All Words Domain Adapted WSD: Finding a Middle Ground between Supervision and Unsupervision

Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, pages 1532–1541, Uppsala, Sweden, 11-16 July 2010. c⃝2010 Association for Computational Linguistics All Words Domain Adapted WSD: Finding a Middle Ground between Supervision and Unsupervision Mitesh M. Khapra Anup Kulkarni Saurabh Sohoney Indian Institute of Technology Bombay,Pushpak Bhattacharyya Mumbai - 400076, India. {miteshk,anup,saurabhsohoney,pb}@cse.iitb.ac.in Abstract In spite of decades of research on word sense disambiguation (WSD), all-words general purpose WSD has remained a dis- tant goal. Many supervised WSD systems have been built, but the effort of creat- ing the training corpus - annotated sense marked corpora - has always been a matter of concern. Therefore, attempts have been made to develop unsupervised and knowl- edge based techniques for WSD which do not need sense marked corpora. However such approaches have not proved effective, since they typically do not better Word- net first sense baseline accuracy. Our re- search reported here proposes to stick to the supervised approach, but with far less demand on annotation. We show that if we have ANY sense marked corpora, be it from mixed domain or a specific domain, a small amount of annotation in ANY other domain can deliver the goods almost as if exhaustive sense marking were avail- able in that domain. We have tested our approach across Tourism and Health do- main corpora, using also the well known mixed domain SemCor corpus. Accuracy figures close to self domain training lend credence to the viability of our approach. Our contribution thus lies in finding a con- venient middle ground between pure su- pervised and pure unsupervised WSD. Fi- nally, our approach is not restricted to any specific set of target words, a departure from a commonly observed practice in do- main specific WSD. 1 Introduction Amongst annotation tasks, sense marking surely takes the cake, demanding as it does high level of language competence, topic comprehension and domain sensitivity. This makes supervised ap- proaches to WSD a difficult proposition (Agirre et al., 2009b; Agirre et al., 2009a; McCarthy et al., 2007). Unsupervised and knowledge based ap- proaches have been tried with the hope of creating WSD systems with no need for sense marked cor- pora (Koeling et al., 2005; McCarthy et al., 2007; Agirre et al., 2009b). However, the accuracy fig- ures of such systems are low. Our work here is motivated by the desire to de- velop annotation-lean all-words domain adapted techniques for supervised WSD. It is a common observation that domain specific WSD exhibits high level of accuracy even for the all-words sce- nario (Khapra et al., 2010) - provided training and testing are on the same domain. Also domain adaptation - in which training happens in one do- main and testing in another - often is able to attain good levels of performance, albeit on a specific set of target words (Chan and Ng, 2007; Agirre and de Lacalle, 2009). To the best of our knowledge there does not exist a system that solves the com- bined problem of all …

Content truncated to 500 words