Exploiting Parallel Texts for Word Sense Disambiguation: An Empirical Study

Exploiting Parallel Texts for Word Sense Disambiguation: An Empirical Study Hwee Tou Ng Bin Wang Yee Seng Chan Department of Computer Science National University of Singapore 3 Science Drive 2, Singapore 117543 {nght, wangbin, chanys}@comp.nus.edu.sg Abstract A central problem of word sense disam- biguation (WSD) is the lack of manually sense-tagged data required for supervised learning. In this paper, we evaluate an ap- proach to automatically acquire sense- tagged training data from English-Chinese parallel corpora, which are then used for disambiguating the nouns in the SENSEVAL-2 English lexical sample task. Our investigation reveals that this method of acquiring sense-tagged data is promising. On a subset of the most diffi- cult SENSEVAL-2 nouns, the accuracy difference between the two approaches is only 14.0%, and the difference could nar- row further to 6.5% if we disregard the advantage that manually sense-tagged data have in their sense coverage. Our analysis also highlights the importance of the issue of domain dependence in evalu- ating WSD programs. 1 Introduction The task of word sense disambiguation (WSD) is to determine the correct meaning, or sense of a word in context. It is a fundamental problem in natural language processing (NLP), and the ability to disambiguate word sense accurately is important for applications like machine translation, informa- tion retrieval, etc. Corpus-based, supervised machine learning methods have been used to tackle the WSD task, just like the other NLP tasks. Among the various approaches to WSD, the supervised learning ap- proach is the most successful to date. In this ap- proach, we first collect a corpus in which each occurrence of an ambiguous word w has been manually annotated with the correct sense, accord- ing to some existing sense inventory in a diction- ary. This annotated corpus then serves as the training material for a learning algorithm. After training, a model is automatically learned and it is used to assign the correct sense to any previously unseen occurrence of w in a new context. While the supervised learning approach has been successful, it has the drawback of requiring manually sense-tagged data. This problem is par- ticular severe for WSD, since sense-tagged data must be collected separately for each word in a language. One source to look for potential training data for WSD is parallel texts, as proposed by Resnik and Yarowsky (1997). Given a word-aligned paral- lel corpus, the different translations in a target lan- guage serve as the “sense-tags” of an ambiguous word in the source language. For example, some possible Chinese translations of the English noun channel are listed in Table 1. To illustrate, if the sense of an occurrence of the noun channel is “a path over which electrical signals can pass”, then this occurrence can be translated as “频道” in Chi- nese. WordNet 1.7 sense id Lumped sense id Chinese translations WordNet 1.7 English sense descriptions 1 1 频道 A path over which electrical signals can pass 2 2 水道 水渠 排水渠 A passage for water 3 3 沟 A long narrow furrow 4 …

Content truncated to 500 words