Translation Context Sensitive WSD

Translation Context Sensitive WSD Lucia Specia♦♦♦♦, Maria das Graças Volpe Nunes♦♦♦♦, and Mark Stevenson♣♣♣♣ ♦ ICMC – University of São Paulo, Caixa Postal 668, São Carlos, 13560-970, SP, Brazil ♣ DCS – University of Sheffield, Regent Court, 211 Portobello Street, Sheffield, S1 4DP, UK {lspecia,gracan}@icmc.usp.br, M.Stevenson@dcs.shef.ac.uk Abstract While it is generally agreed that Word Sense Disambiguation (WSD) is an application-dependent task, the great majority of systems pursue application-independent approaches. We propose a strategy to support WSD for Machine Translation which is designed specifically for this application. It relies on the analysis of co-occurrences in the context that refer to words which have already been translated. Experiments on the English-Portuguese translation of 10 verbs using just this knowledge yielded an accuracy of 51%, which outperforms the baseline using the most frequent translation (37%). A less strict evaluation criterion considering the 10 best ranked translations proved the potential for this approach to be used as extra knowledge source for WSD: the correct translation was among the top 10 results in 92% of the cases. 1. Introduction Word Sense Disambiguation (WSD) in Machine Translation (MT) is concerned with the choice of the most appropriate translation of an ambiguous word given its context. The question of whether WSD is useful for MT has recently been debated again. Vickrey et al. (2005), e.g., showed that a WSD module significantly improves the perform- ance of their statistical MT system. Conversely, Carpuat and Wu (2005) claim that WSD does not yield significantly better translation quality than a statistical MT system alone. In this latter work, however, the WSD module was not specifically designed for MT. In fact, although it has been agreed that multi- lingual WSD differs from monolingual WSD (Hutchins and Somers, 1992) and that WSD is ultimately a task which is only relevant in the con- text of a specific application (Wilks and Steven- son, 1998), little has been done on the develop- ment of WSD modules for particular applications. WSD models in general are application independ- ent, and focus on monolingual contexts, particu- larly English. In the case of MT, the application we are dealing with, WSD approaches usually apply traditional sense repositories, such as the one provided by WordNet (Miller, 1990), to iden- tify the monolingual senses, which are then mapped into the target language translations. However, mapping senses between languages is a very complex issue. One of the reasons for this complexity is the dif- ference in the sense inventories of the languages, as already discussed in (Hutchins and Somers, 1992) and recently evidenced by studies with certain pairs of languages. For example, Bentivogli et al. (2004) investigate the sense inventory discrepancies for Eng- lish-Italian, Miháltz (2005), for English-Hungarian, Chatterjee et al. (2005), for English-Hindi, and Specia et al. (2006), for English-Portuguese. They show that there is not a one-to-one relation between the number of senses in the source language and their translations into another language. More specifically, they show that many source language senses are translated into a unique target language …

Content truncated to 500 words