ABSTRACT
In the preceding chapters the major component processes of language understanding have been described. These processes can be divided into three groups: word-level processes, syntactic processes, and what Forster (1979) has called message-level processes – that is to say, those processes that compute the various aspects of the meaning of a discourse or text. The first set of processes determines, on the basis of perceptual and contextual inputs, what words have been presented to the understanding system; the second set computes the structural relations between the words; the last set uses these structural relations, together with context, to ascertain the import or significance (Johnson-Laird, 1977b) of what has been said. These three sets of processes can be ordered with respect to one another in the following sense. All the words in a sentence could, in principle, be recognized before any syntactic structure has been assigned, but a complete syntactic analysis cannot be generated before any words have been recognized. Similarly, parsing could precede the interpretation of a sentence or text, but interpretation cannot proceed in the absence of any syntactic information. However, these logical considerations place only weak constraints on how the three sets of processes work together. Once the lower-level processes have produced partial analyses, those at a higher level can begin their computations, and their outputs may be fed back to lower levels.
