Text to Sign Language Interpreter for Sinhala

Abstract

The deaf and hearing impaired make up a considerable community in Sri Lanka with certain essentials that have researchers recently begun to tar- get. There is no such proper way to communicate with deaf society instead of learning sign language. Sign language is a natural language which can be used as a powerful weapon for communicate with deaf society. Unfortunately, being a deaf make the life di cult in many ways such as, learning a vocal oriented language is a di cult task without hearing voices and di culty with understanding ideas of non-deaf people. Basically, the target objective of re- search is to ll the communication gap between deaf and non-deaf people. In this thesis, this problem was tackled by presenting the "Text to Sign Lan- guage Interpreter (TSLI)" system to translate Sinhala text into Sinhala Sign Language. In order to do that, input sentences are tokenized and individual words are extracted. Then morphological analysis is carried for identify the basic components of individual words. Basic components can be a sign stem, sign marker or ngerspelling characters. After that, basic word components are mapped with corresponding sign video clips on sign library. Moreover, sign library contains prerecorded video clips which were made as part of the research. Final translation is made by concatenating single video clips as a one video stream. Two major evaluation tests were carried to measure the comprehension level of understanding of users about TSLI output; Individual word analysis and semantic meaning analysis. 50 sample sentences were tested in the eval- uation phase. According to the results of individual word analysis, 47% of individual signs were correctly identi ed, 20% of signs were partially identi- ed and 33% of signs were not identi ed. Somehow, MFR value is a ected for the understandability of signs. Regarding with semantic analysis of stu- dent responses, 14% of signing sentences were perfectly understood. 25% iv of sentences were understood very well. However, results were depended on three types of problems such as transition problem between signs, similarity problem of signs and poor Sinhala knowledge of students. In addition to the main evaluation, students were rated the TSLI system and most of were rated as a reasonable application. Finally, the research will helpful to any deaf or non-deaf people who are interesting to learn sign language. Moreover, research output can be used directly to translate Sinhala news programs into sign language.