frequently with the concept punishment. This kind of conceptual relationship is not always reflected at the lexical level. For instance, in legal reports, …

Content truncated to 500 words