Cross-Lingual Word Sense Disambiguation

Word Sense Disambiguation using Cross-Lingual approach has been used successfully for languages like Farsi and Hindi. However, a comparable corpus in the form of Wikipedia articles available in English and Hindi has been used for such a task. This motivated us to further the approach and test the results when a parallel corpus is used. In this project, we specifically wanted to observe if the accuracy would include if we could get a sentence to sentence mapping across translations and then disambiguating the sense using cross lingual approach. The use of a parallel corpus helped us in getting a sentence aligned data across translations. The results show that when we used individual sentences only for disambiguating, the results increased as compared to those when whole texts were consider in a parallel corpus. The result between local disambiguation (when individual sentences were used) and global disambiguation (when the text was used as a whole) also show that local disambiguation scores more on accuracy as compared to global disambiguation.