LOGIN
>> Home
>> Topics
>> Students
>> Partners
>> Statistics


Information for topics

Topic Id:
ID topic: 444
Partner Email: L.J.M.Rothkrantz@tudelft.nl
Project Title: Using dialogue history in dialogue management of automatic inquiry systems
Abstract: Automatic inquiry systems are systems that allow a user to access a database using a spoken natural language dialogue. As well as components for speech recognition, language understanding, speech synthesis, etc., automatic inquiry systems have a dialogue manager. Two tasks of the dialogue manager are dialogue control and knowledge processing. The knowledge processing consists of processing abstract representations of the user utterance in each dialogue turn, thereby updating the system knowledge. This report is about the knowledge representation and the knowledge processing in the dialogue manager of an automatic inquiry system. It has two parts. The first part describes the use of the dialogue history for the knowledge processing. This is done by keeping alternative interpretations of user utterances of previous turns in the system knowledge and using them for knowledge processing. A statistical theory is described for this and two strategies are given based on the theory. The strategies are demonstrated on a model automatic inquiry system and can be used in real applications. Both strategies lead to a multiple belief system. It is shown that the knowledge processing can improve because of the use of dialogue history but that the switching between the different beliefs can also cause new problems. Special attention is paid to interpreting the user utterance in relation to the system prompt, resulting in an abstract representation of the dialogue turn, which is then used in the strategies mentioned above. It is shown that an abstract representation of the system prompt can help the interpretation of the user utterance. The second part describes the extension of the SUSI system with negative knowledge. Negative knowledge is the knowledge that can be extracted from denials and corrections in the user utterance. The SUSI system is a set of modules for the construction of automatic inquiry systems. A way to use negative knowledge in a single belief system having a simple grammar is shown. The negative knowledge handling has been implemented in SUSI and has been demonstrated and tested for the SUSI based application PADIS. It has been shown that certain PADIS dialogues improve significantly.
Advisor: Leon Rothkrantz
Link:
Degree: Master
 Keywords:
Computer Software
Data modeling
Information systems