ABSTRACT
Overview
Research and theory on comprehension grows from two different traditions. Some of the work comes from the psycholinguistic tradition and is a logical continuation of the work described in the previous chapter. Some emerges from the artificial-intelligence-plus-experimental-psychology tradition begun by Collins and Quillian and described in our chapter on Semantic Memory. This chapter describes the psycholinguistic approach to comprehension.
The Concept of the Synthesized Code
As interest turned away from syntactic theories and toward semantic issues, the field of psycholinguistics lost the unifying force of Chomsky's theories. Most contemporary researchers are interested in language comprehension, but no powerful model guides research. Several issues have emerged that can be grouped roughly into issues of form and issues of content. Some research concerns the form of a message when it has been understood, and some issues concern the content of the understood message. One relatively comprehensive model of conversational comprehension has been presented.
Information-processing psychologists believe that complex mental processes like comprehension involve a large number of unseen internal events that occur between input and observable response. These events include speech perception, parsing, word look-up, and many other things we have not yet thought of. We call these events, as a group, input synthesis. The end product of input synthesis is a representation that is compatible with the contents of permanent memory. We call this end product the synthesized code. Much research seeks to explain the form of the code or its contents.
The Form of the Synthesized Code
The major controversy regarding the form of the synthesized code is whether it is 406best characterized as a recoded linguistic string (a proposition) or as an analogue of visual experience (an image).
Propositional Theories Clark has been the major spokesman for the view that the synthesized code is propositional. His first theory (Clark, 1969) relied heavily on Chomsky's ideas and was intended to account for the way people solve reasoning problems. Later versions of the theory made less use of Chomsky's work and were validated by means of the sentence-picture verification task. All these theories conceived of the understanding process as recoding the literal input sentence into an abstract propositional format. Clark's interests have changed, but a descendant of his model has been proposed and defended by Carpenter and Just (1975).
Dual Coding Theory The main advocate of an imagery code has been Paivio. His position is not a "pure" imagery position but holds that synthesizing an imagery code is the normal strategy for understanding and recalling sentences. But when a sentence is abstract, images may not be available and a verbal code may be all that the comprehension process can extract. This dual coding theory (Begg & Paivio, 1969) has been widely challenged, primarily because the experiments that support the theory use abstract and concrete sentences that are not equally understandable in the first place. This difference in comprehensibility may account for the pattern of results that is taken as support for dual coding. However, the relationship between concreteness and comprehensibility remains an important fact to be explained.
Content of the Synthesized Code
A literal account of the contents of an input sentence cannot explain the knowledge a person has as a result of hearing the sentence. Several lines of research converge on this claim.
Constructive Synthesis People do not accurately distinguish sentences actually heard in an experiment from plausible inferences, recombinations, or paraphrases of those sentences.
The Role of Context The meaning given to a word depends on the sentential context in which it is embedded. This is a problem for theories of word meaning, because it suggests that there may be no core of meaning that is always salient when the word is heard. On the other hand, it is implausible to suppose that word meanings are totally flexible. Context also supplies meaning to deictic terms such as this and here.
The Role of Previous Knowledge People bring their previous knowledge to the comprehension process in several different ways.
Inference Sentence-recognition experiments show that inferences are as familiar to subjects as the actually presented sentences, although they can distinguish actual sentences from plausible inferences if asked immediately-
Presuppositions Presupposed information is not evaluated for truth as much as asserted information. This makes it possible for the careful phrasing of questions to manipulate people's memory for events and to cause them to reconstruct events that did not happen.
Nonliteral Comprehension Sometimes what people extract from a message is not part of its literal content at all. Peculiar sentences are reinterpreted to reflect "what the speaker really meant," and conversational requests are easily understood. However, there is some evidence that 407the literal meaning of conversational requests is computed at some point in the comprehension process.
A Model of Conversational Comprehension
Clark (1976) has suggested that speakers and hearers work within the confines of a sort of contract. Assuming that the speaker will adhere to the contract permits an efficient comprehension strategy in which the hearer locates information already in memory and incorporates new information into it. The theory is very new, but it accommodates much of the data reviewed in this chapter.
