Word Extraction Based on Semantic Constraints in Chinese Word-Formation

This paper presents a novel approach to Chinese word extraction based on semantic information of characters. A thesaurus of Chinese characters is conducted. A Chinese lexicon with 63,738 two-character words, together with the thesaurus of characters, are explored to learn semantic constraints between characters in Chinese word-formation, forming a semantic-tag-based HMM. The Baum-Welch re-estimation scheme is then chosen to train parameters of the HMM in the way of unsupervised learning. Various statistical measures for estimating the likelihood of a character string being a word are further tested. Large-scale experiments show that the results are promising: the F-score of this word extraction method can reach 68.5% whereas its counterpart, the character-based mutual information method, can only reach 47.5%.