W. K. Estes

Array Models for Category Learning

Abstract A family of models for category learning is developed, all members being based on a common memory array but differing in memory access and decision processes. Within this framework, fully controlled comparisons of exemplar-similarity, feature-frequency, and prototype models reveal isomorphism between models of different types under some conditions but empirically testable differences under others. It is shown that current exemplar-memory models, in which categorization judgments are based on similarities of perceived and remembered category exemplars, can be interpreted as generalized likelihood models but can be modified in a simple way to yield pure similarity models. Distance-based exemplar models are formulated that provide means of investigating issues concerning deterministic versus probabilistic decision rules and links between categorization and properties of perceptual dimensions. Other theoretical issues discussed include aspects of similarity, the role of memory storage versus computation in category judgments, and the limits of applicability of array models.

Interactions of signal and background variables in visual processing

Three variables which determine the opportunities for signal-noise confusions, display size (D), number of redundant signals per display (N), and number of alternative signals (A) were studied in relation to nature of the noise elements, confusable or nonconfusable with signals. Data were obtained in a forced-choice visual detection situation, the displays being linear arrays of letters on a CRT screen. For all three performance measures used, frequency of correct detections and correct and error latencies, strong interactions were obtained between all of the other variables and signal-noise confusability. The functions obtained, together with other data bearing on the role of confusions and on spatial relations among characters within the display, suggest a model whose initial phase is a parallel feature extraction process involving inhibitory relations among input channels.

Stimulus sampling theory

Abstract : Some of the fundamental mathematical techniques of stimuls sampling theory are presented. The simplest of all learning models - the pattern model for simple learning is presented in which the population of available stimulation is ASSUMED TO COMPRISE A SET OF DISTINCT STIMULUS PATTERNS, EXACTLY ONE OF WHICH IS SAMPLED ON EACH TRIAL. In the important special case of the oneelement model, it is assumed that there is only one such pattern and that it recurs intact at the beginning of each experimenta trial. The oneelement model is worthy of study not only for expositional purposes but also for its value as an analytic device in relation to certain types of learning data. After a treatment of pattern models for simple acquisition an for learning under probabilistic reinforcement schedules, the conceptualization of generalization and transfer is discussed; the component models in which the patterns of stimulation effective on individual trials are treated, not as distinct elements, but as overlapping samples from a common population; and, finally, some examples of the more complex multiple-process models which are becoming increasingly important in the analysis of discrimination learning, concept formation, and related phenomena are discussed. (Author)

Order and Position in Primary Memory for Letter Strings.

A short-term recall task in which to-be-remembered letters were spaced out in different ways by distractor digits, and all were shadowed during input to preclude rehearsal, was used to separate aspects of primary and secondary memory processes and to permit examination of primary memory for position, order, and item independently. Under conditions contrived to minimize the contribution of secondary processes, memory for order appeared entirely derivative to memory for temporal position and all of the principal trends in both aspects were well predicted by an associative coding model assuming a reverberatory recycling process. Variation in intrastring acoustic confusability of letters yielded a complex pattern of interactions that seems unexplainable by any extant model but suggests specific lines of augmentation of the associative coding model.

Serial position functions for letter identification at brief and extended exposure durations

The properties of serial position functions for tachistoscopic report were investigated over a wide range of viewing times. Four-letter strings of random consonants were presented in varying display locations relative to the fixation point with the observers’ eye movements monitored to limit them to a single fixation for each display. Salient properties of the serial position curves include an overall central-peripheral gradient, higher performance at the ends than the interior of letter strings regardless of absolute location, and left-right asymmetry in the visual field, all of these being largely independent of viewing time. Errors reflecting loss of positional information are prominent even at extended viewing times, are more nearly symmetrical in the left and right visual fields than other types of errors, and, in contrast to item errors, occur less frequently in letter sequences that have high frequencies in English. Further, transposition errors exhibit a pronounced peripheral-to-central drift, possibly reflecting gradients of positional uncertainty. Such gradients may be implicated in the peripheral-central asymmetry of the lateral interference effects exerted by other letters on a target letter in a nonfoveal location.

Memory storage and retrieval processes in category learning.

The detailed course of learning is studied for categorization tasks defined by independent or contingent probability distributions over the features of category exemplars. College-age subjects viewed sequences of bar charts that simulated symptom patterns and responded to each chart with a recognition and a categorization judgment. Fuzzy, probabilistically defined categories were learned relatively rapidly when individual features were correlated with category assignment, more slowly when only patterns carried category information. Limits of performance were suboptimal, evidently because of capacity limitations on judgmental processes as well as limitations on memory. Categorization proved systematically related to feature and exemplar probabilities, under different circumstances, and to similarity among exemplars of categories. Unique retrieval cues for exemplar patterns facilitated recognition but entered into categorization only at retention intervals within the range of short-term memory. The findings are interpreted within the framework of a general array model that yields both exemplar-similarity and feature-frequency models as special cases and provides quantitative accounts of the course of learning in each of the categorization tasks studied.

Risks of drawing inferences about cognitive processes from model fits to individual versus average performance

With the goal of drawing inferences about underlying processes from fits of theoretical models to cognitive data, we examined the trade off of risks of depending on model fits to individual performance versus risks of depending on fits to averaged data with respect to estimation of values of a model’s parameters. Comparisons based on several models applied to experiments on recognition and categorization and to artificial, computer-generated data showed that results of using the two types of model fitting are strongly determined by two factors: model complexity and number of subjects. Reasonably accurate information about true parameter values was found only for model fits to individual performance and then only for some of the parameters of a complex model. Suggested guidelines are given for circumventing a variety of obstacles to successful recovery of useful estimates of a model’s parameters from applications to cognitive data.

Phonemic coding and rehearsal in short-term memory for letter strings

Several new procedures were used to control vocalization during display of letter strings and rehearsal during retention intervals. Clear evidence of phonemic encoding of visually presented letters was obtained when vocalization was permitted at input but was absent when vocalization was suppressed by a categorizing task or by reduced exposure durations. However, under the latter condition, phonemic encoding occurred if rehearsal was permitted immediately following input, indicating that representations of letters can be maintained in a nonauditory form in memory for at least 1 sec following a visual display in spite of the overwriting of successive characters in the same location. Both vocalization during input and minimal rehearsal during retention intervals exerted effects on item availability and memory for order beyond those interpretable in terms of phonemic encoding.

Traps in the route to models of memory and decision

It is proposed that the products of investigations of learning, memory, and decision over the last half century that are most likely to endure have resulted from interactions between models and experimental research. In this article, some of the traps that must be coped with to make fruitful interactions possible are examined and illustrated with case studies from research on probability learning, category learning, and recognition memory. Topics addressed include functions of models in research; the logic of model testing; fitting models to signal plus noise; values and hazards of averaging data; and potential contributions of neural science to the development of cognitive models.

On the communication of information by displays of standard errors and confidence intervals

A survey of practices regarding the presentation of information about reliability of means in psychological research publications over the last century reveals some advance in quality of communication, greater for tabular presentations than for graphic presentations, but also substantial room for improvement. In this article, problems of interpretation and communication associated with presentations of standard errors and confidence intervals in research reports are examined from both statistical and psychological perspectives. Four general principles of effective communication are proposed and illustrated in application to presentations of data from common psychological research designs, with special attention to problems arising in connection with repeated measures.