International Conference on Expert Systems for Development, Bangkok, Thailand, March 1994 CONTEXTUALIZED EXPLANATIONS Brézillon P. J. LAFORIA, Box 169, University Paris 6, 4 place Jussieu, 75252 Paris Cedex 05, France Phone: +33 1 44 27 70 08 - Fax: +33 1 44 27 70 00 - E-mail: brezil@laforia.ibp.fr Keywords: explanation process, context, knowledge-based systems, human-computer cooperation Summary: The aim of the paper is to fill the gap between theory and practice in the production of explanations by a system. One reason of this gap arises from the fact that a problem is often solved thanks to a cooperation between the user and the system, and both participants in the cooperation need explanations. Explanations essentially depend on the context in which the user and the system interact. Such contextualized explanations are the result of a process and constitute a medium of communication between the user and the system during the problem solving. We focus on the need to make the context notion explicit in the explanation process. We analyze explanation and context in term of chunks of knowledge. Then we point out what the contribution of the context to explanation is. An example, which is drawn from a real application, introduces what the problem is. INTRODUCTION A main line of research for introducing explanation capabilities in Knowledge-Based Systems (KBSs) supposes that explanations are used to transmit knowledge from the machine to a user and to improve learning ability of the latter. The lack of consideration for the end-user during the KBS design leads end- users to disregard these explanations. Conversely, there are KBSs that have been developed in an industrial context. These contain various types of explanation tailored for the needs of the user and the application, but do not explore all potentialities of explanations [BRE92]. There is a gap between theory and practice. One reason of this gap is that a problem generally is solved in a cooperative way by the user and the system. Then, an explanation is the result of successive transformations of a proposition that must be accepted by both the user and the system. Such explanations essentially depend on the context in which the user and the system interact. Indeed, the context of the user-system cooperation takes into account different pieces of information like [MAS92]: The history of interaction with the user; The transaction history; The characteristics of the user; The user's intentions; The possible sources of ambiguity; The state of the system; The user's profile and system access allowed by the security. Making explicit all these pieces of the context enables to: Tailor explanations to the users' needs; Simplify correctly a complex information for the user; Paraphrase; Structure the explanation; Manage counter-examples; Guide the research and the focus of attention; Resolve ambiguities; Correct misunderstanding; Help learning; Fill possible gaps; Reduce cognitive load; Develop a model of the "user- in-situation"; Make sure that the participants' mental models match; and Improve qualitative and quantitative performance of the explanation. In this paper, we claim that a system must produce context-sensitive explanations to enhance its intervention in its cooperation with a user. A real- world example introduces the importance of the context notion in explanation in Section 1. We then expose what the role of explanation is according to the user's viewpoint in Section 2. We give a representation of the explanation process in Section 3. Section 4 exposes the notion of context that we retain. Section 5 discusses the relationships of context and explanation 1. A REAL-WORLD EXAMPLE The example is drawn from a real application [BAU92]. First, we give information on the domain and the task of diagnosis in which the application has been developed. Then we give some of the various contexts and implicit knowledge that exist. In the following, we describe the two ways that have been chosen to implement the notion of context in an implicit way in our system. We end with a presentation of the advantages of making that notion of context explicit. 1.1 The domain and the diagnosis task The French national power system is composed of controlled systems (Extra High Voltage substations) and control systems. Elements of the controlled system are lines and busbars that constitute a network. Elements of the control system are controllers. A controller links a busbar to a line or