| Summary: | 碩士 === 國立成功大學 === 資訊工程學系碩博士班 === 92 === The hearing-impaired people generally use sign language to express their intention. However, hearing people don’t know how to use sign language and, therefore, the communication obstacle between them are formed. Presently, machine translation researches mainly focus on word-to-word translation, and some syntactic rule-based translation. On the other hand, the lack of parallel corpus of sign language limits the development of machine translation. For this reason, TSL translation system applied present machine translation technique will have no good performance. In this study, we propose a statistical approach using syntactic information for the translation from Chinese to Taiwanese Sign Language (TSL).
More specially, we focuses on 1) establishing a integrated corpus which consist of word, part of speech, semantic role, and semantic feature by combining information of several Chinese corpora, 2) collecting the context free grammar and training its probability by EM algorithm for proposal translation mechanism, 3) proposing a Chinese to Taiwanese Sign Language translation mechanism based on sentence structure and using syntactic information by complete statistical parsing model, and 4) integrating the above approaches into a Chinese to TSL translation system.
In order to evaluate our proposed approaches, 2,036 parallel sentences, in which the mean length of sentence is 5.6 words, were collected. Of this database, 80% was used as the training corpus and the remainder for testing, and 7,931 transfer rules were obtained. The translation performance achieved 81.6% and 91.5% accuracy for top-1 and top-5 candidates respectively, and got 0.087 Alignment Error Rate (AER). All of the above TSL translation evaluations, our proposed approach achieved higher performance than IBM Model 3. In Mean Opinion Score evaluation, the average translation performance of our proposed system also achieved 81% satisfactory degree. Consequently, our proposed system can provide a channel of communication between the deaf and the able-bodied, and applied to TSL education in future.
|
|---|
