TCD-CS-93-01..Jane Howe, M. & Hederman, L.

Text Analysis in Office Systems

January 1993.

Abstract

Office documents are increasingly available in electronic form. In limited domains and for well-defined tasks natrual language processing techniques can be used to automate office tasks which are based on the contents of documents. This report describes a proposed complaints processing system and discusses how key facts might be extracted from complaint letters to feed such a system. A combination of different approaches is used to extract the useful information, using knowledge about the physical layout of a letter, linguistic knowledge about sentence structure, and domain-specific knowledge about the classes of events describes in complaint letters.


Technical Reports Home Page