SenseLearner: Minimally supervised Word Sense Disambiguation for all words in open text

SenseLearner: Minimally Supervised Word Sense Disambiguation for All Words in Open Text Rada Mihalcea and Ehsanul Faruque Department of Computer Science University of North Texas � rada,faruque � @cs.unt.edu Abstract This paper introduces SENSELEARNER – a mini- mally supervised sense tagger that attempts to dis- ambiguate all content words in a text using the senses from WordNet. SENSELEARNER partici- pated in the SENSEVAL-3 English all words task, and achieved an average accuracy of 64.6%. 1 Introduction The task of word sense disambiguation consists of assigning the most appropriate meaning to a polysemous word within a given context. Appli- cations such as machine translation, knowledge acquisition, common sense reasoning, and oth- ers, require knowledge about word meanings, and word sense disambiguation is considered essential for all these applications. Most of the efforts in solving this problem were concentrated so far toward targeted super- vised learning, where each sense tagged occur- rence of a particular word is transformed into a feature vector, which is then used in an automatic learning process. The applicability of such super- vised algorithms is however limited only to those few words for which sense tagged data is avail- able, and their accuracy is strongly connected to the amount of labeled data available at hand. Instead, methods that address all words in open-text have received significantly less atten- tion. While the performance of such methods is usually exceeded by their supervised corpus-based alternatives, they have however the advantage of providing larger coverage. In this paper, we introduce a new method for solving the semantic ambiguity of all content words in a text. The algorithm can be thought of as a minimally supervised WSD algorithm in that it uses a small data set for training purposes, and generalizes the concepts learned from the training data to disambiguate the words in the test data set. As a result, the algorithm does not need a sepa- rate classifier for each word to be disambiguated. Moreover, it does not requires thousands of occur- rences of the same word to be able to disambiguate the word; in fact, it can successfully disambiguate a content word even if it did not appear in the train- ing data. 2 Background For some natural language processing tasks, such as part of speech tagging or named entity recogni- tion, regardless of the approach considered, there is a consensus on what makes a successful algo- rithm (Resnik and Yarowsky, 1997). Instead, no such consensus has been reached yet for the task of word sense disambiguation, and previous work has considered a range of knowledge sources, such as local collocational clues, membership in a se- mantically or topically related word class, seman- tic density, etc. Other related work has been mo- tivated by the intuition that syntactic information in a sentence contains enough information to be able to infer the semantics of words. For example, according to (Gomez, 2001), the syntax of many verbs is determined by their semantics, and thus it is possible to get the later from …

Content truncated to 500 words