ABSTRACT
The paper reports on a series of experiments on the comprehension of written sentences or passages of prose. The main question addressed concerns the amount of syntactic computation performed during reading, and the use of the results of this computation for comprehension of the written material.
The approach taken in the present study assumes that in normal fast reading the reader is likely to rely for comprehension on various heuristics, which allow a first and efficient use of the different cues available in the text. But if this is true, does the reader need to compute a full syntactic representation of the material read? It will be argued that during reading for comprehension the reader normally executes a full syntactic analysis of the material, although the system can rely on other sources of data for the comprehension process.
The position taken here distinguishes computation, that is, full processing of the various sources of information available in the text, from the use of this information for message interpretation. The paper offers some evidence on the notion that the results of the syntactic computation are redundant with respect to the use of this computation made during reading, and that they can be used as a kind of backup device when the text becomes difficult, incoherent, or pragmatically unplausible.
The first experiment provides evidence for the notion of automatic computation of the syntactic structure, which, however, does not seem to be fully exploited for text comprehension. The second experiment offers evidence favouring the hypothesis that the results of syntactic computation become more important with pragmatically unplausible texts. The third experiment shows that readers differ with regard to their sensitivity to the properties of the syntactic structure.
