Unsupervised Word Sense Disambiguation Using Neighborhood Knowledge

Unsupervised Word Sense Disambiguation Using Neighborhood Knowledge1 Huang Heyan 1,2, Yang Zhizhuo 1,2, and Jian Ping 1,2 1Beijing Engineering Applications Research Center of High Volume Language Information Processing and Cloud Computing, Beijing Institute of Technology, No.5 Yard, Zhong Guan Cun South Street Haidian District,Beijing, 100081,China 2Department of Computer Science, Beijing Institute of Technology, No.5 Yard, Zhong Guan Cun South Street Haidian District,Beijing, 100081,China {hhy63,10907029,pjian}@bit.edu.cn Abstract. Usually ambiguous words contained in article appear several times. Almost all existing methods for unsupervised word sense disambiguation make use of information contained only in ambiguous sentence. This paper presents a novel approach by considering neighborhood knowledge. The approach can naturally make full use of the within-sentence relationship from the ambiguous sentence and cross-sentence relationship from the neighborhood knowledge. Experimental results indicate the proposed method can significantly outperform the baseline method. Keywords: Unsupervised WSD, Neighborhood Knowledge, Similarity Measure, Graph- based Ranking Algorithm. 1. Introduction Word Sense Disambiguation (WSD), the task of indentifying the intended meaning (sense) of words in context is one of the most important problem in natural language processing. Though it is often characterized as an intermediate task rather than an end in itself, it has the potential to improve the performance of many applications including information retrieval, machine translation and so on. Existing methods conduct WSD usually using only the information contained in the ambiguous sentence to be disambiguated. They utilize context words within predefined window in sentence together with other syntactic information. It has been proved that expanding context window size around the target ambiguous word can help to enhance the WSD performance. 1 This work is supported by the National Natural Science Foundation of China under Grant No.61132009, the Cloud Computing for High Volume Language Information Intelligent Processing, the Primary Funds of National Defense ,Open Project Program of National Laboratory of Pattern Recognition ,BIT Fundamental Research Projects No.3070012210917 Copyright 2011 by Huang Heyan, Yang Zhizhuo, Jian Ping 25th Pacific Asia Conference on Language, Information and Computation, pages 333–342 333 However, expanding window size unboundedly will bring not only useful information but also some noise which may finally deteriorate the WSD performance. Can we find other way to expand context words without bringing too much noise? In this study, we proposed to conduct WSD using collaborative techniques. One common assumption of existing methods is that the sentences are independent of each other, and WSD is carried out separately without interactions among the sentences. However, some ambiguous words contained in article appear more than one time, the multiple sentences contain the same ambiguous word within the article actually have mutual influences and contain useful clues which will helpful to deduce word sense from each other. For example, sentence containing program design commonly shares similar topic with the sentence containing computer program, thus the meaning of the ambiguous words program in those two sentences can be deduced from each other. The idea is borrowed from the observation that ambiguous words appear in topic related sentences contained in the same article often share common …

Content truncated to 500 words