The Representation of Meaning in Memory (PLE: Memory)

Walter Kintsch

• Publisher: Taylor & Francis
• Year: 2014
• Print ISBN: 9781848723757
• eBook ISBN: 9781317744887
• Edition: 1
• Topic: Psychology
• Subtopic: Cognitive Psychology & Cognition
• Index: Psychology

Originally published in 1974, this volume presents empirical and theoretical investigations of the role of meaning in psychological processes. A theory is proposed for the representation of the meaning of texts, employing ordered lists of propositions. The author explores the adequacy of this representation, with respect to the demands made upon such formulations by logicians and linguists. A sufficiently large number of problems are encompassed by the propositional theory to justify its use in psychological research into memory and language comprehension.

A number of different experiments are reported on a wide variety of topics, and these test central portions of this theory, and any that purports to deal with how humans represent meaning. Among the topics discussed are the role of lexical decomposition in comprehension and memory, propositions as the units of recall, and the effects of the number of propositions in a text base upon reading rate and recall. New problems are explored, such as inferential processes during reading, differences in levels of memory for text, and retrieval speed for textual information. On the other hand, a study of retrieval from semantic memory focusses on a problem of much current research. The final review chapter relates the present work to other current research in the area at the time.

1

ORIENTING ATTITUDES

In recent years cognitive psychology has shaken off the artificial limitations of its subject matter and methodology that had impaired its progress. In focussing their attention once more on the truly important and interesting problems in the field, psychologists are now being forced to disregard the traditional borders of their discipline. Indeed, it is impossible to restrict the study of memory, thought, and language to psychology proper. To do so means cutting off one’s investigation from rich bases of knowledge that have been accumulated by linguists and philosophers throughout history. To be sure, linguistics and, even more so, philosophy have their own methods and their own problems, and their results cannot simply be incorporated into psychology. But though these disciplines will certainly remain separate, the cognitive psychologist can neglect the knowledge that is available from neighboring disciplines only at the risk of letting his own work become provincial and trivial.

It is not enough to describe behavior in certain well-controlled laboratory situations; such descriptions must be powerful and general enough to deal with a much broader range of observations about cognition and language derived from other disciplines. Chomsky (1957) was entirely right when he claimed that, on the basis of purely linguistic considerations, certain requirements can be set up which a psychological theory must be able to meet. One could debate about the exact nature of these requirements, but there can be no doubt about the reality of these constraints. Attempts to begin the psychological study of language without inquiring what others have to say about the topic from their own, nonpsychological standpoints are doomed to failure. The neglect with which Skinner’s Verbal Behavior is treated today attests to this point. Indeed, the constraints that sister sciences impose upon psychological theory should not be viewed as foreign intrusions into psychological territory, but welcomed as a tool to reduce significantly the all too numerous alternative hypotheses that are usually available to the psychologist. When we know as little about cognitive processes as we do now, we have every reason to grasp at every straw that is offered.

Psycholinguistics is a relatively new subarea of cognitive psychology and has developed rather curiously. It is probably fair to say that during the 1960’s, a period of rapid growth for psycholinguistics, experimental psychology proper and psycholinguistics managed to avoid almost all interaction. Fillenbaum (1973) claims that “As a matter of historical fact, the concerns of the student of verbal learning and memory and those of the psychologist of language have been, and are different [p. 79].” In support he cites, among others, the almost complete lack of overlap in the references of two representative review articles, “Memory and Verbal Learning” by Tulving and Madigan (1970) and “Psycholinguistics” by Fillenbaum (1971). Fillenbaum is, of course, correct historically; students of verbal learning and memory have for a long time disregarded the new psycholinguistics, and the psycholinguists themselves were often merely interested in testing in the psychological laboratory certain linguistic theses (as for example the psychological reality of sentence transformation), with little regard for such mundane matters as short- and long-term memory. Neither party profited from this mutual neglect, and I believe that times are changing in this respect, too. Psycholinguistics is changing its character, and the memory and learning fraternity is beginning to realize the fascinating possibilities of the new subject matter. The recent publication of Anderson and Bower’s Human Associative Memory must be a milestone in the integration of psycholinguistics into cognitive experimental psychology. That this is a broadly based development, and not a passing fashion, can be appreciated by thumbing through recent issues of the Journal of Verbal Learning and Verbal Behavior. There is a definite trend there in terms of the preferred experimental material: The 1950’s were still dominated by the nonsense syllables (with paired-associate learning as the most suitable experimental design), the 1960’s were characterized by the use of word lists (with free recall replacing paired-associate learning in the psychologists’ favor), while the present decade is witnessing a shift to even more complex learning materials. At present, we have reached the point where lists of sentences are being substituted for word lists in studies of recall and recognition. Hopefully, this will not be the endpoint of this development, and we shall soon see psychologists handle effectively the problems posed by the analysis of connected texts.

