The author’s intention is to present a computational system called NEXUS, which is able to construct representations of narrative texts with the aid of a dictionary containing a certain number of event/state concepts.

Let me point out at the beginning of the review: this intention has been successfully realized, and in a clear and understandable style. NEXUS is described in a way that one can see how it works, where its limitations are, and how it can be embedded in larger systems.

NEXUS is to be a module of proposed, more complex systems (i.e., the input and the output of NEXUS are well-defined for better processing within such systems). This feature allows us to describe NEXUS without other possible components and to run tests with manually established input (the input text first has to be converted into case notation). Possible applications (such as the modules SUM and QUEST, which use NEXUS’ output) are only briefly mentioned; they are, however, described in detail in the author’s dissertation, on which the whole paper is based [1].

The author’s hypothesis was “(1) that text is composed of structured chunks of concept coherent event/state descriptions, (2) that it is possible to collect together events without explicitly working out all the details of their semantic (e.g., causal) inter-relationships, and (3) the initial grouping and structuring of a text could be accomplished by augmenting case relationships with a handful of inter-event relations.” This was to be realized with the aid of a dictionary of event/state concepts which can be thought of as a semantic net. The relations between the concepts are defined as their relative proximity within this net, which “allows NEXUS to represent the similarities between two pieces of texts that use the same event/state concepts, but have different causal interpretations.”

The author first gives detailed examples of the structure of the dictionary, illustrating the sort of relations being used within, and then describes how NEXUS produces the coherence representations. Included are examples of the programs, especially of a Horn clause theorem prover, the flow of control, and various pieces of texts having been processed by NEXUS. This allows the reader to understand the system and to follow the author’s conclusions. He states that “the representation produced by NEXUS is . . . an initial structuring of the text. . . . It provides a computational handle on the text by analyzing it into structured coherent chunks of event descriptions,” showing NEXUS as a step within a row of processes beginning with the computerized transformation of a given text into a case representation and ending, for example, with the establishing of story trees.