Multimodal discrimination learning in humans: Evidence for configural theory

Human contingency learning was used to compare the predictions of configural and elemental theories. In three experiments, participants were required to learn which indicators were associated with an increase in core temperature of a fictitious nuclear plant. Experiments 1 and 2 investigated the rate at which a triple-element stimulus (ABC) could be discriminated from either single-element stimuli (A, B, and C) or double-element stimuli (AB, BC, and AC). Experiment 1 used visual stimuli, whilst Experiment 2 used visual, auditory, and tactile stimuli. In both experiments the participants took longer to discriminate the triple-element stimulus from the more similar double-element stimuli than from the less similar single-element stimuli. Experiment 3 tested for summation with stimuli from either a single or multiple modalities, and summation was found only in the latter case. Thus, the pattern of results seen in Experiments 1 and 2 was not dependent on whether the stimuli were single modal or multimodal, nor was it dependent on whether the stimuli could elicit summation. This pattern of results is consistent with predictions derived from Pearce's (1987, 1994) configural theory.     

Multimodal discrimination learning in humans: evidence for configural theory

Human contingency learning was used to compare the predictions of configural and elemental theories. In three experiments, participants were required to learn which indicators were associated with an increase in core temperature of a fictitious nuclear plant. Experiments 1 and 2 investigated the rate at which a triple-element stimulus (ABC) could be discriminated from either single-element stimuli (A, B, and C) or double-element stimuli (AB, BC, and AC). Experiment 1 used visual stimuli, whilst Experiment 2 used visual, auditory, and tactile stimuli. In both experiments the participants took longer to discriminate the triple-element stimulus from the more similar double-element stimuli than from the less similar single-element stimuli. Experiment 3 tested for summation with stimuli from either a single or multiple modalities, and summation was found only in the latter case. Thus, the pattern of results seen in Experiments 1 and 2 was not dependent on whether the stimuli were single modal or multimodal, nor was it dependent on whether the stimuli could elicit summation. This pattern of results is consistent with predictions derived from Pearce's (1987, 1994) configural theory.     

Transfer of configural learning between the components of a preexposed stimulus compound: implications for elemental and configural models of learning

In 2 experiments, rats received preexposure to 2 compound contexts: AB and CD for the congruent group and AC and BD for the incongruent group. Subsequently, all rats received a configural discrimination in which separate placement in contexts A or B indicated that presentations of stimulus X would be followed by food and presentations of Y would not, and separate placement in contexts C and D indicated that Y would be followed by food and X would not. In both experiments, rats in the congruent group acquired the conditional discrimination more rapidly than those in the incongruent group. These results are inconsistent with conventional associative accounts of either stimulus preexposure effects or configural learning and instead provide support for a connectionist account.     

Summation: Further assessment of a configural theory

Rats received Pavlovian conditioning in which food was signalled by a visual stimulus, A+, an auditory stimulus, B+, and a compound composed of different visual and auditory stimuli, CD+. Test trials were then given with the compound AB. Experiments 1 and 2A revealed stronger responding during AB than during CD. In Experiment 2B, there was no evidence of a summation of responding during AB when A+ B+ training was conducted in the absence of CD+ trials. A further failure to observe abnormally strong responding during ABwas found in Experiment 3 for which the training trials with A+ B+ CD+ were accompanied by trials in which C and D were separately paired with food. The results are explained in terms of a configural theory of conditioning, which assumes that responding during a compound is determined by generalization from its components, as well as from other compounds to which it is similar.     

Configural learning without configural training.

Rats learned discriminations in which 2 familiar compound visual stimuli AB and CD were nonreinforced, whereas less-familiar compounds (EF, GH, IJ...) were reinforced (constant-negative paradigm). Such discriminations can in principle be acquired through elemental (nonconfigural) learning. Three different types of compound--object/object, object/position, and shape/fill-pattern--were used, and the number of familiar compounds extended to 3 or 4. In all cases, when rats were tested for the first time with previously unseen recombinations of the familiar elements--for example, AC, BD--they preferred these to the familiar compounds, implying that they had encoded configural information during the original training. Moreover, preference remained constant despite extended exposure to configural training in the test phase. The findings extend the evidence that configural representations are formed independently of explicit configural training.     

