COLEPL and COLSLM: An Unsupervised WSD Approach to Multilingual Lexical Substitution, Tasks 2 and 3 SemEval 2010

Proceedings of the 5th International Workshop on Semantic Evaluation, ACL 2010, pages 129–133, Uppsala, Sweden, 15-16 July 2010. c⃝2010 Association for Computational Linguistics COLEUR and COLSLM: A WSD approach to Multilingual Lexical Substitution, Tasks 2 and 3 SemEval 2010 Weiwei Guo and Mona Diab Center for Computational Learning Systems Columbia University {weiwei,mdiab}@ccls.columbia.edu Abstract In this paper, we present a word sense disambiguation (WSD) based system for multilingual lexical substitution. Our method depends on having a WSD system for English and an automatic word align- ment method. Crucially the approach re- lies on having parallel corpora. For Task 2 (Sinha et al., 2009) we apply a super- vised WSD system to derive the English word senses. For Task 3 (Lefever & Hoste, 2009), we apply an unsupervised approach to the training and test data. Both of our systems that participated in Task 2 achieve a decent ranking among the participating systems. For Task 3 we achieve the highest ranking on several of the language pairs: French, German and Italian. 1 Introduction In this paper, we present our system that was ap- plied to the cross lingual substitution for two tasks in SEMEVAL 2010, Tasks 2 and 3. We adopt the same approach for both tasks with some dif- ferences in the basic set-up. Our basic approach relies on applying a word sense disambiguation (WSD) system to the English data that comes from a parallel corpus for English and a language of relevance to the task, language 2 (l2). Then we automatically induce the English word sense cor- respondences to l2. Accordingly, for a given test target word, we return its equivalent l2 words as- suming that we are able to disambiguate the target word in context. 2 Our Detailed Approach We approach the problem of multilingual lexical substitution from a WSD perspective. We adopt the hypothesis that the different word senses of ambiguous words in one language probably trans- late to different lexical items in another language. Hence, our approach relies on two crucial compo- nents: a WSD module for the source language (our target test words, in our case these are the English target test words) and an automatic word align- ment module to discover the target word sense cor- respondences with the foreign words in a second language. Our approach to both tasks is unsuper- vised since we don’t have real training data anno- tated with the target words and their corresponding translations into l2 at the onset of the problem. Accordingly, at training time, we rely on auto- matically tagging large amounts of English data (target word instances) with their relevant senses and finding their l2 correspondences based on au- tomatically induced word alignments. Each of these English sense and l2 correspondence pairs has an associated translation probability value de- pending on frequency of co-occurrence. This in- formation is aggregated in a look-up table over the entire training set. An entry in the table would have a target word sense type paired with all the observed translation correspondences …

Content truncated to 500 words