A Hybrid Distributional and Knowledge-based Model of Lexical Semantics

Proceedings of the Fourth Joint Conference on Lexical and Computational Semantics (*SEM 2015), pages 20–29, Denver, Colorado, June 4–5, 2015. A Hybrid Distributional and Knowledge-based Model of Lexical Semantics Nikolaos Aletras Department of Computer Science, University College London, Gower Street, London WC1E 6BT United Kingdom nikos.aletras@gmail.com Mark Stevenson Department of Computer Science, University of Sheffield, Regent Court, 211 Portobello, Sheffield S1 4DP United Kingdom mark.stevenson@sheffield.ac.uk Abstract A range of approaches to the representa- tion of lexical semantics have been explored within Computational Linguistics. Two of the most popular are distributional and knowledge- based models. This paper proposes hybrid models of lexical semantics that combine the advantages of these two approaches. Our mod- els provide robust representations of synony- mous words derived from WordNet. We also make use of WordNet’s hierarcy to refine the synset vectors. The models are evaluated on two widely explored tasks involving lexical semantics: lexical similarity and Word Sense Disambiguation. The hybrid models are found to perform better than standard distributional models and have the additional benefit of mod- elling polysemy. 1 Introduction The representation of lexical semantics is a core prob- lem in Computational Linguistics and a variety of approaches have been developed. Two of the most widely explored have been knowledge-based and dis- tributional semantics. Knowledge-based approaches make use of some external information source which defines the set of possible meanings for each lexical item. The most widely used information source is WordNet (Fell- baum, 1998), although other resources, such as Ma- chine Readable Dictionaries, thesaurii and ontologies have also been used (see Navigli (2009)). One advantage of these resources is that they rep- resent the various possible meanings of lexical items which makes it straightforward to identify ones that are ambiguous. For example, these resources would include multiple meanings for the word ball includ- ing the ‘event’ and ‘sports equipment’ senses. How- ever, the fact that there are multiple meanings as- sociated with ambiguous lexical items can also be problematic since it may not be straightforward to identify which one is being used for an instance of an ambiguous word in text. This issue has lead to signif- icant exploration of the problem of Word Sense Dis- ambiguation (Ide and V´eronis, 1998; Navigli, 2009). More recently distributional semantics has become a popular approach to representing lexical semantics (Turney and Pantel, 2010; Erk, 2012). These ap- proaches are based on the premise that the semantics of lexical items can be modelled by their context (Firth, 1957; Harris, 1985). Distributional seman- tic models have the advantages of being robust and straightforward to create from unannotated corpora. However, problems can arise when they are used to represent the semantics of polysemous words. Distri- butional semantic models are generally constructed by examining the context of lexical items in unanno- tated corpora. But for ambiguous words, like ball, it is not clear if a particular instance of the word in a corpus refers to the ‘event’, ‘sports equipment’ or another sense which can lead to the distributional se- mantic …

Content truncated to 500 words