Natural Language Generation.
Abstract
The goal of natural language generation is to replicate human writers or speakers: to generate fluent, grammatical, and coherent text or speech. Produced language, using both explicit and implicit means, must clearly and effectively express some intended message. This demands the use of a lexicon and a grammar together with mechanisms which exploit semantic, discourse and pragmatic knowledge to constrain production. Furthermore, special processors may be required to guide focus, extract presuppositions, and maintain coherency. As with interpretation, generation may require knowledge of the world, including information about the discourse participants as well as knowledge of the specific domain of discourse. All of these processes and knowledge sources must cooperate to produce well-written, unambiguous language. Natural language generation has received less attention than language interpretation due to the nature of language: it is important to interpret all the ways of expressing a message but we need to generate only one. Furthermore, the generative task can often be accomplished by canned text (eg error messages or user instructions). The advent of more sophisticated computer systems, however, has intensified the need to express multisentential English. (sdw)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 01, 1988
- Accession Number
- ADA197321
Entities
People
- Mark T. Maybury
Organizations
- Air Force Institute of Technology