Abstract
In order to handle free speech while driving, we constructed spoken dialogue system combining a question-and-answer database and a largescale language model. We analyzed user behavior and the usefulness of the system by collecting utterances using an actual vehicle. Among the utterances that could not be answered by the conventional system, the large-scale language model was able to respond to more than half of them. This indicated the effectiveness of the introduction of the large-scale language model.
