AI-KU: Using Substitute Vectors and Co-Occurrence Modeling For Word Sense Induction and Disambiguation

Second Joint Conference on Lexical and Computational Semantics (*SEM), Volume 2: Seventh International Workshop on Semantic Evaluation (SemEval 2013), pages 300–306, Atlanta, Georgia, June 14-15, 2013. c⃝2013 Association for Computational Linguistics AI-KU: Using Substitute Vectors and Co-Occurrence Modeling for Word Sense Induction and Disambiguation Osman Bas¸kaya Enis Sert Volkan Cirik Deniz Yuret Artificial Intelligence Laboratory Koc¸ University, ˙Istanbul, Turkey {obaskaya,esert,vcirik,dyuret}@ku.edu.tr Abstract Word sense induction aims to discover differ- ent senses of a word from a corpus by us- ing unsupervised learning approaches. Once a sense inventory is obtained for an ambiguous word, word sense discrimination approaches choose the best-fitting single sense for a given context from the induced sense inventory. However, there may not be a clear distinction between one sense and another, although for a context, more than one induced sense can be suitable. Graded word sense method al- lows for labeling a word in more than one sense. In contrast to the most common ap- proach which is to apply clustering or graph partitioning on a representation of first or sec- ond order co-occurrences of a word, we pro- pose a system that creates a substitute vec- tor for each target word from the most likely substitutes suggested by a statistical language model. Word samples are then taken accord- ing to probabilities of these substitutes and the results of the co-occurrence model are clus- tered. This approach outperforms the other systems on graded word sense induction task in SemEval-2013. 1 Introduction There exists several drawbacks of representing the word senses with a fixed-list of definitions of a man- ually constructed lexical database. There is no guar- antee that they reflect the exact meaning of a tar- get word in a given context since they usually con- tain definitions that are too general (V´eronis, 2004). More so, lexical databases often include many rare senses while missing corpus/domain-specific senses (Pantel and Lin, 2004). The goal of Word Sense In- duction (WSI) is to solve these problems by auto- matically discovering the meanings of a target word from a text, not pre-defined sense inventories. Word Sense Discrimination (WSD) approaches determine best-fitting sense among the meanings that are dis- covered for an ambiguous word. However, (Erk et al., 2009) suggested that annotators often gave high ratings to more than one WordNet sense for the same occurrence. They introduced a novel annota- tion paradigm allowing that words have more than one sense with a degree of applicability. Unlike previous SemEval tasks in which systems labeled a target word’s meaning with only one sense, word sense induction task in SemEval-2013 relaxes this by allowing a target word to have more than one sense if applicable. Word sense induction approaches can be catego- rized into graph based models, bayesian, and vector- space ones. In graph-based approaches, every con- text word is represented as a vertex and if two con- text words co-occur in one or more instances of a target word, then two vertices are connected with an edge. When the graph is obtained, one …

Content truncated to 500 words