The Structure of Long-term Memory

A Connectivity Model of Semantic Processing

Wolfgang Klimesch

• Publisher: Psychology Press
• Year: 2013
• Print ISBN:
  9781138876385
• eBook ISBN:
  9781134777778
• Topic: Psychology
• Subtopic: Cognitive Psychology & Cognition
• Index: Psychology

How is information stored and retrieved from long-term memory? It is argued that any systematic attempt to answer this question should be based on a particular set of specific representational assumptions that have led to the development of a new memory theory -- the connectivity model. One of the crucial predictions of this model is that, in sharp contrast to traditional theories, the speed of processing information increases as the amount and complexity of integrated knowledge increases. In this volume, the predictions of the model are examined by analyzing the results of a variety of different experiments and by studying the outcome of the simulation program CONN1, which illustrates the representation of complex semantic structures. In the final chapter, the representational assumptions of the connectivity model are evaluated on the basis of neuroanatomical and physiological evidence -- suggesting that neuroscience provides valuable knowledge which should guide the development of memory theories.

1

The Representational Problem: A Historical Perspective

Those who concern themselves with the history of experimental psychology will no doubt conclude that the topics this discipline has dealt with in the last 100 years have not changed much (cf. with this the synoptic works of Boring, 1950; Flugel, n. d.; Lück, Miller, & Rechtien, 1984). In cognitive psychology, for example, there are many experimental paradigms and theoretical concepts that have been dealt with in similar fashion, but under different titles, over many historical periods. Consider the cognitive-psychological concept of a limited capacity of short-term memory (STM). James McKeen Cattell had already carried out studies on the attention span at the Leipzig Institute of Wilhelm Wundt and observed that in a simultaneous, tachistoscopic display of several stimuli only 4 to 6 units — be they lines, letters, or words — could be understood and remembered (Flugel, n. d., p. 157). The terms capacity of STM and attention span (cf. Ebbinghaus, 1885) refer to one and the same empirical phenomenon. Where they differ is in their historical context and in the ways in which they happen to be embedded in overlapping theoretical relations (e.g., Bahrick, 1985). There is a long list of historical concepts and empirical phenomena that have been redeployed. Here, however, we content ourselves with a few references and, in doing so, recall the historical relations between the following concepts:

• the similarity between Donder’s “subtractive procedure” (Donders, 1868; cf. Massaro, 1975, p. 44; Sanders, 1971, p. 17) and the experimental paradigms in cognitive psychology (Posner, 1978; Posner, Boies, Eichelman, & Taylor, 1969);
• the importance that reaction time paradigms have assumed in experimental psychology ever since Wilhelm Wundt conducted his extensive experiments at the Leipzig Institute;
• the continued importance of verbal association paradigms, which have been an important feature in memory psychology since Ebbinghaus (1885; e.g., McGeoch, 1942; Slamecka, 1985a, 1985b);
• the experimental documentation concerning the superiority of visual memory over verbal memory, which has been known since Kirkpatrick (1894) and Calkins (1898), and not just since Shepard (1967) or Paivio (1971);
• and, within the confines of cognitive psychology, the “rediscovered” findings — dating back to Cattell (1886) — that words can be identified more quickly than images (Potter & Faulconer, 1975).

Seen in this light, it is not surprising that critical voices continually claim that little, if any, scientific progress is being made in our discipline. We attempt to show that this view is misleading because it ignores important recent developments. It can be seen from the comparison of the most important historical approaches that it is the specific preoccupation with the representational problem that is in fact the new contribution of cognitive psychology as it emerged in the Anglo-American sphere. Only the explicit consideration of the representational problem can build a foundation on which to arrive at a consistent interpretation of memory phenomena. Chapters 2, 3, and 4 consider this question in greater detail.

