Paper: A Discourse-Aware Graph-Based Content-Selection Framework

ACL ID W10-4202
Title A Discourse-Aware Graph-Based Content-Selection Framework
Venue International Conference on Natural Language Generation
Session Main Conference
Year 2010

This paper presents an easy-to-adapt, discourse-aware framework that can be utilized as the content selection compo- nent of a generation system whose goal is to deliver descriptive texts in several turns. Our framework involves a novel use of a graph-based ranking algorithm, to itera- tively determine what content to convey to a given request while taking into account various considerations such as capturing a priori importance of information, convey- ing related information, avoiding redun- dancy, and incorporating the effects of dis- course history. We illustrate and evaluate this framework in an accessibility system for sight-impaired individuals.