An Information Repository Model for Advanced Question Answering Systems

Abstract

This paper presents the design and implementation of the information repository which is the central core of the JAVELIN open-domain question answering system. JAVELIN is comprised of several modules that perform a wide variety of question answering (QA) tasks, such as question analysis, document and passage retrieval, answer candidate extraction, answer selection, answer justification, and planning. The architecture is designed to support comparative component-level evaluation, so that different strategies for each module can be integrated and tested in a straightforward way. Each time a module uses a particular piece of information to produce an output, a dependency is created. To support answer justification and introspective learning, the system can use this long-term memory to trace the origin of each answer it produces for a particular question. The JAVELIN Repository implements a complete, consistent relational model for all of the information associated with a question answering scenario.

Resources

Details