SRCB-WSD: Supervised Chinese Word Sense Disambiguation with Key Features

Proceedings of the 4th International Workshop on Semantic Evaluations (SemEval-2007), pages 300–303, Prague, June 2007. c⃝2007 Association for Computational Linguistics SRCB-WSD: Supervised Chinese Word Sense Disambiguation with Key Features Yun Xing Ricoh Software Research Center Beijing Co., Ltd Beijing, China yun.xing@srcb.ricoh.com Abstract This article describes the implementation of Word Sense Disambiguation system that participated in the SemEval-2007 multilin- gual Chinese-English lexical sample task. We adopted a supervised learning approach with Maximum Entropy classifier. The fea- tures used were neighboring words and their part-of-speech, as well as single words in the context, and other syntactic features based on shallow parsing. In addition, we used word category information of a Chinese the- saurus as features for verb disambiguation. For the task we participated in, we obtained precision of 0.716 in micro-average, which is the best among all participated systems. 1 Introduction Word Sense Disambiguation(WSD) is the process of assigning a meaning to a word based on the context in which it occurs. It is very important to many re- search fields such as Machine Translation, Informa- tion Retrieval. The goal of the multilingual Chinese- English lexical sample task in SemEval-2007 is to predict the correct English translation for an am- biguous Chinese word w. We considered this task as a classification prob- lem, and our system adopted a supervised learning approach with Maximum Entropy classifier, which is widely used in natural language processing(NLP). Within the Maximum Entropy framework, evidence from different features can be combined with no as- sumptions of feature independence. The used fea- tures include neighboring words and their part-of- speech(POS), single words in the context, and other syntactic features based on shallow parsing. In ad- dition, we used word category information of a Chi- nese thesaurus for verb disambiguation. Note that we did not do any feature selection in this work. Next, we will describe the Maximum Entropy framework and detail the features used in our WSD system. 2 Maximum Entropy Maximum entropy modelling is a framework for in- tegrating information from many heterogeneous in- formation sources for classification (Manning and Sch¨utze, 1999). It has been successfully applied to a wide range of NLP tasks, including sentence boundary detection, POS tagging, and parsing (Rat- naparkhi, 1998) . The system estimates the condi- tional probability that an ambiguous word has sense x given that it occurs in context y, where y is a con- junction of features. The estimated probability is derived from feature weights which are determined automatically from training data so as to produce a probability distribution that has maximum entropy, under the constraint that it is consistent with ob- served evidence (Dang et al., 2002). We used the im- plementation of Maximum Entropy framework with OpenNLP MAXENT1, where each nominal feature was represented as “feature code=value”. Based on this framework, we defined the feature set and im- plemented the interface of feature extraction. For the convenient of evaluation, the default parameters 1http://maxent.sourceforge.net/ 300 of training model were used. 3 Used Features Many research (Stevenson and Wilks, 2001; …

Content truncated to 500 words