Research on Word Segmentation for Chinese Sign Language

Research on Word Segmentation for Chinese Sign Language Yinchao Cheng1, Baocai Yin, and Yanfeng Sun Key Laboratory of Multimedia and Intelligent Software, Beijing University of Technology, Beijing, 100022, China haohuanc@emails.bjut.edu.cn, {ybc,yfsun}@bjut.edu.cn Abstract. It remains to be a difficult issue to convert Chinese language into Chinese sign language, which makes it hard to implement an obstacle free Chinese sign language information service under pervasive environment. This paper presents an improved algorithm of forward maximum match approach (MM) and backward maximum match approach (FMM), by taking the characteristics of Chinese sign language into consideration. Besides, a method to reorganize the Chinese sign language dictionary is presented. This paper also proposes a novel strategy of disambiguation based on the statistical information of the context and the mutual information. The experiment results indicate that the accuracy and the efficiency of word segmentation can improve significantly compared to conventional algorithms. Keywords: Word Segmentation; Chinese Sign Language; Disambiguation; Mutual Information 1 Introduction In human languages, a combination of multiple continuous words, or phrases, is usually a minimum meaningful unit, and word segmentation (WS) is one of the major issues of information processing in character-based languages. Because there are no explicit word boundaries in these languages, WS is important for information retrieval, machine translation, lexicon construction, digital libraries, and Chinese sign language. The conventional WS is different from the one in Chinese sign language environment. It should not be based on the subjective evaluation of the views, but be evaluated by whether it could contribute to improving the accuracy and precision of Chinese sign language synthesis. The basic approaches of word segmentation in character-based languages can be partitioned into two categories: statistic-based [8] [9] and dictionary-based [1]. Statistic-based approaches make use of statistical properties, such as frequencies of characters and character sequences in the corpus [2]. In practice, however, to choose a corpus which is big enough and includes all categories is impossible, a statistic dictionary that contains all possible words is unfeasible, costly and unnecessary [3]. Dictionary- based approaches use a dictionary to identify words. When matched in the dictionary, a sequence of characters will be extracted as a word. There are many match criteria in literature, such as maximum match, minimum match and hybrid approach. The maximum match approach can be further divided into MM method and FMM method. However, WS under Chinese sign language environment is quite different from conventional WS. Firstly, speed and accuracy must be considered for the approach of WS under Chinese sign language environment. Each word is translated into a specific series of sign language action, so the accuracy of WS has direct impact on the accuracy of animation synthesis for Chinese sign language. Secondly, Chinese sign language has its particularity in following aspects: (1) Two dictionaries, basic word dictionary (BWDIC) and finger word dictionary (FWDIC); (2) Low vocabulary in BWDIC, only about 6,000 items; (3) Distribution disparity in BWDIC, words made up of single and double characters take up a considerable proportion (total 90.98%). Because of the particularity, new …

Content truncated to 500 words