A general framework for the analysis of concept identification tasks

A general framework is presented for concept identification based on hypothesis-testing theory. It is a modification of the duoprocess theory presented by Chumbley (1972). It is shown how Markov models for various complex concept identification tasks may be derived from this framework and how such models may be analyzed by making use of probability generating functions. Two experiments are described. In experiment 1 three tasks were used: two simple tasks, where the subject either only had to select the relevant dimension in order to solve the problem or only had to learn a short list of paired-associates, and a more complex task, where both processes were needed to reach the solution. The results were in general favorable to the theory. Experiment 2 was designed to test the application of the theory to the four-choice concept problem. The predictions of the theory are compared to those of the subproblem learning theory of trabasso and Bower (1964), modified to include a ‘learning-on-errors’ assumption. The fit of the duoprocess theory was reasonably good and superior to that of the subproblem learning theory.     

Discrimination nets as psychological models

Simulations of human cognitive processes often employ discrimination nets to model the access of permanent memory. We consider two types of discrimination nets—EPAM and positive-proper-only nets—and argue that they have insufficient psychological validity. Their deficiencies arise from negative properties, insufficient sensitivity to the discriminativeness of properties, extreme sensitivity to missing or incorrect properties, inefficiency in representing multiple knowledge domains, and seriality. We argue that these deficiencies stem from a high degree of test contingency in utilizing property information during acquisition and memory search. Discrimination nets are compared to other models that have less or no test contingency (e.g., PANDEMONIUM) and that thereby avoid the problems of discrimination nets. We propose that understanding test contingency and discovering psychologically valid ways to implement it will be central to understanding and simulating memory indexing in human cognition.     

知识表征的新观点--知觉符号理论

在认知科学中一个经典的争论是知识在人的头脑中是如何表征的。其中占主流的理论一直是命题符号理论,而最近Barsalou又提出了一种新的观点——知觉符号理论。本文对这一新理论做了介绍,阐述了它的主要特征,比较了它与命题符号理论的差异,并提供了现今对这一理论的实验证实。希望借此能启发国内的研究者也来关注和研究这一理论。     

A theoretical analysis of the PAN ambiguous-cue problem

The PAN ambiguous-cue problem consists of three stimuli: P, the positive (consistently rewarded) cue; N, the negative (consistently nonrewarded) cue; and A, the ambiguous cue which is negative (nonrewarded) when paired with P, but positive (rewarded) when paired with N. This paper demonstrates how the finding of superior performance on the NA trials as compared to the PA trials (with stereometric objects) can be rigorously derived from a recent extension of Hull-Spence discrimination learning theory, the Stimulus Interaction Hypothesis. It is also shown how this theory can account for the converse outcome of better performance on PA than on NA trials when planometric plaques are used as stimuli. Finally, alternative explanations of the findings are described and evaluated.     

The response-stimulus contingency and reinforcement learning as a context for considering two non-behavior-analytic views of contingency learning

This paper introduces a special section on the contingency. Bower and Watson were invited to present their views of contingency learning in human infants from outside the context of behavior analysis, and Cigales, Marr, and Lattal and Shahan provided commentaries that point out some of the more interesting and controversial aspects of those views from a behavior-analytic perspective. The debate turns on how to conceptualize the response-stimulus contingency of operant learning. The present paper introduces the contingency concept and contingency detection by subjects, as well as research practices in behavior analysis, in a context in which the dependency between infant responding and the presentation of environmental consequences may be disrupted through procedures in which ordinarily consequent events occur before the response or in its absence. These points can relate to and serve as an introduction to the Bower and Watson papers on infant contingency learning as well as to the three commentaries that follow.     

Fragment and schema models for recall

Two general conceptualizations of recall may be distinguished. One holds that recall is sub-served by memory traces in which representations of different aspects of external events are linked directly to each other. The other holds that these representations are connected via a mediating concept. Formalizations of these views are provided by two theories examined by Ross and Bower (1981a): the “fragment” and “schema” models, respectively. Ross and Bower found that patterns of cued recall of clusters of words with a common theme were predicted well by the schema model, but not by the fragment model. Two experiments reported here show that the reverse is true if the clusters of words are selected randomly. Ross and Bower also considered separately a third theory, the “horizontal” model. However, the horizontal model is shown here to be a constrained form of the fragment model. Another constrained form of the fragment model, termed the “link” model, is introduced and is shown to provide an account of encoding as well as of retrieval that, although it is outstandingly parsimonious, is also approximately accurate. Finally, it is shown that the memory structures characterized by the fragment model may be interpreted as either graphs or digraphs: that is, networks with either bidirectional or unidirectional connections between nodes.     