The name cognitive psychology has often led to misunderstandings about the actual concerns of this comparatively young field of research. A more precise and specific description would be “the psychology of information processing.” Those critics who claim that the “cognitive trend” will soon turn into an “action trend” (Graumann, 1983, p. 68) or “emotional trend,” are not taking into consideration the specific contribution made by cognitive psychology. Its task is to examine how information is encoded, represented, and processed. The type of information involved — whether “cognitive,” “emotional,” or “action relevant” — has no immediate impact on the study of the representational problem. Cognitive psychology is not the counterpart of an “emotional” or “action” psychology. Nevertheless, it emphasizes the cognitive content, because for methodical reasons emotional processes are much more difficult to examine empirically than cognitive processes.

The next section considers the fundamental concepts needed to explain and elucidate the representational problem. Subsequent sections are then devoted to the historical development of the representational problem. Based on these sections, we show that the consideration of the representational problem leads to new and important discoveries.

1.1 Defining the Representational Problem: Code, Coding, and the Coding Format

Which processes enable sensory information to be recognized, stored, and recalled? This is the main research topic in cognitive psychology, and characterizes what is generally understood by the encoding or representational problem. Encoding or coding is the transformation of sensory information into a certain format of a memory representation, resulting in the formation of a memory code. The form, composition, and structure of the internal representation, on the other hand, is known as the coding format. It is precisely this interest in how information is “represented” (i.e., how information is stored in memory) that has led to the preeminent position of memory research in the field of cognitive psychology. Because coding is considered a process of transformation — reflecting different stages of information processes, such as perception, recognition, and selective attention — it becomes clear that memory can be described only if empirically validated assumptions regarding the entire information-processing system are made. This idea of a close interdependence between the properties of the encoding format and the structure of the entire information-processing system is discussed in chap. 4.

The description of the elementary properties of codes is crucial here. Consequently, memory research is the main focus of the following historical survey.

1.2 Memory Research: A Historical Perspective

It is worth noting that Ebbinghaus, as the founder of empirical memory research, had no interest in representational assumptions. Following his description of the “Mangelhaftigkeit des Wissens über das Gedächtnis” (The inadequate knowledge about memory), Ebbinghaus (1885) wrote:

And because all our knowledge is so uncertain and imprecise, it has remained unfruitful for an understanding of a theory of memory, recall, and association processes. In our ideas on its physical basis, we use different metaphors such as stored images, imprinted ideas, and encarved traces etc., of which we only know that one thing is certain that they are not correct, (p. 7)

Ebbinghaus could hardly have expressed his rejection of the representational problem more clearly. Nevertheless, it would be wrong to assume that memory psychology could have managed without representational assumptions in its early stages. These were more implicit than explicit in nature and were based on the fundamental conceptions of associationism, which was the predominant psychological trend in the second half of the 19th century. Ebbinghaus saw the goal of his work not as an attempt to empirically examine associationistic representational assumptions, but rather to substantiate the scientific claim of his experimental approach in psychology. The status of implicit and explicit representational assumptions within memory psychology is discussed in chaps. 2 and 3.

In order to evaluate the importance of Ebbinghaus’ work, we must consider the historical factors that influenced it. Two factors need to be taken into consideration: On the one hand, there was the dismissive attitude toward psychology as an experimental science and, on the other hand, the limited prospect of ever arriving at an empirically validated theory of memory. Ebbinghaus, like other empirical psychologists of his day, was primarily concerned with showing that mental processes — like physical and biological processes — could also be examined and understood using scientific methods. His approach to the problem was therefore primarily one of method as opposed to content. It consisted of the classical scientific procedure of proving what effect the specific variation of one or more independent variables has on one or more dependent variables. The dependent variable was memory performance or the extent of forgetting. Among the most important independent variables were the number of repetitions and the retention interval (i.e., the time that elapses between presentation and test), as well as the nature and amount of material to be learned. Ebbinghaus arrived at a series of rules governing the examined variables, whereby the rule governing the length of the retention interval and memory performance, known as the “forgetting curve,” is only one of the better-known examples.

