Knowledge Based Text Generation
Abstract
This report develops a theory of text generation and describes an implemented computational model of this theory. The theory attempts both domain independency at the knowledge level and language independency at the linguistic level by drawing and expanding upon previous work in discourse schema and grammatical relations, respectively. The implemented system, GENNY, generates coherent texts by employing discourse strategies (which occur in human produced text) in parallel with pragmatic constraints (e.g., focus and context). The report begins with an introduction and summary of the research performed. This is followed by a survey of the text generation literature which places GENNY in the context of past language production research. Next, the motivation for the theoretical position adopted is discussed followed by detail of the theory on a knowledge, pragmatic, semantic, relational, and syntactic level, with illustration of the practical implementation throughout. Results of GENNY's test production from two frame knowledge bases (neuropsychology and photography) are then presented together with preliminary interlingual test results (English and Italian). GENNY is evaluated with respect to state-of-the-art generators and is shown to be equivalent and, in some respects, superior, in competence and performance. In conclusion, the contributions and limitations of the system are discussed for further development are suggested. Keywords: Natural language generation. (kr)
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 1989
- Accession Number
- ADA214273
Entities
People
- Mark T. Maybury
Organizations
- Rome Laboratory