Referent Grammar In Text Generation

Abstract

A text may be seen as the result of input from the external world and human internal aspects on it. The paper shows how the same event — a submarine following a trawler — may give rise to different types (genres) of text, e.g., a report on the facts, an evaluation of the facts, a summary of the most important features against the background of previous events, or a mixture of some or all of these.

As is well known, it is difficult to teach a computer how to generate interesting and varying text including more abstract, evaluative and summarizing comments. The paper shows how different predicates, e.g., be, go, sail, follow, attack give information of varying interest and how the computer may calculate the news value of an event on the basis of the amount of interest of the predicates and the objects involved. The bulk of the paper demonstrates how Referent Grammar (RG) — a generalized phrase structure grammar with built-in referents — offers a convenient way of keeping track of both the sentence and the discourse referents.