Semi-Supervised Word Sense Disambiguation Using Word Embeddings in General and Specific Domains

Human Language Technologies: The 2015 Annual Conference of the North American Chapter of the ACL, pages 314–323, Denver, Colorado, May 31 – June 5, 2015. c⃝2015 Association for Computational Linguistics Semi-Supervised Word Sense Disambiguation Using Word Embeddings in General and Specific Domains Kaveh Taghipour Department of Computer Science National University of Singapore 13 Computing Drive Singapore 117417 kaveh@comp.nus.edu.sg Hwee Tou Ng Department of Computer Science National University of Singapore 13 Computing Drive Singapore 117417 nght@comp.nus.edu.sg Abstract One of the weaknesses of current supervised word sense disambiguation (WSD) systems is that they only treat a word as a discrete en- tity. However, a continuous-space represen- tation of words (word embeddings) can pro- vide valuable information and thus improve generalization accuracy. Since word embed- dings are typically obtained from unlabeled data using unsupervised methods, this method can be seen as a semi-supervised word sense disambiguation approach. This paper investi- gates two ways of incorporating word embed- dings in a word sense disambiguation setting and evaluates these two methods on some Sen- sEval/SemEval lexical sample and all-words tasks and also a domain-specific lexical sam- ple task. The obtained results show that such representations consistently improve the ac- curacy of the selected supervised WSD sys- tem. Moreover, our experiments on a domain- specific dataset show that our supervised base- line system beats the best knowledge-based systems by a large margin. 1 Introduction Because of the ambiguity of natural language, many words can have different meanings in different con- texts. For example, the word “bank” has two differ- ent meanings in “the bank of a river” and “a bank loan”. While it seems simple for humans to iden- tify the meaning of a word according to the con- text, word sense disambiguation (WSD) (Ng and Lee, 1996; Lee and Ng, 2002) is a difficult task for computers and thus requires sophisticated means to achieve its goal. Part of this ambiguity may be resolved by considering part-of-speech (POS) tags but the word senses are still highly ambiguous even for the same part-of-speech. Machine translation is probably the most important application of word sense disambiguation. In machine translation, dif- ferent senses of a word cause a great amount of am- biguity for automated translation and it negatively affects the results. Hence, an accurate WSD system can benefit machine translation significantly and im- prove the results (Chan et al., 2007; Carpuat and Wu, 2007; Vickrey et al., 2005). Moreover, Zhong and Ng (2012) have shown that word sense disambigua- tion improves information retrieval by proposing a method to use word senses in a language modeling approach to information retrieval. The rest of this paper is organized as follows. Section 2 gives a literature review of related work, including a review of semi-supervised word sense disambiguation and distributed word representation called word embeddings. The method and frame- work used in this paper are explained in Section 3. Finally, we evaluate the system in Section 4 and con- clude the paper in Section 5. 2 Related Work The method that …

Content truncated to 500 words