From the perspective of the then-dominant school of thought, these results were a sweeping success. Ebbinghaus was now able to prove that the study of higher mental processes was also possible for a psychology using scientific methods. Thus, together with Wilhelm Wundt, Ebbinghaus made a significant contribution toward the founding of a scientifically and experimentally oriented psychology. It is interesting to note, however, that Wundt had a negative attitude toward the study of higher mental processes — as they represented memory performance — and seven years after first publishing Philosophische Studien (the journal founded by Wundt in 1890) Ebbinghaus published Zeitschrift für Psychologie und Physiologie der Sinnesorgane, which to a certain extent provided a forum for independent researchers outside of the Wundtian School (Boring, 1950; Flugel, n. d., p. 167).

Müller, Jost, and Pilzecker (Jost, 1897; Muller & Pilzecker, 1900) followed a procedure similar to Ebbinghaus, but even this was completely derived from associationism (e.g., Müller, 1917). The first 30 years of empirical memory research are thus essentially characterized by two factors: on the one hand, by the methodical-scientific orientation that predominates, and, on the other hand, by simple associationistic conceptions of memory. As a result, one finds a wealth of important rules that up to now remain untouched in applied memory psychology. What was missing, however, were approaches to general, overlapping memory theories. Therefore, within the framework of classic memory psychology, it was not the investigation of representational assumptions that predominated, but in effect only the question of how associations develop between memory contents.

After this first classical period of memory research, the emphasis of scientific research shifted from Germany to the Anglo-American sphere. Even there the focus remained for a long time — up to the early 1960s — within the framework of associationism and behavioristic approaches. There were, however, a few important exceptions. For example, F. C. Bartlett, then a Cambridge psychologist, was among the first to introduce the concept of a mental “schema” into memory psychology. Bartlett (1932) assumed that sensory information is structured and stored alongside these mental schemas, which are themselves represented in memory. Schemas, which are derived by means of abstraction, represent the essential characteristics of a whole class of stimuli. The similarity to the Gestalt concept, but also to Rosch’s (1975) “typicality concept” in the area of concept formation, should not be overlooked. According to Bartlett, the associationistic viewpoint of memory as a passive store was abandoned, and the active, structuring character of memory came to the fore. Bartlett assumed that schemas are of crucial importance for perception and thinking as well as memory. Thus it became clear that memory could not be studied and understood in isolation from other phenomena of the human mind.

Oldfield joined others in adopting Bartlett’s schema concept, which Evans and his colleagues (Evans, 1967; Evans & Arnoult, 1967; Evans & Edmonds, 1966) subsequently applied in the area of concept formation (Homa & Cultice, 1984). Seen from our perspective, Bartlett’s works occupy a special position, because they were among the first ideas that — after 50 years of empirical memory research — started out with clearly drawn representational assumptions. It was all the more surprising, therefore, that these important ideas — apart from those exceptions mentioned earlier — did not receive widespread attention.

In the late 1940s, Shannon’s information theory and Wiener’s cybernetics theory stimulated new interest among researchers in psychology and in other scientific disciplines (see Wiener, 1968). In the hope of a promising and fruitful approach to the study of perception, memory, and thinking, information theory and cybernetics were frequently introduced into psychology with a lack of critical insight. The number of works relating to the concept of information theory, above all in the Anglo-American sphere, soon became vast (Garner, 1962). One of the best-known experiments was carried out by Miller (1956) on the “magical number seven.” He was able to show that the human “channel capacity” lies within a range of two and three bits. Thus, STM can store at maximum between four and eight unrelated information units. If this capacity limit is exceeded, information stored in STM is lost. Miller, however, emphasized that his results were closely related to Külpe’s, which were reported at the first Congress of Experimental Psychology in Gieβen in 1904 (which was, incidentally, inaugurated by Ebbinghaus). Miller failed to mention that similar experiments had already been conducted by James McKeen Cattel and Wilhelm Wundt, who came to the conclusion that the immediate attention span comprises approximately four to six different units.

Psychological research, based on information theory, was often accused of merely altering the names of already well-established results and concepts. This criticism is certainly justified. Information theory was inadequate for psychological resarch because it neither led to a better and deeper understanding of results already known, nor contributed to the discovery of new experimental paradigms. It therefore comes as no surprise that the significance of information theory began to wane by the mid-1960s.

