NRC: A Machine Translation Approach to Cross-Lingual Word Sense Disambiguation (SemEval-2013 Task 10)

Second Joint Conference on Lexical and Computational Semantics (*SEM), Volume 2: Seventh International Workshop on Semantic Evaluation (SemEval 2013), pages 188–192, Atlanta, Georgia, June 14-15, 2013. c⃝2013 Association for Computational Linguistics NRC: A Machine Translation Approach to Cross-Lingual Word Sense Disambiguation (SemEval-2013 Task 10) Marine Carpuat National Research Council Ottawa, Canada Marine.Carpuat@nrc.gc.ca Abstract This paper describes the NRC submission to the Spanish Cross-Lingual Word Sense Dis- ambiguation task at SemEval-2013. Since this word sense disambiguation task uses Spanish translations of English words as gold annota- tion, it can be cast as a machine translation problem. We therefore submitted the output of a standard phrase-based system as a baseline, and investigated ways to improve its sense dis- ambiguation performance. Using only local context information and no linguistic analy- sis beyond lemmatization, our machine trans- lation system surprisingly yields top precision score based on the best predictions. However, its top 5 predictions are weaker than those from other systems. 1 Introduction This paper describes the systems submitted by the National Research Council Canada (NRC) for the Cross-Lingual Word Sense Disambiguation task at SemEval 2013 (Lefever and Hoste, 2013). As in the previous edition (Lefever and Hoste, 2010), this word sense disambiguation task asks systems to dis- ambiguate English words by providing translations in other languages. It is therefore closely related to machine translation. Our work aims to explore this connection between machine translation and cross- lingual word sense disambiguation, by providing a machine translation baseline and investigating ways to improve the sense disambiguation performance of a standard machine translation system. Machine Translation (MT) has often been used indirectly for SemEval Word Sense Disambiguation (WSD) tasks: as a tool to automatically create train- ing data (Guo and Diab, 2010, for instance) ; as a source of parallel data that can be used to train WSD systems (Ng and Chan, 2007; van Gompel, 2010; Lefever et al., 2011); or as an application which can use the predictions of WSD systems de- veloped for SemEval tasks (Carpuat and Wu, 2005; Chan et al., 2007; Carpuat and Wu, 2007). This Se- mEval shared task gives us the opportunity to com- pare the performance of machine translation systems with other submissions which use very different ap- proaches. Our goal is to provide machine transla- tion output which is representative of state-of-the-art approaches, and provide a basis for comparing its strength and weaknesses with that of other systems submitted to this task. We submitted two systems to the Spanish Cross-Lingual WSD (CLWSD) task: 1. BASIC, a baseline machine translation system trained on the parallel corpus used to define the sense inventory; 2. ADAPT, a machine translation system that has been adapted to perform better on this task. After describing these systems in Sections 2 and 3, we give an overview of the results in Section 4. 2 BASIC: A Baseline Phrase-Based Machine Translation System We use a phrase-based SMT (PBSMT) architec- ture, and set-up our system to perform English-to- Spanish translation. We use a standard SMT system set-up, …

Content truncated to 500 words