Robert M. Nosofsky

Choice, similarity, and the context theory of classification.

Medin and Schaffers (1978) context theory of classification learning is interpreted in terms of Luces (1963) choice theory and in terms of theoretical results obtained in multidimensional scaling theory. En route to this interpretation, quantitative relationships that may exist between identification and classification performance are investigated. It is suggested that the same basic choice processes may operate in the two paradigms but that the similarity parameters that determine performance change systematically according to the structure of the choice paradigm. In particular, when subjects are able to attend selectively to the component dimensions that compose the stimuli, the similarity parameters may tend toward what is optimal for maximizing performance.

Rule-plus-exception model of classification learning

The authors propose a rule-plus-exception model (RULEX) of classification learning. According to RULEX, people learn to classify objects by forming simple logical rules and remembering occasional exceptions to those rules. Because the learning process in RULEX is stochastic, the model predicts that individual Ss will vary greatly in the particular rules that are formed and the exceptions that are stored. Averaged classification data are presumed to represent mixtures of these highly idiosyncratic rules and exceptions. RULEX accounts for numerous fundamental classification phenomena, including prototype and specific exemplar effects, sensitivity to correlational information, difficulty of learning linearly separable versus nonlinearly separable categories, selective attention effects, and difficulty of learning concepts with rules of differing complexity. RULEX also predicts distributions of generalization patterns observed at the individual subject level.

Dissociations Between Categorization and Recognition in Amnesic and Normal Individuals: An Exemplar-Based Interpretation

In recent work, the finding of dissociations between categorization and recognition in amnesic and normal individuals has been taken as evidence of multiple memory systems mediating these tasks. The present research provides support for the alternative idea that these dissociations can be interpreted in terms of a single-system exemplar-memory model that makes allowance for parameter differences across groups. In one experiment, a parameter change in memory sensitivity was induced by testing classification and recognition at varying delays; the results closely matched the ones observed by Knowlton and Squire (1993) for normal and amnesic participants. The exemplar model also yielded good quantitative predictions of the categorization-recognition dissociation. A second analysis demonstrated that dissociations between early versus late probabilistic classification learning and memory sensitivity were also well predicted by the single-system exemplar model. Limitations of the exemplar interpretation and future research directions are also discussed.

Rules and Exemplars in Categorization, Identification, and Recognition

Subjects learned to classify perceptual stimuli varying along continuous, separable dimensions into rule-described categories. The categories were designed to contrast the predictions of a selective-attention exemplar model and a simple rule-based model formalizing an economy-of-description view. Converging evidence about categorization strategies was obtained by also collecting identification and recognition data and by manipulating strategies via instructions. In free-strategy conditions, the exemplar model generally provided an accurate quantitative account of identification, categorization, and recognition performance, and it allowed for the interrelationship of these paradigms within a unified framework. Analyses of individual subject data also provided some evidence for the use of rules, but in general, the rules seemed to have a great deal in common with exemplar storage processes. Classification and recognition performance for subjects given explicit instructions to use specific rules contrasted dramatically with performance in the free-strategy conditions and could not be predicted by the exemplar model.

Comparing modes of rule-based classification learning: A replication and extension of Shepard, Hovland, and Jenkins (1961)

We partially replicate and extend Shepard, Hovland, and Jenkinss (1961) classic study of task difficulty for learning six fundamental types of rule-based categorization problems. Our main results mirrored those of Shepard et al., with the ordering of task difficulty being the same as in the original study. A much richer data set was collected, however, which enabled the generation of block-by-block learning curves suitable for quantitative fitting. Four current computational models of classification learning were fitted to the learning data: ALCOVE (Kruschke, 1992), the rational model (Anderson, 1991), the configural-cue model (Gluck & Bower, 1988b), and an extended version of the configural-cue model with dimensionalized, adaptive learning rate mechanisms. Although all of the models captured important qualitative aspects of the learning data, ALCOVE provided the best overall quantitative fit. The results suggest the need to incorporate some form of selective attention to dimensions in category-learning models based on stimulus generalization and cue conditioning.

