An Intelligent Agent Based Spoken Dialog System for Content Based Image Retrieval

Current spoken dialog systems are not robust and scalable. This is due to the lack of a systematic approach for studying natural language understanding problems. We hope to propose a method to build robust dialog systems by exploring the fundamental relationship between language and mind. We argue that an intelligent agent is necessary for building viable natural language understanding systems for complex applications. By utilizing language knowledge related to each individual word, we claim scalable and robust semantic analysis can be achieved under the framework of an intelligent agent. A novel knowledge representation structure designed for language---the Word Concept Model---is proposed to separate world knowledge and language knowledge so that an intelligent agent can have integrated information processing for multimodal interface. An extensible layered concept structure is proposed for storing concepts abstracted from the internal world of the intelligent agent in a relatively open structure to enable expansion of the concept space. The word concept model then stores information between the words and this concept space. Syntax information related to a specific word can also be stored in the word concept model. The semantic analysis algorithm processes sentences according to their sentence types to achieve uniform processing. We tested this idea by building a spoken dialog interface for a content-based image retrieval system. An evaluation of the system was conducted, and a relatively robust performance was achieved.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Electrical Engineering

Grantor

University of Illinois at Urbana-Champaign

Year dc:date

2015

Author dc:creator

• Li, Yang

Contributors dc:contributor

• Levinson, Stephen E.
• Huang, Thomas S.

• Artificial Intelligence

Language dc:language

eng

Identifier

(MiAaPQ)AAI3202128

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/80949