In recent years, much research work on spoken dialogue systems has been done all over the world. Hundreds of spoken dialogue systems have been provided by many organizations. However, most of these systems are based on limitative tasks, such as querying for travel information or weather forecast information. This paper presents our research on a spoken dialogue system that is different from those systems based on limitative tasks. This system is provided mostly for entertainment rather than getting useful information. Our research work includes designing man-machine interactive content, analyzing speech recognition result, offering answers, and also properly leading the conversation process in order to make the topic not overstep the range the system can analyze. All the work can be summarized as follows: 1. Establish a question-answering system for chatting by speech, which is based on combination of keywords. We analyze lots of dialogue corpus to design FAQ, and also, design more than one answers for each question and memorize each question and its answers in a rule; 2. Propose three ways of extending FAQ; 3. Propose a method of computing sentence similarity, which takes account of the characteristics of spoken language and the influence of speech recognition module; 4. Provide some valid methods to achieve better system performance and show higher-level intelligence.
修改评论