Referring Via Document Parts

Documents in a wide range of genres often contain references to their own sections, pictures etc. We call such referring expressions instances of Document Deixis. The present work focuses on the generation of Document Deixis in the context of a particular kind of natural language generation system in which these descriptions are not specified as part of the input, i.e., when it is up to the system to decide whether a reference is called for and, if so, which document entity it should refer to. We ask under what circumstances it is advantageous to describe domain objects in terms of the document parts where they are mentioned (as in "the insulin described in section 2"). We report on an experiment suggesting that such indirect descriptions are preferred by human readers whenever they cause the generated descriptions to be shorter than they would otherwise be.

• Type: Conference
• Year: 2006
• Where: CICLING
• Authors: Ivandré Paraboni, Kees van Deemter
• Keywords: CICLING 2006, Document, Document Deixis, Natural Language Processing, Present Work Focuses