Human Associative Memory

John R. Anderson, G. H. Bower

• Publisher: Psychology Press
• Year: 2013
• ISBN:
  9781134916931
• Edition: 1
• Topic: Psychology
• Subtopic: History & Theory in Psychology
• Index: Psychology

Published in 1980, part of the Experimental Psychology series. This book proposes and tests a theory about human memory, about how a person encodes, retains, and retrieves information from memory. This edition contains two major parts. First is the historical analysis of associationism and its countertraditions. This still provides the framework that has been used to relate the current research to an important intellectual tradition. This is reproduced without comment from the original book; historical analyses do not need as rapid revision as theoretical analyses. The second part of the book reproduces the major components of the HAM theory.

1

INTRODUCTION

And I gave my heart to seek and search out by wisdom concerning all things that are done under heaven; this sore travail hath God given to the sons of men to be exercised therewith.

–Solomon

Two years ago, we set out to develop a theory of human memory, a theory which was to span a wide range of mnemonic phenomena. We are now humbled by the immensity of this task; human memory is a complex mental capacity, and our ability to comprehend man's mind appears at times quite limited. But Solomon calls us to the task of understanding, to be “exercised” by its sore travail. And so we tried. In countless hours of conversations, we discussed, proposed, role-played, argued, laughed, cajoled, reasoned, debunked, and just plain talked to one another about the problems of human memory. The time has come for us to commit to print a fraction of the things we have thought about human memory in the hope of helping others to think about this problem–which we consider to be the supreme intellectual puzzle of the century.

The theory of human memory which we will articulate will seem overly ambitious but still terribly programmatic; no one can realize this better than we ourselves. So why bother? What does Psychology need with another fragmentary theory of memory? After all, a long parade of memory theories since Plato's have been offered with great fanfare, hopeful enthusiasm, and persuasive arguments. Most of these were soon consigned to the loneliness of library tombs, accumulating dust to hide their insignificance. A very few of these writings become classics. But no one really believes the classics; they are read only to provide jousting partners for later opponents and voyagers on the seas of the unknown.

It is commonplace that the Zeitgeist in current psychology opposes global theories such as the one to be presented. It is said, instead, that one ought to work on limited hypotheses for small, manageable problems–categorization effects in free recall, verification latencies for negative sentences, search of items in short-term memory, and so on ad infinitum. Indeed, we have been told by many respected colleagues in psychology that we will surely fail because we “are trying to explain everything.” Of course, we are not. Human memory is but a very small part of the psychological domain. To make a salient contrast, a criticism we are apt to receive from colleagues outside of psychology (e.g., artificial intelligence) is that we are far too narrow in our perspective and aims.

The reason for writing a theoretical book on human memory is the belief that we have something important to offer. In rejecting the earlier global theories, modern research on human memory has overreacted to the opposite extreme; it has become far too narrow, particulate, constricted, and limited. There is no overall conception of what the field is about or even what it should be about. There is no set of overarching theoretical beliefs generally agreed upon which provide a framework within which to fit new data and by which to measure progress. Were we describing an unhappy personality, we would say that the contemporary study of memory has lost its sense of direction, its sense of purpose, and it is drifting aimlessly with much talent but little focus. This point was stated forcibly by a recent, informed but highly critical review of the field (Tulving & Madigan, 1970).

1.1. CONCERN FOR SUFFICIENCY CONDITIONS

Laboratory studies of memory appear under the inexorable control of a distinct set of “experimental paradigms,” a standard set of “tasks,” which seem by their nature to spew out an unending string of methodological variations and empirical studies. But the phenomena studied are becoming further and further removed from the manifestations of memory in everyday life. There would be nothing necessarily wrong with this esoterica provided psychologists had some clear conception of how their research and theories would eventually fit together into a system adequate to explain the complexities of everyday human memory. But, on the contrary, it appears that we psychologists are totally unconcerned about having our psychological theories meet certain sufficiency conditions. It is not enough that a theory make adequate ordinal predictions for a particular situation and experiment; in addition, it should be shown that its principles are sufficient to play a part in the explanation of the total complexity of human behavior. For instance, one could require of a model of memory that it be sufficiently powerful to succeed in simulating question-answering behavior.

When we began to concern ourselves with sufficiency conditions, we were forced to fundamental reconceptualizations regarding the nature of memory. We found that memory could no longer be conceived as a haphazard jumble of associations that blindly record contiguities between elements of experience. Rather, memory now had to be viewed as a highly structured system designed to record facts about the world and to utilize that knowledge in guiding a variety of performances. We were forced to postulate entities existing in memory which have no one-to-one correspondence with external stimuli or responses. As discussed in Chapter 2, such structures violate the Terminal Meta-Postulate of classical associationism and stimulus-response psychology. It also became necessary to postulate the existence in the mind of highly complex parsing and inferential systems which function to interface the memory component with the external world. Furthermore, we were forced to postulate the existence of innately specified ideas in the form of semantic primitives and relations. We will therefore be proposing and arguing for a radical shift from the associationist conceptions that have heretofore dominated theorizing on human memory.

