• DocumentCode
    972630
  • Title

    A new level of language generation technology: capabilities and possibilities

  • Author

    Hovy, E.H.

  • Author_Institution
    Inf. Sci. Inst., Univ. of Southern California, Marina del Rey, CA, USA
  • Volume
    7
  • Issue
    2
  • fYear
    1992
  • fDate
    4/1/1992 12:00:00 AM
  • Firstpage
    12
  • Lastpage
    17
  • Abstract
    The different tasks and types of language generation systems are discussed. There are four types of generation systems: canned-text, template, phrase-based, and feature-based systems. Canned text is the simplest way of producing output and is used in the vast majority of software. Templates, the next level of sophistication, are used when a single message must be produced multiple times with slight alterations. Phrases are generalized templates and can be developed to considerable complexity. Features are the limit point of phrase generalization. Each possible minimal alternative of expression is represented by a single feature. These systems can be applied to produce multiparagraph texts, summaries of numerical information, explanations, help systems, multimedia interfaces, software documentation and machine translation. Penman, one of several general-purpose single-sentence generators that are gradually evolving toward full-scale commercial strength, is discussed.<>
  • Keywords
    linguistics; natural languages; word processing; Penman; canned-text; explanations; feature-based systems; full-scale commercial strength; general-purpose single-sentence generators; generalized templates; help systems; language generation systems; machine translation; multimedia interfaces; multiparagraph texts; numerical information; phrase generalization; software documentation; Airplanes; Artificial intelligence; Computer displays; Databases; Degradation; Expert systems; Information retrieval; Marine vehicles; Power generation; Problem-solving;
  • fLanguage
    English
  • Journal_Title
    IEEE Expert
  • Publisher
    ieee
  • ISSN
    0885-9000
  • Type

    jour

  • DOI
    10.1109/64.129278
  • Filename
    129278