Slide 17-18

Ambiguity is a huge problem in IR. Traditional word sense disambiguation (WSD) research comes from machine translation and NLP and presupposes the existence of a rich textual context. Such contextual information is absent in IR and so new, innovative methods have to be used. [Kwo98] describes a method based on WordNet [Fel98]. [MS99] describes statistical WSD methods. Many IR scientists maintain that there is no need for complex WSD techniques as entering many terms in a query implicitly performs the disambiguation task.

Knowledge based techniques are popular in combating the problem of polysemy. Efforts are made to use existing resources, such as multilingual thesauri [FSE+97] and WordNet [SQ96]. In the past, many researchers tried to construct thesauri automatically using cluster analysis techniques [SM83], but these methods were not successful enough.