The research reported in the present volume is a product of this time of transition. The theory aims at generality and at encompassing as broad a range as possible of phenomena relevant to problems of language processing and memory. In formulating psychological explanations of language and thinking, a conscious attempt has been made to keep in mind the constraints imposed by other disciplines, while at the same time maintaining the traditional standards of experimental psychology. Linguistic and logical scholarship, as well as computer programs, have their uses in guiding the psychologist’s thinking, but in the final analysis experimentation under controlled circumstances is necessary to sort out the bewildering array of observations we are faced with. Even if we cannot tell at this particular stage of investigation which particular experiment might be the best and the most fruitful one in the long run, it is nevertheless possible to perform informative and useful experiments.

The point we have just made, that one should not conceive psychological theories so narrow as to exclude linguistic observations about language, can also be turned around: The new theories must be able to deal with the old facts. It would be ill advised to forget that cognitive psychology has available a very solid body of experimental data, the list-learning research of the last decades. Whatever the limitations of this research, it is well-established experimentally and provides a valuable proving ground for any theory of learning and memory.

The approach advocated here conflicts with several widely held views in both psychology and linguistics. It has no use at all for the competence-performance distinction which, in my opinion, is merely an excuse for both the linguist and psychologist to justify the neglect of each others’ findings. Chomsky (1965) makes a “… fundamental distinction bewteen competence (the speaker-hearer’s knowledge of his language) and performance (the actual use of the language in concrete situations) [p. 4].” Competence is “… unaffected by such grammatically irrelevant conditions as memory limitations, distractions, shifts of attention and interest, and errors in applying knowledge [p. 3].” Performance reflects competence only in indirect ways: Knowing the linguistic rules to generate a sentence will tell us very little about how people actually generate a sentence. This strict separation permits the linguist to deal with convenient abstractions, uninhibited by psychological reality, and it provides the psychologist with the facetious argument that linguistic theories have nothing to do with processes anyway. As long as a linguistic theory is strictly a competence theory, it is of no interest to the psychologist. Indeed I doubt that it should be of much interest to linguists either, but that is for them to decide. The linguistic theories discussed below are not of that type, and process notions, appeals to a person’s beliefs, or to memory processes, are inextricable parts of many of them. Equally to be rejected is the neat separation between structure and process, upon which many psychologists insist today. Such a separation is both impossible and misleading. There is no structure apart from psychological processes; structure is the result of processes. Two examples must do for the moment to illustrate the close interconnection between process and structure in the present approach. In Chapter 2, the set of all semantically acceptable sentences is discussed; this set is not something fixed, something that exists separately from process considerations. It is the result of psychological processing, of applying production rules to information stored in memory. The set of all acceptable sentences is a potential set that could be generated from a given semantic memory, if certain processes take place. Thus, what is a semantically acceptable sentence will differ for different individuals as well as for the same individual at different times. Similarly, in Chapter 3, (psychological) implication is discussed not as a fixed structural relationship that holds between sentences, independent of the beholder, but as the consequence of certain psychological processes performed by a given individual. Thus, whether sentence A implies sentence B is not something fixed for all time, but depends upon the occurrence of specifiable psychological processes; indeed, it is possible that for a given individual at a given time there would be no implicational relationship between two sentences, while under somewhat different circumstances, for the same individual, such a relationship might be constructed between the same two sentences. Structures are what is being generated, but they have no existence of their own.

