Projecting Parameters for Multilingual Word Sense Disambiguation

Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 459–467, Singapore, 6-7 August 2009. c⃝2009 ACL and AFNLP Projecting Parameters for Multilingual Word Sense Disambiguation Mitesh M. Khapra Sapan Shah Piyush Kedia Pushpak Bhattacharyya Department of Computer Science and Engineering Indian Institute of Technology, Bombay Powai, Mumbai – 400076, Maharashtra, India. {miteshk,sapan,charasi,pb}@cse.iitb.ac.in Abstract We report in this paper a way of doing Word Sense Disambiguation (WSD) that has its ori- gin in multilingual MT and that is cognizant of the fact that parallel corpora, wordnets and sense annotated corpora are scarce re- sources. With respect to these resources, lan- guages show different levels of readiness; however a more resource fortunate language can help a less resource fortunate language. Our WSD method can be applied to a lan- guage even when no sense tagged corpora for that language is available. This is achieved by projecting wordnet and corpus parameters from another language to the language in question. The approach is centered around a novel synset based multilingual dictionary and the empirical observation that within a domain the distribution of senses remains more or less invariant across languages. The effectiveness of our approach is verified by doing parameter projection and then running two different WSD algorithms. The accuracy values of ap- proximately 75% (F1-score) for three lan- guages in two different domains establish the fact that within a domain it is possible to cir- cumvent the problem of scarcity of resources by projecting parameters like sense distribu- tions, corpus-co-occurrences, conceptual dis- tance, etc. from one language to another. 1 Introduction Currently efforts are on in India to build large scale Machine Translation and Cross Lingual Search systems in consortia mode. These efforts are large, in the sense that 10-11 institutes and 6-7 languages spanning the length and breadth of the country are involved. The approach taken for translation is transfer based which needs to tackle the problem of word sense disambiguation (WSD) (Sergei et. al., 2003). Since 90s machine learning based ap- proaches to WSD using sense marked corpora have gained ground (Eneko Agirre & Philip Edmonds, 2007). However, the creation of sense marked cor- pora has always remained a costly proposition. Statistical MT has obviated the need for elaborate resources for WSD, because WSD in SMT hap- pens implicitly through parallel corpora (Brown et. al., 1993). But parallel corpora too are a very cost- ly resource. The above situation brings out the challenges involved in Indian language MT and CLIR. Lack of resources coupled with the multiplicity of Indian languages severely affects the performance of sev- eral NLP tasks. In the light of this, we focus on the problem of developing methodologies that reuse resources. The idea is to do the annotation work for one language and find ways of using them for another language. Our work on WSD takes place in a multilingual setting involving Hindi (national language of India; 500 million speaker base), Marathi (20 million speaker base), Bengali (185 million speaker base) and Tamil …

Content truncated to 500 words