Sign Language Translation

Analysis of Notation Systems for Machine Translation of Sign Languages

Principle researcher: Jessica Jeanne Hutchinson, Computer Science Honours student at Rhodes University
Project Supervisor: Mr. James Connan, lecturer at Rhodes University.

Machine translation of sign languages is complicated by the fact that there are few standards for sign languages, both in terms of the actual languages used by signers within regions and dialogue groups, and also in terms of the notations with which sign languages are represented in written form. A standard textual representation of sign languages would aid in optimising the translation process.

This area of research still needs to determine the best, most efficient and scalable techniques for translation of sign languages. Being a young field of research, there is still great scope for introducing new techniques, or greatly improving on previous techniques, which makes comparing and evaluating the techniques difficult to do. The methods used are factors which contribute to the process of translation and need to be considered in an evaluation of optimising translation systems.

This project analyses sign language notation systems; what systems exists, what data is currently available, and which of them might be best suited for machine translation purposes. The question being asked is how using a textual representation of signs could aid machine translation, and which notation would best suit the task.
A small corpus of SignWriting data was built and this notation was shown to be the most accessible. The data was cleaned and run through a statistical machine translation system. The results had limitations, but overall are comparable to other translation systems, showing that translation using a notation is possible, but can be greatly improved upon.