The Predication Semantics Model

This paper presents and tests the predication semantics model, a computational model of text comprehension. It goes beyond previous case grammar approaches to text comprehension in employing a propositional rather than a rigid hierarchical tree notion, attempting to maintain a coherent set of propositions in working memory.

The authors’ assertion is that predicate class contains semantic information that readers use to make generally accurate predictions of a given proposition. Thus, the main purpose of the model-which works as a series of input and reduction cycles-is to explore the extent to which predicate categories play a role in reading comprehension and recall. In the reduction phase of the model, the propositions entered into the memory during the input phase are decreased while coherence is maintained among them. In an examination of the working memory at the end of each cycle, the computational model maintained coherence for 70% of cycles. The model appeared prone to serial dependence in errors: the coherence problem appears to occur because (unlike real readers) the simulation docs not reread when necessary.

Overall, the experiment suggested that the predication semantics model is robust. The results suggested that the model emulates a primary process in text comprehension: predicate categories provide semantic information that helps to initiate and control automatic processes in reading, and allows people to grasp the gist of a text even when they have only minimal background knowledge. While needing refinement in several areas presenting minor problems—for example, the lack of a sufficiently complex memory to ensure that when the simulation of the model goes wrong it does not, as at present, stay wrong for successive intervals—the success of the model even at the current restrictive level of detail demonstrates the importance of the semantic information in predicate categories.

Working Paper No. 78(425.10 KB)