Efficient spoken dialogue control depending on the speech recognition rate and system's database

We present dialogue control methods (the dual-cost method and the trial dual-cost method) that enable a spoken dialogue system to convey information to the user in as short a dialogue as possible depending on the speech recognition rate and the content of its database. Both methods control a dialogue so as to minimize the sum of two costs: the confirmation cost (Ccost) and the information transfer cost (I-cost). The C-cost is the length of a subdialogue for confirming a user query, and the I-cost is the length of a system response generated after the confirmations. The dual-cost method can avoid the unnecessary confirmations that are inevitable in conventional methods. The trial dual-cost method is an improved version of the dualcost method. Whereas the dual-cost method has the limitation that it generates a system response based on only the content of a query that the user has acknowledged in the confirmation subdialogue, the trial dual-cost method does not. Dialogue experiments prove that the trial dual-cost method outperforms the dual-cost method and that both methods outperform conventional ones.