Human abstracting will be with us for some time, so abstractor’s assistants are welcome. This paper gives a list of useful features, and reports on a study of the use of a prototype called TEXNET, with emphasis on one feature: the provision of a list of important words (condition 1) or words and phrases (condition 2) extracted from the text of the document to be abstracted based on frequency (eight or higher). The interface consists of three panes: the document text, suggestions (extracted terms), and an empty pane for the abstract. The main issue is the relative performance of the word versus the phrase suggestion list in terms of ease of use and abstract quality.
Unfortunately, the prototype and the study design are so flawed that the results are fairly useless. The interface is a throwback to the early 1980s, well below the standards of even 1995 (when the study appears to have been conducted). The phrase selection and the text editor used are poor. The subjects, and even three of the evaluators of abstract quality, were not abstractors but student volunteers. To find out the extent to which the abstractors use copy-and-paste from document text to abstract, the study relies on indirect measures rather than recording such moves directly. While there is the potential for useful work here, it is hard to fathom why this particular paper was published, particularly as it appears five years after the fact.