A connectionist theory of phenomenal experience

When cognitive scientists apply computational theory to the problem of phenomenal consciousness, as many have been doing recently, there are two fundamentally distinct approaches available. Consciousness is to be explained either in terms of the nature of the representational vehicles the brain deploys or in terms of the computational processes defined over these vehicles. We call versions of these two approaches vehicle and process theories of consciousness, respectively. However, although there may be space for vehicle theories of consciousness in cognitive science, they are relatively rare. This is because of the influence exerted, on the one hand, by a large body of research that purports to show that the explicit representation of information in the brain and conscious experience are dissociable, and on the other, by the classical computational theory of mind--the theory that takes human cognition to be a species of symbol manipulation. Two recent developments in cognitive science combine to suggest that a reappraisal of this situation is in order. First, a number of theorists have recently been highly critical of the experimental methodologies used in the dissociation studies--so critical, in fact, that it is no longer reasonable to assume that the dissociability of conscious experience and explicit representation has been adequately demonstrated. Second, classicism, as a theory of human cognition, is no longer as dominant in cognitive science as it once was. It now has a lively competitor in the form of connectionism; and connectionism, unlike classicism, does have the computational resources to support a robust vehicle theory of consciousness. In this target article we develop and defend this connectionist vehicle theory of consciousness. It takes the form of the following simple empirical hypothesis: phenomenal experience consists of the explicit representation of information in neurally realized parallel distributed processing (PDP) networks. This hypothesis leads us to reassess some common wisdom about consciousness, but, we argue, in fruitful and ultimately plausible ways.     

Prejudiced learning: A connectionist account

Connectionist simulation was employed to investigate processes that may underlie the relationships between prior expectancies or prejudices and the acquisition of attitudes, under conditions where learners can only discover the valence of attitude objects through directly experiencing them. We compared contexts analogous to learners holding either false negative expectancies (‘prejudices’) about a subclass of objects that were actually good or false positive expectancies about objects that were actually bad. We introduced expectancy‐related bias either by altering the probability of approach, or by varying the rate of learning following experience with good or bad objects. Where feedback was contingent on approach, the false positive expectancies were corrected by experience, but negative prejudices resisted change, since the network avoided objects deemed to be bad, and so received less corrective feedback. These findings are discussed in relation to the effects of intergroup contact and expectancy‐confirmation processes in reducing or sustaining prejudice.     

Cue interactions in flavor preference learning: a configural analysis

Four experiments showed that the preference normally established to a neutral flavor cue that was paired with maltodextrin was attenuated when that cue was conditioned in compound with another flavor--overshadowing. Furthermore, two experiments showed that the preference for a neutral flavor conditioned as part of a compound was further attenuated if the other element in that compound was separately paired with the reinforcer--blocking. These results stand in contrast to a number of previous compound flavor preference conditioning experiments, which have not revealed reliable cue competition effects. These discrepant findings are discussed in terms of the effects of within-compound associations and a configural perspective on potentiation. Modeling of this configural perspective predicts that a compound of two separately trained cues will elicit a similar response to the individual cues themselves--absence of summation. Two experiments confirmed this prediction.     

Navigation assistance: a trade-off between wayfinding support and configural learning support

Current GPS-based mobile navigation assistance systems support wayfinding, but they do not support learning about the spatial configuration of an environment. The present study examined effects of visual presentation modes for navigation assistance on wayfinding accuracy, route learning, and configural learning. Participants (high-school students) visited a university campus for the first time and took a predefined assisted tour. In Experiment 1 (n = 84, 42 females), a presentation mode showing wayfinding information from eye-level was contrasted with presentation modes showing wayfinding information included in views that provided comprehensive configural information. In Experiment 2 (n = 48, 24 females), wayfinding information was included in map fragments. A presentation mode which always showed north on top of the device was compared with a mode which rotated according to the orientation of the user. Wayfinding accuracy (deviations from the route), route learning, and configural learning (direction estimates, sketch maps) were assessed. Results indicated a trade-off between wayfinding and configural learning: Presentation modes providing comprehensive configural information supported the acquisition of configural knowledge at the cost of accurate wayfinding. The route presentation mode supported wayfinding at the cost of configural knowledge acquisition. Both presentation modes based on map fragments supported wayfinding. Individual differences in visual-spatial working memory capacity explained a considerable portion of the variance in wayfinding accuracy, route learning, and configural learning. It is concluded that learning about an unknown environment during assisted navigation is based on the integration of spatial information from multiple sources and can be supported by appropriate visualization.     

