AWeb-Deployed Swedish Spoken CALL System Based on a Large Shared English/Swedish Feature Grammar

We describe a Swedish version of CALL-SLT, a web-deployed CALL system that allows beginner/intermediate students to practise generative spoken language skills. Speech recognition is grammar-based, with language models derived, using the Regulus platform, from substantial domain-independent feature grammars. The paper focusses on the Swedish grammar resources, which were developed by generalising the existing English feature grammar into a shared grammar for English and Swedish. It turns out that this can be done very economically: all but a handful of rules and features are shared, and English grammar essentially ends up being treated as a reduced form of Swedish. We conclude by presenting a simple evaluation which compares the Swedish and French versions of CALL-SLT.