Perspectives on Belief Ascription

A 1/2-day Tutorial at FAPR96,
the International Conference on Formal and Applied Practical Reasoning,
Bonn, June 1996

Anthony Jameson, University of Saarbrücken, Germany

To obtain copies of the overhead slides and the reference list, either as paper copies or in electronic form, please send e-mail to the author (jameson@cs.uni-sb.de).

Summary

Subfields of several scientific disciplines address aspects of the general issue of how agents ascribe beliefs to other agents. Some research paradigms adopt a normative perspective, considering mainly how belief ascription should be performed in particular contexts. In some of these normatively oriented paradigms--such as epistemic logic, distributed artificial intelligence, and user and student modeling--the focus is on ascriptions made with formal inference methods. In others--such as educational testing, attitude assessment, and knowledge acquisition for expert systems--a human is often assumed to be in the loop.

In the social sciences, belief ascription is mostly studied from a more descriptive perspective. Some of the research within psychology's subfields of social perception, attitude change, cognitive development, and metacognition examines the inferences that people make about the beliefs of other persons or themselves. Studies of linguistic communication conducted within various disciplines examine the often indirect inferences that listeners make about a speakers' beliefs.

The aim of this tutorial is not to provide a thorough overview of any of these research paradigms but rather (a) to convey a view of the relationships among the various perspectives on belief ascription and (b) to bring to the participants' attention ideas from relatively unfamiliar paradigms which may be relevant to their own work.

The tutorial begins by clarifying the relationships among the paradigms, characterizing each one in terms of (a) the the belief ascription contexts studied, (b) the nature of the inferences that are made in these contexts, (c) the research methods employed, and (d) some typical results of the research. It then considers cases where one paradigm could benefit from increased use of concepts, methods, or results of other paradigms. Among the normatively oriented paradigms, which inference techniques developed within one paradigm could be employed to a greater extent within other paradigms? Among the descriptively oriented paradigms, which empirical phenomena highlighted within one paradigm also occur in the situations studied by other paradigms? And regarding the relationship between the normatively and descriptively oriented paradigms: In what cases is increased cross-fertilization desirable in view of the fact that everyday human inference is often more rational than it at first appears to be?

For concreteness, each paradigm will be discussed largely with reference to typical examples of research conducted within it. Time will be allowed for discussion and debate among participants who work within different paradigms.