AUTOMATIC ABSTRACTING EVALUATION SUPPORT.
AMERICAN INSTITUTES FOR RESEARCH PITTSBURGH PA
Pagination or Media Count:
Techniques for producing two types of abstracts of technical literature were developed. One abstract was designed to be a general-purpose abstract which could be used to support a variety of text-mediated tasks. The second abstract was a modification of the first, and was designed to contain only that information necessary to the performance of a single given task. Four types of text-supported tasks were identified--screening, comprehension-retention, fact retrieval, and problem solving. Paper and pencil tests were prepared--the tests designed to serve as analogs of the four tasks. Performance accuracy and performance time on the tasks of subjects using the abstracts were compared to those of subjects using original text. For both college students and professional scientists, it was found that the abstracts effectively served in place of full text in supporting performance on all tasks except fact retrieval. Accuracy loss was slight, while time savings were substantial. The abstracts were more effectively used for screening and comprehension-retention than for problem solving. The task-specific abstracts were superior to the general-purpose abstracts in amount of time saved in performance of the screening and the comprehension-retention tasks. The implications of these results for future work were discussed. Author