7th International Conference on Spoken Language Processing
September 16-20, 2002
This paper addresses an error-tolerant retrieval algorithm for generating sentences from ill-formed sign sequences in Taiwanese Sign Language (TSL). The design methodology is motivated by the kinematics of hand gestures for sign language. In order to increase the input rate and retrieval accuracy, the basic design strategy leads to develop an efficient and effective sign feature retrieval method. In this approach, a Multi-list Code Tree (MCT) data structure for sign feature indexing and an error-tolerant matching algorithm are proposed. For text generation, the optimal path in the word graph, generated from the sign input sequence, is incrementally estimated by using a translation model. Several interface design and evaluation methods are also conducted for empirical study. Evaluation results show that the generation rate and retrieval accuracy for sentence construction are significantly improved.
Bibliographic reference. Wu, Chung-Hsien / Chiu, Yu-Hsien / Cheng, Kung-Wei (2002): "Sign language translation using an error tolerant retrieval algorithm", In ICSLP-2002, 637-640.