Enhancing Readability of Automatic Summaries by Using Schemas

The automatic summaries produced by extraction processes have been subject to several critics because of the quality of their textual substance (i.e, lack of cohesion and coherence). In this paper, we propose the use of summary-schemas conceived on the base of an empirical study of the argumentative structure of scientific articles and their author abstracts. These schemas enclose guidelines which help to achieve four properties to the summary textual material: completeness, non-redundancy, canonical organisation, and connectivity. These properties are achieved by means of three operations (addition, suppression and permutation) applied to the initial extract by agents that use various resources and interact with source text.