The conventional automatic document abstraction systems are based on an identification of document sentences believed to reflect document content, followed by the extraction and stringing together of the most important sentences. The author of the paper under review remarks that in such extracted sentences there is no coherence between the individual sentences. He proposes a manual (or rather intellectual) document-encoding process designed to identify anaphoric expressions in the texts, together with the corresponding full referents of each anaphoric expression. This makes it possible to later use the encoded texts to construct document abstracts of various kinds, in which each concept is fully expressed when it first appears in the abstract instead of being identified by a possibly ambiguous referent.
There are two main problems with the work: first, the human encoding and identification of anaphoric expressions appears laborious and is not mechanizable under current conditions; and second, no indication is given about what kind of abstract is useful for the users under particular circumstances. One must conclude that this effort is an academic exercise rather than a really practical approach to an existing problem.