Copyright © 2012 Peng Jin et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

Distributional Similarity has attracted considerable attention in the field of natural language processing as an automatic means of countering the ubiquitous problem of sparse data. As a logographic language, Chinese words consist of characters and each of them is composed of one or more radicals. The meanings of characters are usually highly related to the words which contain them. Likewise, radicals often make a predictable contribution to the meaning of a character: characters that have the same components tend to have similar or related meanings. In this paper, we utilize these properties of the Chinese language to improve Chinese word similarity computation. Given a content word, we first extract similar words based on a large corpus and a similarity score for ranking. This rank is then adjusted according to the characters and components shared between the similar word and the target word. Experiments on two gold standard datasets show that the adjusted rank is superior and closer to human judgments than the original rank. In addition to quantitative evaluation, we examine the reasons behind errors drawing on linguistic phenomena for our explanations.