Search

For the speech-to-speech translation system verbmobil the dialogue component has the task of providing contextual information for other verbmobil subcomponents and to follow the flow of the dialogue. Since verbmobil operates on demand, only parts of the dialogue are processed using syntactic and semantic knowledge. Therefore, robustness with respect to incomplete structures is a basic requirement. We show the performance of the statistic prediction module and the planning module under such conditions.

In this paper, we present a statistical approach for speech act prediction in the dialogue component of the speech-to-speech translation system Verbmobil. The prediction algorithm is based on work known from language modelling and uses N-gram information computed from a training corpus. We demonstrate the performance of this method with 10 experiments. These experiments vary in two dimensions, namely whether the N-gram information is updated while processing, and whether deviations from the standard dialogue structure are processed. Six of the experiments use complete dialogues, while four process only the speech acts of one dialogue partner. It is shown that the predictions are best when using the update feature and deviations are not processed. Even the processing of incomplete dialogues then yields acceptable results. Another experiment shows that a training corpus size of about 40 dialogues is sufficient for the prediction task, and that the structure of the dialogues of the Verbmobil corpus we use differs remarkably with respect to the predictions.

We present the application of statistical language modeling methods for the prediction of the next dialogue act. This prediction is used by different modules of the speech-to-speech translation system VERBMOBIL. The statistical approach uses deleted interpolation of n-gram frequencies as basis and determines the interpolation weights by a modified version of the standard optimization algorithm. Additionally, we present and evaluate different approaches to improve the prediction process, e.g. including knowledge from a dialogue grammar. Evaluation shows that including the speaker information and mirroring the data delivers the best results.