Developing an interlingual translation lexicon using WordNets and Grammatical Framework

Machine translation using the Grammatical Framework (GF) was originally developed for controlled languages with well-defined abstract semantics that acts as an interlingua. More recently, the coverage of GF resource grammars and its processing capabilities have reached to a level, where open-domain tasks such as arbitrary text parsing and translation look a possibility. But, we need several new advances, including robust parsing, parse- tree disambiguation, word sense disambiguation (WSD), and wide-coverage interlingual lexicons. In this paper, we address the later two issues. First, we build a wide coverage interlingual translation lexicon using the Princeton and Universal WordNet data. Then, we propose a technique to do WSD in GF, by integrating an existing WSD tool and replacing the usual GF style lexicons, which give one target word for each source word, by the reported WordNet based lexicons. The result is that, with the help of these new lexicons and WSD, the quality of translations improves in most of the cases, as we show by examples. Both WSD in general, and WordNets are of course well known, but this is the first time these tools have been coupled with GF.