Chunking mechanisms in human learning

Pioneering work in the 1940s and 1950s suggested that the concept of 'chunking' might be important in many processes of perception, learning and cognition in humans and animals. We summarize here the major sources of evidence for chunking mechanisms, and consider how such mechanisms have been implemented in computational models of the learning process. We distinguish two forms of chunking: the first deliberate, under strategic control, and goal-oriented; the second automatic, continuous, and linked to perceptual processes. Recent work with discrimination-network computational models of long- and short-term memory (EPAM/CHREST) has produced a diverse range of applications of perceptual chunking. We focus on recent successes in verbal learning, expert memory, language acquisition and learning multiple representations, to illustrate the implementation and use of chunking mechanisms within contemporary models of human learning.     

Prior knowledge and correlational structure in unsupervised learning.

Prior knowledge has been shown to facilitate both supervised and unsupervised category learning, but questions remain about how this facilitation occurs. This article describes two experiments that investigate the effects of prior knowledge on unsupervised learning, using the exemplar-memory task of Clapper and Bower (2002). Experiment 1 demonstrates that prior knowledge facilitates learning in this task, as expected, and that this facilitation extends to both knowledge-relevant and knowledge-irrelevant features of the new categories. Experiment 2 shows that knowledge facilitates learning not only by increasing the probability that people will discover separate categories, but also by making the features of different categories seem less interchangeable, thereby reducing interference and confusion among them. Taken together, these experiments demonstrate that prior knowledge has multiple effects on unsupervised learning and suggests that the exemplar-memory task may provide a useful procedure for disentangling and investigating these effects.     

A simulation of memory for chess positions

This paper describes an information-processing model, MAPP, implemented as a computer program, that simulates the processes subjects use to remember and reproduce chess positions they have seen briefly. The model incorporates processes adapted from PERCEIVER, an information-processsing theory of eye movements in chess perception, and EPAM, a theory of rote verbal learning. The data from MAPP show good agreement with the performance of strong chess players in identical tasks.     

Brief escape from a dangerous place: The role of reinforcement in the rat's one-way avoidance acquisition

Five experiments are reported which attempt to eliminate two possible sources of reinforcement in one-way avoidance learning: a period of escape from aversive apparatus cues and the termination of the warning signal (WS). A “brief escape”/one-way avoidance procedure was developed to minimize the time rats spent away from the shock box. It was found that, although prolonged escape from shock box cues contributes to the acquisition rate of one-way avoidance, it is not essential for relatively rapid acquisition to occur. It was also found, in agreement with earlier evidence, that WS termination makes no discernible contribution to learning in the one-way situation. Thus, neither source of reinforcement appears to be necessary for acquisition. At this point, one may argue either that reinforcement is not necessary for rapid one-way avoidance acquisition [e.g., Bolles' 1972 species-specific defense reaction hypothesis (In G. H. Bower (Ed.), The psychology of learning and motivation: Advances in research and theory (Vol. 6). New York: Academic Press, 1972)] or that there are other sources of reinforcement. As an example of the latter approach, a consummatory stimulus reward hypothesis is advanced.     

Ad hoc category restructuring

Participants learned to classify seemingly arbitrary words into categories that also corresponded to ad hoc categories (see, e.g., Barsalou, 1983). By adapting experimental mechanisms previously used to study knowledge restructuring in perceptual categorization, we provide a novel account of how experimental and preexperimental knowledge interact. Participants were told of the existence of the ad hoc categories either at the beginning or the end of training. When the ad hoc labels were revealed at the end of training, participants switched from categorization based on experimental learning to categorization based on preexperimental knowledge in some, but not all, circumstances. Important mediators of the extent of that switch were the amount of performance error experienced during prior learning and whether or not prior knowledge was in conflict with experimental learning. We present a computational model of the trade-off between preexperimental knowledge and experimental learning that accounts for the main results.     

Some neurolinguistic implications of prearticulatory editing in production

This paper argues that a complete model of language production must include a prearticulatory editing component. The function of this component is to monitor planned speech for deviations from the speaker's intention and repair any deviations that are found. It is claimed that adding such an editing component onto a production model fundamentally changes any account of aphasic symptoms using that model. As a case in point it is shown that E. M. Saffran's (1982, British Journal of Psychology, 73, 317-337) argument that agrammatic Broca's aphasia involves a deficit at the functional level of M. F. Garrett's (1975, in G. H. Bower (Ed.). The psychology of learning and motivation, New York: Academic Press) production model is no longer sound when prearticulatory editing processes are considered.     

The role of object motion in forging long-term representations of objects

It has been suggested that objects are represented as collections of two dimensional images. Although possible in theory, such a representation must first be built, which presupposes a mechanism for collating these images. This is not trivial. Any such mechanism would have to be able to group images which differ considerably in appearance. One possible solution to this problem is to associate those images whose appearance is closely temporally correlated, on the assumption that multiple views of an object are frequently experienced in close temporal succession. This paper describes evidence for the influence of just such a mechanism in human observers, and its effect on long-term representations of initially novel faces. Discrimination performance for previously viewed faces is shown to depend on whether views of the two faces being discriminated, had previously been seen in close temporal succession or not. Irrespective of the preferred theory of how humans represent objects, these results reveal strong evidence for a novel, time-based learning mechanism that strongly influences within-category discrimination.     

