MIXCD: System Description for Evaluating Chinese Word Similarity at SemEval-2012

First Joint Conference on Lexical and Computational Semantics (*SEM), pages 425–429, Montr´eal, Canada, June 7-8, 2012. c⃝2012 Association for Computational Linguistics MIXCD: System Description for Evaluating Chinese Word Similarity at SemEval-2012 Yingjie Zhang Nanjing University 22 Hankou Road Jiangsu P. R. China jillzhyj@139.c om Bin Li Nanjing University Nanjing Normal University 122 Ninghai Road Jiangsu P. R. China gothere@126.com Xinyu Dai Nanjing University 22 Hankou Road Jiangsu P. R. China dxy@nju.edu.cn Jiajun Chen Nanjing University 22 Hankou Road Jiangsu P. R. China cjj@nju.eud.cn Abstract This document describes three systems calcu- lating semantic similarity between two Chi- nese words. One is based on Machine Readable Dictionaries and the others utilize both MRDs and Corpus. These systems are performed on SemEval-2012 Task 4: Evaluat- ing Chinese Word Similarity. 1 Introduction The characteristics of polysemy and synonymy that exist in words of natural language have always been a challenge in the fields of Natural Language Processing (NLP) and Information Retrieval (IR). In many cases, humans have little difficulty in de- termining the intended meaning of an ambiguous word, while it is extremely difficult to replicate this process computationally. For many tasks in psycholinguistics and NLP, a job is often decom- posed to the requirement of resolving the semantic similarity between words or concepts. There are two ways to get the similarity between two words. One is to utilize the machine readable dictionary (MRD). The other is to use the corpus. For the 4th task in SemEval-2012 we are re- quired to evaluate the semantic similarity of Chi- nese word pairs. We consider 3 methods in this study. One uses MRDs only and the other two use both MRD and corpus. A post processing will be done on the results of these methods to treat syno- nyms. In chapter 2 we introduce the previous works on the evaluation of Semantic Similarity. Chapter 3 shows three methods used in this task. Chapter 4 reveals the results of these methods. And conclu- sion is stated in chapter 5. 2 Related Work For words may have more than one sense, similari- ty between two words can be determined by the best score among all the concept pairs which their various senses belong to. Before constructed dictionary is built, Lesk similarity (Lesk, 1986) which is proposed as a so- lution for word sense disambiguation is often used to evaluating the similarity between two concepts. This method calculates the overlap between the corresponding definitions as provided by a diction- ary. ( ) | ( ) ( )| Since the availability of computational lexicons such as WordNet, the taxonomy can be represented as a hierarchical structure. Then we use the struc- ture information to evaluate the semantic similarity. In these methods, the hierarchical structure is often seen as a tree and concepts as the nodes of the tree while relations between two concepts as the edges. (Resnik, 1995) determines the conceptual simi- larity of two concepts by calculating the infor- mation content (IC) of the least common subsumer (LCS) of …

Content truncated to 500 words