Getting Serious about Word Sense Disambiguation

Getting Serious about Word Sense Disambiguation Hwee Tou Ng Defence Science Organisation 20 Science Park Drive Singapore 118230 Republic of Singapore nhweet ou@dso, gov. sg Abstract Recent advances in large-scale, broad cov- erage part-of-speech tagging and syntactic parsing have been achieved in no small part due to the availability of large amounts of online, human-annotated corpora. In this paper, I argue that a large, human sense- tagged corpus is also critical as well as ne- cessary to achieve broad coverage, high ac- curacy word sense disambiguation, where the sense distinction is at the level of a good desk-top dictionary such as WORD- NET. Using the sense-tagged corpus of 192,800 word occurrences reported in (Ng and Lee, 1996), I examine the effect of the number of training examples on the accur- acy of an exemplar-based classifier versus the base-line, most-frequent-sense classi- tier. I also estimate the amount of hu- man sense-tagged corpus and the manual annotation effort needed to build a large- scale, broad coverage word sense disambig- uation program which can significantly out- perform the most-frequent-sense classifier. Finally, I suggest that intelligent example selection techniques may significantly re- duce the amount of sense-tagged corpus needed and offer this research problem as a fruitful area for word sense disambiguation research. 1 Introduction Much recent research in the field of natural lan- guage processing (NLP) has focused on an empirical, corpus-based approach (Church and Mercer, 1993). The high accuracy achieved by a corpus-based ap- proach to part-of-speech tagging and noun phrase parsing, as demonstrated by (Church, 1988), has in- spired similar approaches to other problems in nat- ural language processing, including syntactic parsing and word sense disambiguation (WSD). The availability of large quantities of part-of- speech tagged and syntactically parsed sentences like the Penn Treebank corpus (Marcus, Santorini, and Marcinkiewicz, 1993) has contributed greatly to the development of robust, broad coverage part- of-speech taggers and syntactic parsers. The Penn Treebank corpus contains a sufficient number of part- of-speech tagged and syntactically parsed sentences to serve as adequate training material for building broad coverage part-of-speech taggers and parsers. Unfortunately, an analogous sense-tagged corpus large enough to achieve broad coverage, high accur- acy word sense disambiguation is not available at present. In this paper, I argue that, given the cur- rent state-of-the-art capability of automated machine learning algorithms, a supervised learning approach using a large sense-tagged corpus is a viable way to build a robust, wide coverage, and high accuracy WSD program. In this view, a large sense-tagged corpus is critical as well as necessary to achieve broad coverage, high accuracy WSD. The rest of this paper is organized as follows. In Section 2, I briefly discuss the utility of WSD in practical NLP tasks like information retrieval and machine translation. I also address some objections to WSD research. In Section 3, I examine the size of the training corpus on the accuracy of WSD, us- ing a corpus of 192,800 occurrences of 191 words hand tagged with WORDNET senses (Ng and Lee, 1996). …

Content truncated to 500 words