Comprehension and Recoding-Time of Transformed Sentences

Abstract

Two groups of subjects, psychiatric patients and normals, were required to demonstrate their comprehension of sentences by reconstructing the situations which the sentences described. Twelve sentence-types were employed which differed in syntactic structure. The recoding-time (T) and the number of propositions correctly translated (C) were taken as primary measures of performance. Another performance-measure called " information correctly translated " (S) was calculated according to four different sets of assumptions (models). The " parenthetical load " of each sentence-type (PL) was calculated in terms of interpolated words per sentence according to four different models and in terms of interpolated semantic information per sentence according to eight different models.

The results suggest that S should not be interpreted as a measure of informational work done in recoding a sentence, since it correlates positively with C and negatively with T. The opposite is the case with the PL. The PL of a sentence, however calculated, appears to be an overall measure of its inherent difficulty of recoding and may be likened to " input noise" which interferes with the rapid, accurate recoding of its semantic information content. Informational measures of PL compare well with measures based on counting words.

A comparison of the informational models used to calculate PL suggests that in this task, performance is dominated by a " compiling " process in which the sentence is scanned from left to right and its constituent propositions are decoded in the same order as they are extracted from the sentence with the aid of syntactical transformations. As the beginning of each new clause or proposition is encountered in the scan a new " storage location " is opened, which is only closed to further input when the end of the clause or proposition is recognized by encountering a " terminator ", which may be a semantically redundant word.

All these conclusions carry more weight for the patients than for the normals, for whom the different variables span a narrower range and yield non-significant correlations in many cases.