BOGAR_LN: An Agent Based Component Framework for Developing Multi-modal Services Using Natural Language

This paper describes BOGAR_LN an agent-based component-ware framework which consist of a multi-layered library and support tools for component creation, retrieval, management, and reuse. BOGAR_LN library provides application developers with four categories of reusable component models: Agent Organization models, Agent models, Resource models, and Basic computing entities. For each category, there are generic components which represents abstract reusable patterns and application oriented components. Component instances are made up of three blocks of information: design description in UML, code implementation in Java, and an extensible collection of attribute-value descriptors. The initial repertoire of repository components, comes from previous experiences in developing Agent based telecom services. The paper also presents the metrics and the evaluation approach to assess the benefits of the framework. Evaluation data have been gathered by using the framework to develop a mixed-initiative spoken dialog system for appointment management over the telephone. Results showed significant reductions on both project duration and cost. Compared to previous developments the time and the engineering effort required to build the service, was on average 65% less when using BOGART_LN.