Determining Word Sense Dominance Using a Thesaurus

Determining Word Sense Dominance Using a Thesaurus Saif Mohammad and Graeme Hirst Department of Computer Science University of Toronto Toronto, ON M5S 3G4, Canada fsmm,gh g@cs.toronto.edu Abstract The degree of dominance of a sense of a word is the proportion of occurrences of that sense in text. We propose four new methods to accurately determine word sense dominance using raw text and a pub- lished thesaurus. Unlike the McCarthy et al. (2004) system, these methods can be used on relatively small target texts, without the need for a similarly-sense- distributed auxiliary text. We perform an extensive evaluation using artificially gen- erated thesaurus-sense-tagged data. In the process, we create a word–category co- occurrence matrix, which can be used for unsupervised word sense disambiguation and estimating distributional similarity of word senses, as well. 1 Introduction The occurrences of the senses of a word usually have skewed distribution in text. Further, the dis- tribution varies in accordance with the domain or topic of discussion. For example, the ‘assertion of illegality’ sense of charge is more frequent in the judicial domain, while in the domain of eco- nomics, the ‘expense/cost’ sense occurs more of- ten. Formally, the degree of dominance of a par- ticular sense of a word (target word) in a given text (target text) may be defined as the ratio of the occurrences of the sense to the total occurrences of the target word. The sense with the highest domi- nance in the target text is called the predominant sense of the target word. Determination of word sense dominance has many uses. An unsupervised system will benefit by backing off to the predominant sense in case of insufficient evidence. The dominance values may be used as prior probabilities for the differ- ent senses, obviating the need for labeled train- ing data in a sense disambiguation task. Natural language systems can choose to ignore infrequent senses of words or consider only the most domi- nant senses (McCarthy et al., 2004). An unsuper- vised algorithm that discriminates instances into different usages can use word sense dominance to assign senses to the different clusters generated. Sense dominance may be determined by sim- ple counting in sense-tagged data. However, dom- inance varies with domain, and existing sense- tagged data is largely insufficient. McCarthy et al. (2004) automatically determine domain- specific predominant senses of words, where the domain may be specified in the form of an un- tagged target text or simply by name (for exam- ple, financial domain). The system (Figure 1) au- tomatically generates a thesaurus (Lin, 1998) us- ing a measure of distributional similarity and an untagged corpus. The target text is used for this purpose, provided it is large enough to learn a the- saurus from. Otherwise a large corpus with sense distribution similar to the target text (text pertain- ing to the specified domain) must be used. The thesaurus has an entry for each word type, which lists a limited number of words (neigh- bors) that are distributionally most similar to it. …

Content truncated to 500 words