Knowledge sources for disambiguating highly ambiguous verbs in machine translation

摘要: Word sense disambiguation (WSD) is one of the most challenging outstanding problems in the current machine translation systems. An effective proposal in this context will rely on the use relevant knowledge sources. Moreover, it must perform better than the current traditional approaches. We present some experiments with machine learning algorithms traditionally applied to WSD, aiming to discover both the best knowledge sources and the performance of these approaches. The results confirmed those already reported in monolingual WSD, indicating collocations and semantic word associations as the best word sense distinctive characteristics. In future work, we will use the best knowledge sources discovered, along with good rules produced by a symbolic algorithm, in a new WSD approach.