Automating the Librarian: Belief Revision as a base for System Action and Communication with the User

Abstract

This paper describes a current research project investigating belief revision in intelligent systems by modelling the librarian in interaction with a literature-seeking user. The work is designed both to test a theory of agent behaviour based on belief revision proposed by Galliers, and to evaluate a model of the librarian developed by Belkin, Brooks and Daniels, through computational implementation. Agent communication is seen as motivated by and motivating belief changes, where belief revision is determined by coherence, combining endorsement, connectivity and conservatism. The librarian is viewed as a distributed expert system with many individual specialised functions operating in particular belief domains. The paper describes our first implementation of the belief revision mechanism and of a very primitive librarian, designed to test the basic viability of our ideas and to allow us to explore different forms of the distributed system architecture. The work described in this paper is basic research aimed at a challenging and necessarily very long-term goal, automating the librarian. Searching online bibliographic databases to obtain literature references or documents for end users is an important component of a modern librarian or information officer's work. It requires professional knowledge and skill, so providing convenient direct access to bibliographic services for end users instead calls for sophisticated interfaces able both to determine the user's need and to express this in a way suited to searching the bibliographic file. In general, that is, it is necessary both to identify the user's topic and to specify this in the indexing or classification language used to describe documents in the file. But even when the search language is the natural language of the file documents' titles, abstracts or texts, professional knowledge and skill is required for effective searching. Some first steps have already been taken in automating the intermediary by, for example, Pollitt, 22 Vickery et alP and Brajnik et a/.; 3 but what has been done so far has been very limited, especially in the system's subject scope. More effective systems would call on artificial intelligence (AI) techniques for reasoning on knowledge in interacting with and acting for the user. The project described in this paper is therefore concerned on the one hand with appropriate general mechanisms for agents manipulating beliefs and conducting dialogue, and on the other with deploying these mechanisms within the framework supplied by the literature-searching task and by a model of the librarian's characteristic knowledge and actions. The project is intended to investigate key ideas about what is in principle involved in automating the intermediary. But given the complexity of the real librarian's task it can only attempt an initial, very simplistic, laboratory system: the foundational character of the work which is needed for eventual proper systems means we cannot envisage realistic prototyping. The project is also in progress, so this paper describes the work's starting point, plan, and what has been done so far, but cannot provide final results or evaluate these. It must further be emphasised that while the research is presented here as directed towards document retrieval, because automating the intermediary requires general capabilities as well as specialised knowledge and developing these capabilities is a concern of AI this work is also seeking to contribute to AI as a whole. Thus while from one point of view the aim is to apply AI ideas to information retrieval (IR), from another IR provides a valuable study context for modelling the way any agents adopt or change their beliefs about the world, particularly through engagement in dialogue. At the same time, while the work is concerned with helping the user to obtain literature, it is not primarily concerned with the type of display-oriented support system developed by e.g. McAlpine and Ingwersen, 21 or with 'plain' public access as in Okapi. 24 Section 2 of the paper describes the general theory of belief revision underlying the whole; Section 3 the view of the intermediary adopted as a project starting point; and Section 4 the current state of the research and intentions for the future. Cawsey et al. provides a fuller