Combining Orthogonal Monolingual and Multilingual Sources of Evidence for All Words WSD

Proceedings of the 48th Annual Meeting of the Association for Computational Linguistics, pages 1542–1551, Uppsala, Sweden, 11-16 July 2010. c⃝2010 Association for Computational Linguistics Combining Orthogonal Monolingual and Multilingual Sources of Evidence for All Words WSD Weiwei Guo Computer Science Department Columbia University New York, NY, 10115 weiwei@cs.columbia.edu Mona Diab Center for Computational Learning Systems Columbia University New York, NY, 10115 mdiab@ccls.columbia.edu Abstract Word Sense Disambiguation remains one of the most complex problems facing com- putational linguists to date. In this pa- per we present a system that combines evidence from a monolingual WSD sys- tem together with that from a multilingual WSD system to yield state of the art per- formance on standard All-Words data sets. The monolingual system is based on a modification of the graph based state of the art algorithm In-Degree. The multilingual system is an improvement over an All- Words unsupervised approach, SALAAM. SALAAM exploits multilingual evidence as a means of disambiguation. In this paper, we present modifications to both of the original approaches and then their combination. We finally report the highest results obtained to date on the SENSEVAL 2 standard data set using an unsupervised method, we achieve an overall F measure of 64.58 using a voting scheme. 1 Introduction Despite advances in natural language processing (NLP), Word Sense Disambiguation (WSD) is still considered one of the most challenging problems in the field. Ever since the field’s inception, WSD has been perceived as one of the central problems in NLP. WSD is viewed as an enabling technology that could potentially have far reaching impact on NLP applications in general. We are starting to see the beginnings of a positive effect of WSD in NLP applications such as Machine Translation (Carpuat and Wu, 2007; Chan et al., 2007). Advances in WSD research in the current mil- lennium can be attributed to several key factors: the availability of large scale computational lexi- cal resources such as WordNets (Fellbaum, 1998; Miller, 1990), the availability of large scale cor- pora, the existence and dissemination of standard- ized data sets over the past 10 years through differ- ent testbeds such as SENSEVAL and SEMEVAL competitions,1 devising more robust computing algorithms to handle large scale data sets, and sim- ply advancement in hardware machinery. In this paper, we address the problem of WSD of all content words in a sentence, All-Words data. In this framework, the task is to associate all to- kens with their contextually relevant meaning defi- nitions from some computational lexical resource. Our work hinges upon combining two high qual- ity WSD systems that rely on essentially differ- ent sources of evidence. The two WSD systems are a monolingual system RelCont and a multi- lingual system TransCont. RelCont is an en- hancement on an existing graph based algorithm, In-Degree, first described in (Navigli and Lapata, 2007). TransCont is an enhancement over an existing approach that leverages multilingual evi- dence through projection, SALAAM, described in detail in (Diab and Resnik, 2002). Similar to the leveraged systems, the …

Content truncated to 500 words