Explaining and justifying the advice of a decision support system: a natural language generation approach

This paper describes a method for generating explanations in decision analytic contexts. Unlike other approaches, we use natural language generation techniques. The novelty of the work stems from the development of a library, of text plans that structure the explanation messages conveyed. This makes our approach generic and easily adjusted to different contexts. In order to demonstrate the applicability of the method, we have developed a natural language generator that explains and justifies the advice of a decision support system for nuclear emergencies. The generator outputs two reports: a comparative report that explains the rationale behind the ranking of the alternatives and a sensitivity analysis report that gives an overall assessment of the decision model and describes the effect of varying a decision parameter. (C) 2002 Elsevier Science Ltd. All rights reserved.