The distinction between episodic and semantic memory, on the other hand, has proven very valuable in the present context. Semantic memory refers to a person’s knowledge; episodic memory, to his store of experiences. Obviously, most or all knowledge is derived in some way from personal experiences at one time or another, but these experiences have become depersonalized, at least in part, separated from their original context, and are thus much more broadly useful than specific, personal, context-bound experiences. This distinction was introduced by Tulving (1972), and the reader is referred to the original discussion for a more detailed treatment, as well as to some complementary remarks in the following Chapters. No sharp line can be drawn between episodic memory and semantic memory, but nevertheless the two separate terms come in handy in sorting out various aspects of memory. If for no other reason, Tulving’s distinction is valuable because it helps to focus attention on some problems that have been traditionally neglected in both verbal learning research and psycholinguistics. These are the problems concerned with the processing, retention, and retrieval of knowledge, as distinguished from personal, context-bound memories. Surely, this is one of the reasons why psychological research (as distinguished from programmatic theories without appreciable bases in experimental cognitive psychology) has been so notoriously ineffective in directing educational practice. Most of the experimental research concerning memory has never really dealt with problems of the acquisition and retention of knowledge, but with episodic memory, which is not at all the problem of interest in education. Simply replacing the words with sentences in our experiments will make the research no more relevant to education than it was before (or only very little so). An educational technology squarely based upon psychological research needs research concerned with problems of knowledge. Even the most sophisticated understanding of episodic memory phenomena would be of little use in itself.

Historically, knowledge was considered an important topic for psychologists until the last few decades. What is called semantic memory today was at one time referred to as the apperceptive mass. The term is from Leibniz, but was brought into prominence in psychology by Herbart (1816). Herbart’s fundamental insight was that new learning presupposes the availability of memory structures with which the new information can be connected. Learning does not consist in the passive recording of new information, but in apperception: The new information is connected in a meaningful way with a person’s background of experience and knowledge. Apperception and apperceptive mass were enormously influential concepts in education and psychology throughout the 19th century. Today they are relegated to the history of psychology.

There is no reason to try to reintroduce these terms (partly because apperception underwent some subtle but confusing changes in meaning between Leibniz and Wundt, who used it more in the sense of the present-day term “attention”). But it is important that we study the subject matter and the problems that were at one time designated with these terms. In order to do so, we need some sort of a language to talk about “knowledge.” The approach taken here is to represent knowledge in terms of propositions. A proposition is a k-tuple of word concepts, one serving as a predicator and the others as arguments. Word concepts are abstract entities to be defined below; words are used to express word concepts verbally. Thus, a distinction is made between different levels of representation. Memory may refer to any of these levels, or to a combination thereof. For example, a sentence may be represented physically as an acoustic pattern, graphically as a string of letters, linguistically as a sequence of words, and conceptually in terms of its meaning. The last representation is the propositional one, and the study of the properties of this level of representation in memory, both theoretically and empirically, is the main concern of this book.

The question arises whether propositional representations as conceived here are indeed at the proper level of analysis for the study of language and thought. The problem is an old one, but it still resists satisfactory solution. None can be provided here either, but it is necessary to point out where and how the present approach fits into the range of solutions proposed by others. The problem can be formulated as “What is an idea?” or, more precisely, “How is an idea to be represented?” It is suggested here that propositions represent ideas, and that language (or imagery) expresses propositions, and hence ideas. Thinking occurs at the propositional level; language is the expression of thought. This is a squarely anti-behaviorist position; arguments against this position have been frequently made, with especial brilliance by Kantor (1936). There is no need to refute these arguments again, since this discussion has left many well-known public records (see, for instance, Humphrey, 1963). A more acute problem is posed today by attacks upon propositional theories from the opposite camp. It is sometimes claimed that propositions do not represent ideas, but that ideas are at a deeper level of analysis than propositions. Ideas are claimed to be unarticulated, pre-propositional schemes of thought; propositions specify ideas, express them in some sense, but rob them of their all-encompassing nature. It is difficult to state this problem clearly because of its very nature, but it has been a recurring question in the study of thinking. Humphrey (1963, p. 260) presents a good historical discussion of the early work concerning this question. He distinguishes various levels of representation of a sentence: the actual choice of words (today we would call this the surface form of the sentence), the sentence schema (a Jamesian term, which has found a contemporary parallel in today’s linguistic deep structure), the well-articulated sentence meaning (which is what the propositional representation is concerned with), and finally a state of consciousness from which all the other representations are ultimately derived.

