Machine translation is one of the oldest tasks of computational linguistics, and it has achieved a certain amount of success, especially in genres with limited vocabulary.  Machine translation into or out of signed languages is still in its infancy, however, and is often confused with sign synthesis and recognition.

This report discusses the progress of a project to create an application to translate weather reports prepared in English by the National Oceanic and Atmospheric Administration (NOAA) into American Sign Language (ASL), the principal language of the Deaf community in the United States and Canada.  The nonwritten nature of ASL poses a problem for the output of the application: there exist a number of writing systems for the language, but none have a wide following among signers.  The system currently produces output in the Newkirk 1986 writing system, but in order for it to be widely used, it will have to be linked with a sign synthesis application.

Weather reports were chosen as input because they are both useful and readily available on the World Wide Web, and use a well-defined set of terms, but the the existing vocabulary of weather terms in ASL was not extensive enough to match the NOAA English terminology, so the project includes a significant amount of corpus planning.  As counterpoint to the current predominance of formal theories of syntax in computational linguistics, the project also examines functional syntactic models and possible implementations in machine translation.  The presentation will include a brief demonstration of the system.

Back to the Conference Schedule