Abstract
This book is about word sense disambiguation, the process of figuring out word meanings in a discourse which is an essential task in natural language processing. Computational linguists’ efforts over several decades have led to an apparently plateaued performance in state-of-the-art systems, but considerable unknowns regarding the lexical sensitivity of the task still remain. We propose to address this issue through a better synergy between the computational and cognitive paradigms, which had once closely supported and mutually advanced each other. We start off with an introduction to the word sense disambiguation problem and the notion of word senses in this chapter. While the psychological reality of word senses is beyond doubt, the boundaries between senses could be fuzzy. We discuss various models for representing senses and suggest that the discreteness assumption held by most mainstream systems is relevant to the perception of word senses rather than their definition.
This is a preview of subscription content, log in via an institution to check access.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
Stevenson (2003) distinguishes sense tagging from sense disambiguation, where the former attempts to disambiguate all words in a text but the latter only does so for a restricted set of words. In both cases, the annotations applied are senses from some lexicon.
- 2.
SENSEVAL and SEMEVAL provide a common platform for evaluating WSD systems, where participating systems submit their results on the same test data and are evaluated against the same gold standard. WSD evaluation will be further discussed in Chap. 3.
- 3.
Homonymy and polysemy may occur within or across syntactic word classes, and for the latter they are more often known as categorial ambiguities. Since POS tagging would have taken care of most categorial ambiguities, our main concern for WSD will be the ambiguities within the same POS.
- 4.
See for example Evens (1988) for a fuller account of relational models.
- 5.
e.g. logical metonymy (a fast car, a fast typist, a fast motorway, etc.), broadening (e.g. a cloud of something), nominal metonymy (e.g. the ham sandwich), portioning (e.g. three beers = three portions of beer), grinding (e.g. lamb as the animal or the meat), and so on.
- 6.
The “Bank Model” suggests that most words are like the word “bank” which has perfectly distinct senses. It predicts that each occurrence of “bank” would refer to one sense (a money bank) or the other (a river bank), which should be instant and effortless for English speakers to recognise. Kilgarriff (1993) tried to match each occurrence of his test words in the Lancaster Oslo-Bergen (LOB) corpus to one of their senses listed in LDOCE, and found that 87% (60 out of 69) of the test words had at least one usage which could not be confidently assigned only one sense, with which he concluded that word senses are by no means discrete. Wilks (1998) pointed out that the 87%-claim has no contradiction to saying that the majority of the text usage could be associated with only one dictionary sense.
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
Copyright information
© 2013 The Author(s)
About this chapter
Cite this chapter
Kwong, O.Y. (2013). Word Senses and Problem Definition. In: New Perspectives on Computational and Cognitive Strategies for Word Sense Disambiguation. SpringerBriefs in Electrical and Computer Engineering(). Springer, New York, NY. https://doi.org/10.1007/978-1-4614-1320-2_1
Download citation
DOI: https://doi.org/10.1007/978-1-4614-1320-2_1
Published:
Publisher Name: Springer, New York, NY
Print ISBN: 978-1-4614-1319-6
Online ISBN: 978-1-4614-1320-2
eBook Packages: EngineeringEngineering (R0)