Full text loading...
-
Building a parallel corpus of German/Swiss German Sign Language train announcements
- Source: International Journal of Corpus Linguistics, Volume 21, Issue 1, Apr 2016, p. 116 - 129
Abstract
We present a parallel corpus of German/Swiss German Sign Language train announcements. The corpus is used in a statistical machine translation system that translates from German to Swiss German Sign Language. The output of the translation system is then passed on to an animation system, the result being a sign language avatar representation on a mobile phone. Building the parallel corpus consisted of four steps: translating the written German train announcements into Swiss German Sign Language glosses, signing the announcements in front of a camera on the basis of the gloss transcriptions, notating the signs in the video recordings in a form-based sign language notation system, and adding information about non-manual features. The resulting corpus contains 3,241 sentence pairs, which makes it a large parallel corpus involving sign language.