What is the evidence (other than introspection, which might as well be neglected here) for the existence of pre-propositional schemes of thought? Consider, for example, any sufficiently complex event and the propositional representation of its description. Several propositional paraphrases that are all representations of the same event can be devised. In Chapter 2, in the section on discourse analysis, such problems are treated in some detail. A given propositional representation may be complete and sufficient to the extent that everything else about the event in question may be deduced from it, but it is not unique: There may be another way of representing the same state of affairs, just as complete and just as sufficient as the one chosen. Hence, some psychologists have assumed a deeper, more basic, level of representation underlying the propositional one.

The argument just presented is, of course, not very precise. Indeed, if it could be stated with precision, we would probably be able to decide the issue one way or another. As it is, it is difficult to argue on principle against the existence of pre-propositional, unarticulated thoughts. One may maintain, however, that no matter how this problem is eventually decided, it makes little difference for all practical purposes. No one has clear enough notions to permit a serious, scientific investigation of such thought schemes at the present time. We can, and should, do something about the study of propositional representation, on the other hand. If they are not the deepest, most basic elements of thought, it appears almost certain that they are the proper units of analysis for one level of representation. Clearly, thoughts do not stay unarticulated forever and must be formulated precisely during some stage of processing. Thus, whatever we find out about the characteristics of the propositional representations would not be wasted, even if that representation is not the most basic one. At this time so little is known about pre-propositional thought schemas that it seems quite defensible to disregard them altogether and to assume, as a working hypothesis, that ideas are represented propositionally.

Before leaving this discussion, one possible misunderstanding must be anticipated. Whatever pre-propositional thought schemas might be, they are not images. Images pose the same problem as propositions: Suppose a person has an image of an object which is sufficient to specify the object in all relevant detail (such an image would be more like a map or a plan than a photograph, presumably). The problem is that there may be alternative maps or plans that would be just as good as the original one. There are as many imagery paraphrases for a scene as propositional paraphrases. Imagery and propositional representations are probably more similar than is commonly admitted. There is good evidence (Pylyshyn, 1973) that imagery is based upon propositional representations, much as verbal expressions are, the differences between the two lying primarily in the way these representations can be processed and accessed in memory.

PART I

THEORY

Understanding a sentence means
understanding a language.

Ludwig Wittgenstein

2

A PROPOSITIONAL THEORY FOR THE REPRESENTATION OF MEANING IN KNOWLEDGE AND MEMORY

The theory to be outlined here is intended to serve as basis for the development of psychological process models. These models should, eventually, provide an account of such topics as the use and acquisition of knowledge and the comprehension and memorization of text, even though only small steps toward the development of such models can be undertaken at the present time. As a consequence of these goals, the formalism for the representation of meaning that is used in the theory must be general enough to be usefully employed in all these different situations. The formalism itself is motivated by various linguistic and logical considerations, which will be detailed in this and the following chapter.

Whenever the term meaning is used here it is used in a psychological sense; we are only concerned with the psychological representation of meaning, that is, the meaning of a concept is defined only with respect to a particular semantic memory. A concept is defined through its relationship with other concepts in semantic memory, but each semantic memory is subjective. Each person’s knowledge consists of only a (small) subset of the culturally shared knowledge. We shall neglect these individual differences whenever possible and talk about semantic memory in general terms, though it must be remembered that this is merely an idealization, useful for o...

Table of contents

• Cover Page
• Half Title page
• Title Page
• Copyright Page
• Orginal Title Page
• Orginal Copyright Page
• Contents
• 1 Orienting Attitudes
• Part I Theory
• Part II Experimental Investigations
• References
• Author Index
• Subject Index