This paper examines the effect of abstracting instructions on free-text retrieval systems which rely on abstracts. The approach used in the paper is to consider each topic in an abstracting policy, examine the implications of options for that policy, and then apply what is known about information retrieval to predict the behavior of the system when that option is chosen.

Although no empirical data are given to support the conclusions of this study, the conclusions are carefully thought out and based on well-documented research.

The effects of instructions related to type, length, content, and language are analyzed in this way. The author convincingly argues that the effects of type and length instructions are relatively unpredictable, while the effects of content and language descriptions are indeed predictable.