The bow and sequential effects in absolute identification

The bow and sequential effects in absolute identification are investigated in this paper by following two strategies: (1) Experiments are performed in which sequential dependencies in signal presentations are manipulated, and 12) analyses are conducted (some of which are largely free of model-specific assumptions) which bear directly on the question of the origin of the sequential effects. The main result of the study is that absolute identification performance is greatly improved in a design in which each signal lies close to the preceding signal presented, even though the entire range of signals used is the same as in a random presentation design. This finding is consistent with the attention-band model of Luce, Green, and Weber (1976) and rejects hypotheses that suggest that the variability in the signal representation in absolute identification is a function solely of the range of signals being used. However, nonparametric analyses of sequential response errors show that a plausible assumption concerning the trial by-trial movement of the attention band provides an incomplete explanation of Seluential effects in absolute identification. These results are far better explained in terms of systematic shifts of category boundaries in a Thurstonian model, as suggested by Purks, Callahan, Braida, and Durlach (1980). Experiments are also performed which suggest that memory decay is not the major factor accounting for the bow effect in absolute identification.

Similarity-scaling studies of dot-pattern classification and recognition

Classification performance in the dot-pattern, prototype-distortion paradigm (e.g., Posner & Keele, 1968) was modeled within a multidimensional scaling (MDS) framework. MDS solutions were derived for sets of dot patterns that were generated from prototypes. These MDS solutions were then used in conjunction with exemplar, prototype, and combined models to predict classification and recognition performance. Across 3 experiments, an MDS-based exemplar model accounted for the effects of several fundamental learning variables, including level of distortion of the patterns, category size, delay of transfer phase, and item frequency. Most important, the model quantitatively predicted classification probabilities for individual dot patterns in the sets, not simply general trends of performance. There was little evidence for the existence of a prototype-abstraction process that operated above and beyond pure exemplar-based generalization.

Seven plus or minus two: a commentary on capacity limitations.

Millers classic 1956 article is best known today for its discussion of capacity limitations in short-term memory, but the bulk of the article dealt with capacity limitations in absolute judgment tasks and the relation of such limitations to information theory. Many of the puzzles of absolute judgment first raised by Miller remain a puzzle today. The authors review some of the literature directed toward this issue and discuss a few models that attempt to elucidate the phenomena. Since 1956 there has been an enormous research effort aimed at understanding the mechanisms and limitations of short-term memory, resulting in considerable progress. The authors briefly discuss some of these advances. The authors conclude, as Miller did, by noting the probable lack of connection between the limitations observed in these 2 areas of inquiry.

A response-time approach to comparing generalized rational and take-the-best models of decision making

The authors develop and test generalized versions of take-the-best (TTB) and rational (RAT) models of multiattribute paired-comparison inference. The generalized models make allowances for subjective attribute weighting, probabilistic orders of attribute inspection, and noisy decision making. A key new test involves a response-time (RT) approach. TTB predicts that RT is determined solely by the expected time required to locate the 1st discriminating attribute, whereas RAT predicts that RT is determined by the difference in summed evidence between the 2 alternatives. Critical test pairs are used that partially decouple these 2 factors. Under conditions in which ideal observer TTB and RAT strategies yield equivalent decisions, both the RT results and the estimated attribute weights suggest that the vast majority of subjects adopted the generalized TTB strategy. The RT approach is also validated in an experimental condition in which use of a RAT strategy is essentially forced upon subjects.

Stimulus Bias, Asymmetric Similarity, and Classification.

Abstract This article proposes that patterns of proximity data that have been characterized in terms of “asymmetric similarity” may be alternatively characterized in terms of differential “bias.” Bias is a characteristic pertaining to an individual object, as opposed to similarity, which is a relation between two objects. It is proposed that biases can be stimulus based as well as response based, and numerous examples are provided. Part 1 of the article reviews an additive similarity and bias model proposed by Holman (1979, Journal of Mathematical Psychology, 20, 1–15), which generalizes various extant models that have successfully characterized asymmetric proximities. Part 1 then discusses relations between asymmetric proximities and differences in self-proximities, and also discusses multidimensional scaling models that are supplemented with stimulus bias terms. Part 2 of the article reviews and integrates a variety of phenomena in the perceptual classification literature involving asymmetries that can be characterized in terms of symmetric similarity together with differential stimulus bias. Part 3 provides examples of limitations of the additive similarity and bias model. A main thesis of the article is that models of proximity and classification data that incorporate properties of the individual stimulus may not always require recourse to the positing of asymmetric similarities.