This chapter discusses the basic concepts of Word Sense Disambiguation (WSD) and the approaches to solving this problem. Both general purpose WSD and domain specific WSD are presented. The first part of the discussion focuses on existing approaches for WSD, including knowledge-based, supervised, semi-supervised, unsupervised, hybrid, and bilingual approaches. The accuracy value for general purpose WSD as the current state of affairs seems to be pegged at around 65%. This has motivated investigations into domain specific WSD, which is the current trend in the field. In the latter part of the chapter, we present a greedy neural network inspired algorithm for domain specific WSD and compare its performance with other state-of-the-art algorithms for WSD. Our experiments suggest that for domain-specific WSD, simply selecting the most frequent sense of a word does as well as any state-of-the-art algorithm.
Top1. Introduction
Word Sense Disambiguation (WSD) is the problem of finding the correct sense (i.e., meaning) of a word by looking at the context in which it appears. It is one of the central challenges in NLP and is ubiquitous across all languages. Almost every language that we know has polysemy (poly means “many” and semy means “signs” or “meanings”) to a certain degree. For example, consider the two different meanings of the word bank in English:
I am going to the bank to withdraw money.
I am going to take a walk along the river bank.
In the first sentence, the word ‘bank’ refers to a “financial institution” whereas in the second sentence it refers to a “sloping land beside a water body (river, in this case).” When a human being reads the first sentence, he sees the words “withdraw” and “money” in the context and uses his world knowledge to decide that the word bank here refers to a “financial institution.” Similarly, he sees the word “river” in the second sentence and easily infers that the word bank here refers to a “sloping land near the river.” Identifying the correct meaning of a word can serve as a building block for many Natural Language Processing (NLP) tasks, such as Information Retrieval (IR), Machine Translation (MT), Information Extraction (IE), and more recently for Subjectivity and Sentiment Analysis. In IR, WSD can help in identifying the correct sense of a word in the query and thereby improve the precision of the results fetched (Harman, 2005). In MT, identifying the correct sense of a word in the source language can help in selecting its appropriate translation in the target language (Carpuat & Wu, 2007). Similarly, in IE, knowing the correct sense of every word in a document may help in doing an accurate analysis of the text. More recently, Balamurali et al. (2011) have shown that WSD can help in improving the performance of document level sentiment classifiers.
The above-mentioned applications of WSD suggest that distinguishing between different senses of a word is indeed important, but, how do we train a machine to acquire the necessary world knowledge required to perform such distinction or how do we even make a machine aware that a word has such multiple senses or meanings? The first question brings out the hardness of the problem and it is a commonly accepted notion that WSD is an AI-complete problem, i.e. it is as hard as any other AI problem (Navigli, 2009). In fact, several studies (Snyder & Palmer, 2004) have shown that WSD is a hard problem even for human beings. Specifically, these studies have shown that given the task of assigning senses to a large set of words by looking at their context, the agreement in the senses assigned by two humans is only 78%. Considering the difficulty of the task, its importance, and its ubiquitous nature, much work has been done in this area. In this chapter, we describe some of the popular algorithms, which have been proposed to perform WSD and highlight that in some specific conditions, such as when the corpus is restricted to a specific domain, it is possible to achieve near human performance on WSD.
The second question, i.e., “how do we make a machine aware of the different senses of a word” brings us to the concept of a sense repository or a knowledge base. A sense repository is a lexical resource which lists down the different senses of a word. The most popular sense repository used for WSD is WordNet (Fellbaum, 1998) which is a hierarchical lexical database where the basic unit of storage is a synset (short for synonymy set). As the name suggests, each synset contains a set of words, which together define a concept. From now on, we use the words synset and sense interchangeably. In addition to storing the gloss, examples and members for each synset, a wordnet also stores semantic relations between the synsets, e.g., hypernymy/hyponymy (IS-A), holonymy/meronymy (PART-OF), troponymy (TYPE-OF), etc. Below, we give examples of two synsets from the English wordnet along with their relations.