5th International Conference on Spoken Language Processing
The robust estimation of language models for new applications of spoken dialogue systems often suffers from a lack of available training material. An alternative to training is to adapt initial language models to a new task by exploiting material from recognition. We investigate different methods for online-adaptation of language models. Apart from supervised and unsupervised adaptation, we look at two refined approaches: the first allows multiple hypotheses from N-best lists for adaptation and the second uses confidence measures to reject unreliably recognized sentences. We apply adaptation both to the language model used in the recognizer to focus the beam search and to the stochastic language understanding grammar. It turns out that the understanding grammar can be improved quite significantly using N-best lists or confidence measures, whereas unsupervised adaptation may even result in a deterioration of the system. The language model used in the recognizer is also improved very satisfactory.
Bibliographic reference. Souvignier, Bernd / Kellner, Andreas (1998): "Online adaptation of language models in spoken dialogue systems", In ICSLP-1998, paper 0961.