HIT: Web based Scoring Method for English Lexical Substitution

Proceedings of the 4th International Workshop on Semantic Evaluations (SemEval-2007), pages 173–176, Prague, June 2007. c⃝2007 Association for Computational Linguistics HIT: Web based Scoring Method for English Lexical Substitution Shiqi Zhao, Lin Zhao, Yu Zhang, Ting Liu, Sheng Li Information Retrieval Laboratory, School of Computer Science and Technology, Box 321, Harbin Institute of Technology Harbin, P.R. China, 150001 { zhaosq, lzhao, zhangyu, tliu, lisheng }@ir.hit.edu.cn Abstract This paper describes the HIT system and its participation in SemEval-2007 English Lexical Substitution Task. Two main steps are included in our method: candidate sub- stitute extraction and candidate scoring. In the first step, candidate substitutes for each target word in a given sentence are ex- tracted from WordNet. In the second step, the extracted candidates are scored and ranked using a web-based scoring method. The substitute ranked first is selected as the best substitute. For the multiword subtask, a simple WordNet-based approach is em- ployed. 1 Introduction Lexical substitution aims to find alternative words that can occur in given contexts. It is important in many applications, such as query reformulation in question answering, sentence generation, and paraphrasing. There are two key problems in the lexical substitution task, the first of which is candidate substitute extraction. Generally speaking, synonyms can be regarded as candidate substitutes of words. However, some looser lexical relationships can also be considered, such as Hypernyms and Hyponyms defined in WordNet (Fellbaum, 1998). In addition, since lexical substitution is context dependent, some words which do not have similar meanings in general may also be substituted in some certain contexts (Zhao et al., 2007). As a result, finding a lexical knowledge base for substitute extraction is a challenging task. The other problem is candidate scoring and ranking according to given contexts. In the lexical substitution task of SemEval-2007, context is con- strained as a sentence. The system therefore has to score the candidate substitutes of each target word using the given sentence. The following questions should be considered here: (1) What words in the given sentence are “useful” context? (2) How to combine the context words and use them in rank- ing candidate substitutes? For the first question, we can use all words of the sentence, words in a win- dow, or words having syntactic relations with the target word. For the second question, we can re- gard the context words as “bag of words”, n-grams, or syntactic structures. In HIT, we extract candidate substitutes from WordNet, in which both synonyms and hypernyms are investigated (Section 3.1). After that, we score the candidates using a web-based scoring method (Section 3.2). In this method, we first select frag- ments containing the target word from the given sentence. Then we construct queries by replacing the target word in the fragments with the candidate substitute. Finally, we search Google using the constructed queries and score each candidate based on the counts of retrieved snippets. The rest of this paper is organized as follows: Section 2 reviews some related work on lexical substitution. Section 3 describes …

Content truncated to 500 words