A Model for Adapting Explanations to the User‘s Likely Inferences

  • Authors:

  • Helmut Horacek

  • Affiliations:

  • Universität Bielefeld, LILI-Fakultät, Postfach 100 131, D-33501 Bielefeld, Germany

  • Venue:
  • User Modeling and User-Adapted Interaction
  • Year:
  • 1997

Quantified Score

Hi-index 0.00

Visualization

Abstract

In order to generate natural, high quality textualpresentations in technical domains, good explanationsmust not only be adapted to the knowledge attributedto the intended audience, but they must also take intoaccount the inferential capabilities of theaddressees. In this paper, we present a model foranticipating contextually-motivated inferencesaddressees are likely to draw. This model is used tomotivate choices in presenting or omitting individualpieces of information; it takes into account theaddressees‘ domain expertise and expectations aboutlogical consequences of purposefully presentedinformation. Several kinds of empirical evidence areincorporated into a text planning process that aims atexploiting conversational implicature, so that a mostsuitable portion of the plan can be selected for beinguttered explicitly. This way, our method adds todiscourse planners based on Rhetorical StructureTheory (RST) the ability to omit easily inferableinformation. Thus, it overcomes one of the mainshortcomings of RST. In the course of this process,rules anticipating user inferences are invoked todetermine contextually justified derivability ofinformation. In this manner, text variants can becomposed on the basis of a text plan entailingannotations about the inferability of pieces ofinformation. Moreover, pragmatically-motivatedpreference criteria can be used to choose amongseveral plausible variants. The model is formulated ina reasonably domain-independent way, so that the rulesexpressing aspects of conversational implicature canbe incorporated into typical RST-based text planners.