It may seem paradoxical to assume that, in spite of its negative effects, information theory was an essential precursor to cognitive psychology. In order to explain this notion, we must deal separately with the negative and positive influences of information theory.

Its negative influence can be seen in the fact that its concepts — though useful for describing the stimulus material (Garner, 1962; Klimesch, 1974; Miller, 1956) or measuring the amount of information transferred from input to output (cf. choice reaction time experiments and the concept of “channel capacity”) — were inadequate in defining representational assumptions for human information processing. The application of information theory in psychology, therefore, stood (whether intended or not) in the best behavioristic tradition.

Information theory also had a positive influence. For the first time in the history of psychology there was a model to describe basic principles of information processing. It was only after the advent of cybernetics that psychologists began to see a parallel between the technical, machine-related, and the human or, generally speaking, biological information processing. Thus, it was only a small step toward the development of a new system, namely that of artificial intelligence and later of cognitive science. Within the framework of this discipline, which developed somewhat later than cognitive psychology and maintained close connections with it, the representational problem was of crucial importance from the outset (J. R. Anderson & Bower, 1973; Newell & Simon, 1972).

Toward the end of the 1960s, there was a “cognitive revolution,” or an abrupt renunciation of the formal, descriptive approach of information theory toward the special interest shown in “opening the black box,” left in the wake of behaviorism. Opening the black box made it necessary to make assumptions about the internal structure of information processing. Thus, representational assumptions became the central concern of cognitive psychology. Four works can be regarded as milestones in this new development: Neisser’s Cognitive Psychology (1967), which was, from a historical standpoint, probably the most important work (J. R. Anderson, 1985a, p. 9; Bahrick, 1984); Atkinson’s and Shiffrin’s work (1968) on an overlapping theory of memory; the introduction into human information processing by Lindsay and Norman (1972); and the epoch-making Human Associative Memory by J. R. Anderson and Bower (1973). The importance of this new trend was soon noticed and recognized in the German-speaking area (Wimmer & Perner, 1979).

1.3 Representational Assumptions in Historical Approaches to the Psychology of Perception and Thinking

Psychology of perception and thinking confronted representational assumptions long before memory psychology. The reason for this is obvious: In contrast to memory processes, certain structures and principles governing perception and thinking are to some extent more accessible to self-observation and can thus be detected more easily.

Thus, for the Würzburg School, the question of how the contents of consciousness are structured and of what elements they consist were among the most important research topics (cf. Messer, 1924, p. 24). By analyzing the contents of consciousness, which could either be conducted by experimental methods or by the method of systematic self-observation, an attempt was made to classify its elements. These examinations led to differentiation between two types of consciousness elements: sensations and intentional acts (cf. the excellent review in Humphrey, 1963; Munzert, 1984). In order to distinguish their approach from that of elementaristic psychology, they introduced an important restriction: Perception and thought, for example, could not be explained by a decomposition into individual elements. The exclusive preoccupation with phenomena accessible to self-observation, as well as the assumption that consciousness processes could, in principle, not be explained by a knowledge of their elements, were the obvious reasons why access to explicit representational assumptions was denied the Würzburg School. Chapter 4 discusses why a general understanding of perception, memory, and thought processes is impossible without assumptions regarding the unconscious course of informat...

Table of contents

• Cover
• Half Title
• Title Page
• Copyright
• Dedication
• Contents
• Preface
• Introduction
• 1. The Representational Problem: A Historical Perspective
• 2. Traditional Theories of Forgetting
• 3. The Failure of Traditional Forgetting Theories: Misleading Representational Assumptions
• 4. STM Codes: Their Structure and Decay
• 5. Network Theories: Basic Assumptions on the Structure of LTM
• 6. Strictly and Nonstrictly Hierarchical Models of Fact Retrieval
• 7. Strictly and Nonstrictly Hierarchical Models of Word Meaning
• 8. The Connectivity Model
• 9. A Connectivity Model for Semantic Processing
• 10. The Simulation Program CONN1
• 11. Representational Assumptions and their Possible Neural Bases
• 12. Concluding Remarks
• References
• Author Index
• Subject Index