Effects of similarity and experience on discrimination learning: a nonassociative connectionist model of perceptual learning.

This article describes a novel connectionist model of perceptual learning (PL) that provides a mechanism for nonassociative differentiation (J. J. Gibson & E. J. Gibson, 1955). The model begins with the assumption that 2 processes--1 that decreases associability and 1 that increases discriminability--operate during preexposure (S. Channell & G. Hall, 1981). In contrast to other models (e.g., I. P. L. McLaren, H. Kaye, & N. J. Mackintosh, 1989), in the current model the mechanisms for these processes are compatible with a configural model of associative learning. A set of simulations demonstrates that the present model can account for critical PL phenomena such as exposure learning and effects of similarity on discrimination. It is also shown that the model can explain the paradoxical result that preexposure to stimuli can either facilitate or impair subsequent discrimination learning. Predictions made by the model are discussed in relation to extant theories of PL.     

Combining exemplar-based category representations and connectionist learning rules.

Adaptive network and exemplar-similarity models were compared on their ability to predict category learning and transfer data. An exemplar-based network (Kruschke, 1990a, 1990b, 1992) that combines key aspects of both modeling approaches was also tested. The exemplar-based network incorporates an exemplar-based category representation in which exemplars become associated to categories through the same error-driven, interactive learning rules that are assumed in standard adaptive networks. Experiment 1, which partially replicated and extended the probabilistic classification learning paradigm of Gluck and Bower (1988a), demonstrated the importance of an error-driven learning rule. Experiment 2, which extended the classification learning paradigm of Medin and Schaffer (1978) that discriminated between exemplar and prototype models, demonstrated the importance of an exemplar-based category representation. Only the exemplar-based network accounted for all the major qualitative phenomena; it also achieved good quantitative predictions of the learning and transfer data in both experiments.     

A configural analysis for feature-negative discrimination learning.

In 4 experiments, rats were initially trained with an A+ AXo discrimination in which Stimulus A by itself signaled the delivery of food, A+, whereas the simultaneous presentation of A and X was followed by nothing, AXo. In each experiment X was then paired with food prior to a test phase in which A and X were again presented for a discrimination. The discrimination was of the form A+ AXo in Experiments 1 and 2, whereas it was of the form X+ AXo for Experiments 3 and 4. In all 4 experiments the test discrimination was acquired more rapidly than a control discrimination. The results are interpreted in terms of the original A+ AXo discrimination resulting in the growth of an association between a representation of the entire AX compound and the effects of nonreinforcement.     

Generative connectionist networks and constructivist cognitive development

This article presents a novel computational framework for modeling cognitive development. The new modeling paradigm provides a language with which to compare and contrast radically different facets of children's knowledge. Concepts from the study of machine learning are used to explore the power of connectionist networks that construct their own architectures during learning. These so-called generative algorithms are shown to escape from Fodor's (1980) critique of Constructivist development. We describe one generative connectionist algorithm (cascade-correlation) in detail. We report on the successful use of the algorithm to model cognitive development on balance scale phenomena; seriation; the integration of velocity, time, and distance cues; prediction of effect sizes from magnitudes of causal potencies and effect resistances; and the acquisition of English personal pronouns. The article demonstrates that computer models are invaluable for illuminating otherwise obscure discussions.     

