Tiantianzhu7:System Description of Semantic Textual Similarity (STS) in the SemEval-2012 (Task 6)

First Joint Conference on Lexical and Computational Semantics (*SEM), pages 575–578, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics Tiantianzhu7:System Description of Semantic Textual Similarity (STS) in the SemEval-2012 (Task 6) Tiantian Zhu Department of Computer Science and Technology East China Normal University 51111201046@student.ecnu.edu.cn Man Lan Department of Computer Science and Technology East China Normal University mlan@cs.ecnu.edu.cn Abstract This paper briefly reports our submissions to the Semantic Textual Similarity (STS) task in the SemEval 2012 (Task 6). We first use knowledge-based methods to compute word semantic similarity as well as Word Sense Dis- ambiguation (WSD). We also consider word order similarity from the structure of the sen- tence. Finally we sum up several aspects of similarity with different coefficients and get the sentence similarity score. 1 Introduction The task of semantic textual similarity (STS) is to measure the degree of semantic equivalence between two sentences. It plays an increasingly important role in several text-related research and applications, such as text mining, Web page retrieval, automatic question-answering, text summarization, and ma- chine translation. The goal of the Semeval 2012 STS task (task 6) is to build a unified framework for the evaluation of semantic textual similarity modules for different systems and to characterize their impact on NLP applications. Generally, there are two ways to measure sim- ilarity of two sentences, i.e, corpus-based meth- ods and knowledge-based methods. The corpus- based method typically computes sentence similar- ity based on the frequency of word occurrence or the co-occurrence between collocated words. For ex- ample, in (Islam and Inkpen, 2008) they proposed a corpus-based sentence similarity measure as a func- tion of string similarity, word similarity and com- mon word order similarity (CWO). The knowledge- based method computes sentence similarity based on the semantic information collected from knowl- edge bases. With the aid of a number of success- ful computational linguistic projects, many seman- tic knowledge bases are readily available, for ex- ample, WordNet, Spatial Date Transfer Standard, Gene Ontology, etc. Among them, the most widely used one is WordNet, which is organized by mean- ings and developed at Princeton University. Sev- eral methods computed word similarity by using WordNet, such as the Lesk method in (Banerjee and Pedersen, 2003), the lch method in (Leacock and Chodorow, 1998)and the wup method in (Wu and Palmer, 1994). Generally, although the knowledge- based methods heavily depend on the knowledge bases, they performed much better than the corpus- based methods in most cases. Therefore, in our STS system, we use a knowledge-based method to com- pute word similarity. The rest of this paper is organized as follows. Sec- tion 2 describes our system. Section 3 presents the results of our system. 2 System Description Usually, a sentence is composed of some nouns, verbs, adjectives, adverbs and/or some stop words. We found that these words carry a lot of informa- tion, especially the nouns and verbs. Although the adjectives and adverbs also make contribution to the semantic meaning of the sentence, they are much weaker than …

Content truncated to 500 words