Remembering environmental sounds: The role of verbalization at input

Two experiments investigated the role of verbalization in memory for environmental sounds. Experiment i extended earlier research (Bower & Holyoak, 1973) showing that sound recognition is highly dependent upon consistent verbal interpretation at input and test. While such a finding implies an important role for verbalization, Experiment 2 suggested that verbalization is not the only efficacious strategy for encoding environmental sounds. Recognition after presentation of sounds was shown to differ qualitatively from recognition after presentation of sounds accompanied with interpretative verbal labels and from recognition after presentation of verbal labels alone. The results also suggest that encoding physical information about sounds is of greater importance for sound recognition than for verbal free recall, and that verbalization is of greater importance for free recall than for recognition. Several alternative frameworks for the results are presented, and separate retrieval and discrimination processes in recognition are proposed.     

A theory of learning and transfer: II

The theory of discrimination learning developed in Part I of this paper is extended to more complex learning situations such as those used for differentiating relative from absolute habits. The greater difficulty of learning an absolute discrimination than of learning a relative discrimination is predicted. The theory is also extended to cover cases of transfer. It predicts that under the ordinary training conditions transfer to new stimulus situations will be on a relative basis, and it predicts the type of transfer to be expected under other training conditions. A number of additional established facts of learning and transfer can be deduced from the theory. Methods of testing the adequacy of the theory by new and more crucial experiments are suggested.     

Five Seconds or Sixty? Presentation Time in Expert Memory

For many years, the game of chess has provided an invaluable task environment for research on cognition, in particular on the differences between novices and experts and the learning that removes these differences, and upon the structure of human memory and its paramaters. The template theory presented by Gobet and Simon based on the EPAM theory offers precise predictions on cognitive processes during the presentation and recall of chess positions. This article describes the behavior of CHREST, a computer implementation of the template theory, in a memory task when the presentation time is varied from one second to sixty, on the recall of game and random positions, and compares the model to human data. Strong players are better than weak players in both types of positions, especially with long presentation times, but even after brief presentations. CHREST predicts the data, both qualitatively and quantitatively. Strong players' superiority with random positions is explained by the large number of chunks they hold in LTM. Their excellent recall with short presentation times is explained by templates, a special class of chunks. CHREST is compared to other theories of chess skill, which either cannot account for the superiority of Masters in random positions or predict too strong a performance of Masters in such positions.     

The congruency effect: just what is being learned?

G. H. Bower, S. Thompson-Schill, and E. Tulving (1994) found that when stimulus-response sets in A-B, A-C learning belong to unique categories (congruent-triads), learning appropriate responses appear rapid and memory performance on a modified modified free recall (MMFR) test is enhanced. Bower et al. assumed that category cues protect associations from interpair interference, allowing more rapid learning. However, unlike arbitrary pairs, congruent pairs also allow a reliance on preexperimental associations. As a result, MMFR test performance may not be an unbiased test of what was learned. In the present experiment, free recall (FR) demonstrated that responses were learned approximately equally in all conditions and that the pattern of clustering was compatible with the hypothesis that preexperimental associations continue to play a major role in FR test performance.     

Evaluation and development of a connectionist theory of configural learning

A configural theory of associative learning is described that is based on the assumption that conditioning results in associations between the unconditioned stimulus and a representation of the entire pattern of stimulation that was present prior to its delivery. Configural theory was formulated originally to account for generalization and discrimination in Pavlovian conditioning. The first part of the article demonstrates how this theory can be used to explain results from studies of overshadowing, blocking, summation, and discrimination learning. The second part of the article shows how the theory can be developed to explain a broader range of phenomena, including mediated conditioning, reinforcer devaluation effects, the differential outcomes effect, acquired equivalence, sensory preconditioning, and structural discriminations.     

Cue selection in verbal discrimination learning of retarded subjects.

A series of experiments was conducted to examine cue function in trigram verbal discrimination learning by retarded subjects. The issue was to determine the factors that control attention in this type of learning situation. The two variables of chief interest were trigram meaningfulness and reinforcement history. The major general conclusion was that retarded subjects employ complex cue selection strategies in solving a verbal discrimination involving compound verbal stimuli. Major findings were as follows: (a) Retarded subjects exhibit a response bias in favor of words over nonsense trigrams; (b) relatively little active cue selection based on meaningfulness was observed; (c) a compound stimulus discrimination was more difficult to learn than a discrimination involving single verbal stimuli; (d) compound and single stimuli may be processed differently in original discrimination learning; (e) cue position probably controls attention in compound discrimination learning; and (f) a frequency theory of verbal discrimination learning is supported by these data.     

The effect of task difficulty and criteria of learning on a subsequent reversal

Two experiments using rats were carried out in which it was shown that a quadratic function can best describe the relation between amount of initial discrimination training and speed of reversal learning for both a difficult visual and an easy spatial task. The results are used to explain the rarity of the over learning reversal effect (ORE) using easy tasks such as position discriminations. Implications for the attention theory are also discussed.