AUTOMATIC ABSTRACTING C107-3U1
Abstract
A series of additions and refinements to the previous RADC contract study is presented. During this contract, major results were produced: an operating system, and a research methodology. The operating system produces automatic abstracts via programs written for the IBM 7090 and the IBM 1401. The programming involved the preparation of an Edit Program which inputs the text of documents, a Cue Dictionary Program which inputs a fixed word list, and an Abstracting Program which selects and outputs sentences of the document. The research methodology proceeds from linguistic analysis of documents comprising a sample library, the compilation of dictionaries, the formulation of abstracting rules which are applied to new documents of an experimental library, and concludes with testing and evaluation of the final program and dictionaries on documents of a test library. Examples are given of abstracts produced by the four basic methods: Cue method, Key method, Title method, and Location method. In addition, a combined method using Cue-Title-Location is exemplified as the preferred method. Conclusions resulting from this study and recommendations for future research are presented.
Document Details
- Document Type
- Technical Report
- Publication Date
- Feb 02, 1963
- Accession Number
- AD0406155