A Rule-Based Approach to Evaluating Importance in Descriptive Texts

A RULE-BASED APPROACH TO EVALUATING IMPORTANCE IN DESCRIPTIVE TEXTS Danilo Fum(*), Giovanni Gulda(?), Carlo Tasso(~) Isllmm di Matemadca, Informatica • Slstemistica Universi~ di Udlne Udlne, Italy ABSTRACT Importance evaluation is one of the most challenging problems in the field of text processing. In the paper we focus on the notion of importance from a computational standpoint, and we propose a procedural, rule-based approach to importance evaluation. This novel approach is supported by a prototype experimental system, called importance evaluator, that can deal with descriptive texts taken from computer science literature on operating systems. The evaluator relies on a set of importance rules that are used to assign importance values to the different parts of a text and to resolve or explain conflicting evaluations. The system utilizes world knowledge on the subject domain contained in an encyclopedia and takes into account a goal assigned by the user for specifying the pragmatic aspects of the understanding activity. The paper describes the role of the evaluator in the frame of a larger system for text summarization (SUSY); it illustrates its overall mode of operation, and discusses some meaningful examples. I. INTRODUCTION Text understanding has received increasing attention in recent years. A major problem in this area is that of importance evaluation: not all the components of a sufficiently large and structured piece of text are equally important for the reader, and humans are able to evaluate the relative importance of the parts of the texts they read. This issue has been faced so far only in an indirect way in the literature on discourse structure (Kintsch and van Dijk, 1978; van Dijk and Kintsch, 1983), summarization (Lehrert, 1982 and 1984; Wilensky, 1982; Hahn and Reimer, 1984), and inference (Schank, 1979). Moreover, several studies in the field of summarization (e.g.: Schank, 1979 and 1982; Lehnert, 1982 and 1984; Wilensky, 1982) have mostly been concerned with narrative texts (stories), and it is not at all obvious that the approaches which proved successful in this area (') also with: Laboratorio di Psicologia E.E., Universita' di Trieste, Trieste, Italy (t) also with: Milan Polytechnic Artificial Intelligence Project, Milano, Italy (~) also with: CISM International Center for Mechanical Sciences, Udine, Italy could be applied to descriptive texts as well. Expository prose has its own specific features (Graesser, 1981), and it seems to require different understanding processes, different summarization skills, and different cognitive models (Lehnert, 1984). Work on the problem of understanding and summarizing expository prose is still at the very beginning (Hahn and Reimer, 1984). In this paper we focus on the notion of importance from a computational standpoint, and we propose a rule-based approach to importance evaluation. This research is part of a larger project aimed at developi ng a system for understanding and summarizing descriptive texts (SUSY, a SUmmarizing SYstem), which is in progress a~-~Te University" of SUSY proposes an approach to descriptive text understanding and summarization (Fum, Guida, and Tasso, 1982, 1983, and 1984) in which the process of representing the meaning of a natural …

Content truncated to 500 words