This shift is most apparent in the unit of analysis which we adopt. Unlike past associative theories, we will not focus on associations among single items such as letters, nonsense syllables, or words. Rather, we will introduce propositions about the world as the fundamental units. A proposition is a configuration of elements which (a) is structured according to rules of formation, and (b) has a truth value. Intuitively, a proposition conveys an assertion about the world. The exact structural properties of our propositional representation will be set forth in Chapter 5. We will suppose that all information enters memory in propositional packets. On this view, it is not even possible to have simple word-to-word associations. Words can become interassociated only as their corresponding concepts participate in propositions that are encoded into memory. However, propositions will not be treated here as unitary objects or Gestalt wholes in memory having novel, emergent properties. Rather, propositions will be conceived as structured bundles of associations between elementary ideas or concepts. However, our insistence that all input to memory be propositional imposes certain well-formedness conditions on the structure of the interidea associations. This notion of structural well-formedness is one that was completely lacking in past associative theories and was at the heart of many rationalist attacks on associationism.

1.2. NEO-ASSOCIATIONISM

We shall use the term “neo-associationism” to denote this new conception of human memory. While it introduces substantial deviations from past associationist doctrines, it still maintains a strong empiricist bias. We feel that the full significance of these theoretical assumptions can only be appreciated when one understands the associationist tradition out of which they came. Therefore, we have devoted Chapter 2 to an analysis of the associative tradition that extends from Aristotle through current American psychology. We will argue that a defining feature of associationism has been its methodological empiricism. That is, all associationists have accepted as their task the job of taking the immediate sense-data available to them and constructing their theory directly from these, always letting the data dictate the nature of the theory. This is contrasted in Chapter 3 with the methodological rationalism which attempts to first arrive at abstract, sufficient conditions, or constraints for the phenomena at hand, and then tries to relate these abstractions and conceptual constraints to the empirical world.

The contrast we are making between methodological empiricism and methodological rationalism corresponds (not surprisingly) to the more frequently made distinction between empiricism and rationalism. In the strong version of empiricism, the mind begins as tabula rasa, and all knowledge is a consequence of the passive encoding of experience. The strong version of rationalism claims that the mind begins highly structured and all significant knowledge derives from the mind's initial structure. According to the rationalists, the role of experience is simply to stimulate the mind to derive that knowledge. Methodological empiricism and rationalism are not concerned with the origins of human knowledge, but rather with procedures for developing a scientific theory. However, we can almost derive a definition of each by substituting “scientific theory” for “mind” in the above statements of empiricism and rationalism. That is, methodological empiricism claims a scientific theory can be built up from immediate data by the blind procedure of generalization; whereas, methodological rationalism insists the theory builder must bring the essential structure of the theory to the phenomena to be explained.

Neo-associationism represents a profane union of these two methodologies. There is no attempt at a “creative synthesis” of these two positions; we simply pursue both methods in parallel in constructing a theory. The result is a theory that irreverently intermixes connectionism with nativism, reductionism with wholism, sensationalism with intuitionism, and mechanism with vitalism. Depending on the theoretician's propensities, the mixture can be claimed to be either more rationalist than empiricist or vice versa. The mixture we will offer is still strongly empiricist, much more so than the other neo-associationist theories that we will examine.

The various neo-associationist theories of memory (e.g., Simon & Feigenbaum, 1964; Collins & Quillian, 1972b; Rumelhart, Lindsay, & Norman, 1972), including our own, have been cast in the form of computer simulation models of memory. This is no accident. The task of computer simulation simultaneously forces one to consider both whether his theory is sufficient for the task domain to be simulated and also whether it can deal with the particular trends found in particular experiments.

Our therorizing and experimentation are specifically oriented towards memory for linguistically structured material. With such interests, one cannot help but make constant contact with the recent ideas in linguistics. The linguistic work, particularly of Chomsky, Fillmore, Lakoff, Katz, Ross, and their associates, is important for a second reason. These linguists have argued effectively for the importance of sufficiency conditions in linguistics. As a consequence, over the past decade rationalism and mentalism have become strongly entrenched in linguistics. The rationalist “revolution” has been imported from linguistics into psychology. Thus, the developments in linguistics are an important source behind the neo-associationist developments.

Two substantial chapters are being devoted to an extensive historical and theoretical review of efforts related to our own. This is clearly out of character for a typical “research volume.” The usual practice for American psychology is to restrict its focus to the last 5 or 10 years of experimentation centered around a narrowly circumscribed topic. This practice is lamentable since true scholarly endeavor would seem to require an appreciation of the historical and intellectual context within which that scholarship occurs. Without knowledge of that context, it is not possible to discriminate between significant theoretical advance as opposed to elaboration of an established paradigm. Chomsky (1968) has argued persuasively for a similar historical perspective in linguistics.