The relationship between connectionist models and a dynamic data-oriented theory of concept formation

In this paper I shall compare two models of concept formation, both inspired by basic convictions of philosophical empiricism. The first, the connectionist model, will be exemplified by Kohonen maps, and the second will be my own dynamic theory of concept formation. Both can be understood in probabilistic terms, both use a notion of convergence or stabilization in modelling how concepts are built up. Both admit destabilization of concepts and conceptual change. Both do not use a notion of representation in some pregiven language, such as a language of thought or some logical language. Representation in a formal language only plays a role on the meta-level, namely within the theory about concept formation.A short version of this paper was given at the Fachtagung der Gesellschaft für Kognitionswissenschaft. Freiburg, October 12–15th, 1994.     

Overconfidence effects in category learning: a comparison of connectionist and exemplar memory models

Exemplar and connectionist models were compared on their ability to predict overconfidence effects in category learning data. In the standard task, participants learned to classify hypothetical patients with particular symptom patterns into disease categories and reported confidence judgments in the form of probabilities. The connectionist model asserts that classifications and confidence are based on the strength of learned associations between symptoms and diseases. The exemplar retrieval model (ERM) proposes that people learn by storing examples and that their judgments are often based on the first example they happen to retrieve. Experiments 1 and 2 established that overconfidence increases when the classification step of the process is bypassed. Experiments 2 and 3 showed that a direct instruction to retrieve many exemplars reduces overconfidence. Only the ERM predicted the major qualitative phenomena exhibited in these experiments.     

Multialternative decision field theory: a dynamic connectionist model of decision making

The authors interpret decision field theory (J. R. Busemeyer & J. T. Townsend, 1993) as a connectionist network and extend it to accommodate multialternative preferential choice situations. This article shows that the classic weighted additive utility model (see R. L. Keeney & H. Raiffa, 1976) and the classic Thurstone preferential choice model (see L. L. Thurstone, 1959) are special cases of this new multialternative decision field theory (MDFT), which also can emulate the search process of the popular elimination by aspects (EBA) model (see A. Tversky, 1969). The new theory is unique in its ability to explain several central empirical results found in the multialternative preference literature with a common set of principles. These empirical results include the similarity effect, the attraction effect, and the compromise effect, and the complex interactions among these three effects. The dynamic nature of the model also implies strong testable predictions concerning the moderating effect of time pressure on these three effects.     

A connectionist account of asymmetric category learning in early infancy

Young infants show unexplained asymmetries in the exclusivity of categories formed on the basis of visually presented stimuli. A connectionist model is described that shows similar exclusivity asymmetries when categorizing the same stimuli presented to infants. The asymmetries can be explained in terms of an associative learning mechanism, distributed internal representations, and the statistics of the feature distributions in the stimuli. The model was used to explore the robustness of this asymmetry. The model predicts that the asymmetry will persist when a category is acquired in the presence of mixed category exemplars. An experiment with 3-4-month-olds showed that asymmetric exclusivity persisted in the presence of mixed-exemplar familiarization, thereby confirming the model's prediction.     

ALCOVE: an exemplar-based connectionist model of category learning.

ALCOVE (attention learning covering map) is a connectionist model of category learning that incorporates an exemplar-based representation (Medin & Schaffer, 1978; Nosofsky, 1986) with error-driven learning (Gluck & Bower, 1988; Rumelhart, Hinton, & Williams, 1986). Alcove selectively attends to relevant stimulus dimensions, is sensitive to correlated dimensions, can account for a form of base-rate neglect, does not suffer catastrophic forgetting, and can exhibit 3-stage (U-shaped) learning of high-frequency exceptions to rules, whereas such effects are not easily accounted for by models using other combinations of representation and learning method.     

Comparing elemental and configural associative theories in human causal learning: a case for attention

In two causal learning experiments with human participants, the authors compared various associative theories that assumed either elemental (unique cue, modified unique cue, replaced elements model, and Harris' model) or configural processing of stimuli (Pearce's theory and a modification of it). The authors used modified patterning problems initially suggested by Redhead and Pearce (1995). Predictions for all theories were generated by computer simulations. Both configural theories and the unique cue approach failed to account for the observations. The replaced elements model was able to account for part of the data, but only if the replacement parameters could vary across discrimination problems. The Harris model and the modified unique cue approach, assuming that the salience of stimuli decreases with an increasing number of stimuli in a compound, successfully accounted for all of our data. This success implies that attentional factors should be explicitly taken into account in associative learning theory.