A Re-Evaluation of Machine-Generated Abstracts

Abstract

Twenty-five subjects were divided into five matched groups and each group was given a different type of lexical indicator of content. The purpose was to determine their ability to answer a 72-question criterion test and determine the relevance of the documents furnished relative to that task. The results indicated that there were no major differences between the five groups in their ability to select relevant documents. The group utilizing complete text material, however, obtained a significantly higher score on the examination. © 1960, Human Factors and Ergonomics Society. All rights reserved.