Our work began in the typical intellectual isolation of experimental psychology, but we constantly found ourselves being led into discussions of issues about which we know very little. Therefore, we have tried to trace the connections between our work and that which had occurred in past centuries or which was occurring in related fields. Our perception of what questions were important changed; similarly, the character of the theory and research to be presented is very different from what we had originally projected and from the typical fare that one finds in psychology. It can only be appreciated in the perspective of the historical context that we set in the first two chapters. One of the incidental advantages of a theory so constructed is that it provides the reader with an integrated viewpoint from which to perceive his own experimental research, related research in psychology and other fields, and the relationship between this research and what has happened in past centuries.

Following these two review chapters, the remainder of the book serves as a forum for presenting our theory and research. We have many experiments to report that have not appeared before in print. We will also review and comment upon a large number of recent experiments that seem particularly interesting with respect to the issues that we are raising. Although there is no attempt to review extensively the literature in human memory, we do hope to establish theoretical connections among many different areas of experimentation in psychology.

To preview the contents of the later chapters, Chapter 4 provides a general overview of our model of long-term memory. We have christened the model HAM, an acronym for Human Associative Memory. The subsequent three chapters set forth most of the substantive theoretical assumptions of that model. The character of presentation varies considerably from one chapter to the next. In Chapter 5, entitled “The Structure of Knowledge,” we propose a structure in which information will be encoded and stored in long-term memory. In Chapter 6 we will ask how the memory system recognizes that it has experienced something before. This issue, of how current stimuli contact old traces, is a point of notorious difficulty for other accounts of memory. Finally, in Chapter 7 we will present a stochastic model of how incoming information is encoded into long-term memory.

The remaining chapters will be concerned with relating our theory to various areas of research and experimentation. In Chapter 8 we will examine the question of how long-term memory is searched for information, to decide whether or not some fact is known or some statement is true. This is the problem of fact retrieval. In Chapter 9, we will discuss how our model would perform in the typical verbal learning paradigms such as paired-associate learning and free recall. Finally, in Chapter 10 we will discuss how different information inputs interfere with one another to produce forgetting. We will compare our model of this process with past theories of interference and forgetting.

1.3. THE FUNDAMENTAL QUESTIONS

There are well-known advantages to vagueness in constructing a theory; it protects the theory from disconfirmation. The typical strategy is to articulate the theory at those points where it makes contact with confirming evidence, but otherwise to shroud it in sufficient vagueness so that any other present or future data cannot unambiguously disconfirm the model. We have tried to avoid this tactic. Not only is our theory vulnerable to future disconfirmation; it also clearly fails to handle a number of the existing facts. The points of misfit will be openly acknowledged at the appropriate places. It is difficult to determine how serious these failures are. In a complex model like HAM it is always possible to introduce some special assumption that will handle any particular discrepancy. Also, the misfits may indicate a mistake in one particular assumption rather than a flaw in the grand theoretical design.

The fundamental issue at stake with respect to our theory is its neo-associationist character. This is not to be found in any particular assumption, but rather pervades diffusely throughout the whole enterprise. Our strong computer-simulation orientation has led to a class of controversial assumptions. Information processing in HAM tends to be in terms of discrete units called ideas and associations, and it proceeds in sequential steps, whereas parallel, interactive processes are assumed to be minimal. Can one really claim that a human processes information in this discrete, serial manner? But the physiology of the brain is very different from that of a serial, digital computer, and analogue, parallel processes would not seem out of character for that mysterious organ (cf. Von Neuman, 1958). Perhaps, then, our theory has been too strongly determined by what is easy to simulate on a computer rather than by considerations of psychological plausibility. That is one fundamental question.

Another source of difficulty with our theory may arise from our strong empiricist leanings. We have insisted that all knowledge in memory should be built up from input to the memory. We have denied that memory has any capacity to spontaneously restructure itself into more useful forms. Perhaps we have made memory too passive, too much of a tabula rasa. That is a second fundamental question.

On the other hand, we have granted the mind a great deal of self-structuring power in our assumptions regarding the perceptual parsers that transform external stimuli into memory input, or the various inferential and problem-solving abilities that enable the system to make intelligent use of the information recorded in memory. One is forced to postulate such powerful mechanisms in order to interface a memory with the world. The postulated mechanisms are enormously more complicated than any of the theoretical devices that have been previously postulated in associative theories. Perhaps, if we had complicated the proposed memory system, we could have simplified the interfacing apparatus. That is a third fundamental question.

Another possible flaw in the grand design has to do with our insistence on making the propositional representation fundamental. We will want to encode perceptual as well as linguistic input into this uniform propositional base. Perhaps we are choosing a representation that is too logical and abstract. Perhaps the primary representation of knowledge is of some diffuse, sensory sort; and our ability to encode information propositionally in this or...

Table of contents

• Front Cover
• Half Title
• Title Page
• Copyright
• Contents
• Preface to Brief Edition
• Preface to First Edition
• 1. Introduction
• 2. Associationism: A Historical Review
• 3. Rationalist Countertraditions
• 4. An Overview of Ham
• 5. The Structure of Knowledge
• 6. The Recognition Process
• 7. Model for Sentence Learning
• 8. Fact Retrieval
• 9. Verbal Learning
• 10. Interference and Forgetting
• 11. Problems and new Issues
• References
• Author Index
• Subject Index