Multidimensional same--different judgments: evidence against independent comparisons of dimensions

Commonly discussed models of performance in the multidimensional same-different task are based on the assumption that independent same-different decisions are reached for each of the stimulus dimensions used in an experiment. In these models, a higher level mechanism examines th outcome of these independent decisions in order to determine whether the stimuli are the same or different overall. Previous evidence indicating that these models are inconsistent with results from same trials is summarized, and a new test of the above assumption using data from different trials is proposed. Detailed analysis of reaction times from different trials is shown, at least for some pairs of dimensions, to be incompatible with the view that separate same-different decisions are reached for each dimension. The major finding is that there are more relatively fast responses in a condition with two dimensions different, compared with the two conditions with only one of the dimensions different, than any processing model treating the dimensions separately could predict. The results suggest that partial information may be combined across dimensions in order to reach a "different" judgment.     

Folded additive structures: A nonpolynomial model of some factorial interactions

A nonpolynomial measurement model for representing nonadditive compositions of stimulus dimensions is presented. This representation, which is referred to as a folded additive structure, permits the decomposition of a behavioral ordering into independent factors in some of the cases where the behavior itself does not directly reflect this independence. Psychologically, this representation corresponds to a model of behavior which assumes that (1) the effect stimulus attributes have on behavior is mediated by a single relevent psychological dimension, (2) the behavioral ordering is single-peaked over this mediating dimension and (3) the mediating dimension has a conjoint additive ordering over a set of relevent stimulus dimensions. Necessary and sufficient tests for folded additive structures are given. In addition, five psychological theories which make these assumptions are discussed.     

Garner interference is not solely driven by stimulus uncertainty

The speeded classification tasks popularized by Garner (1974), with their accompanying labels of separable and integral dimensions, have become the dominant paradigm for characterizing perceptual interactions between stimulus dimensions. Separable dimensions like color and shape can be selectively attended to at will, but integral dimensions like hue and brightness cannot. When classifying stimuli with respect to a single dimension, integral stimuli lead to faster performance in the baseline block, where only that single dimension varies, than in the filtering block, in which a second dimension is allowed to vary irrelevantly. No such difference is predicted for separable dimensions lead to no difference. The comparison between baseline and filtering confounds a change in the number of stimuli with a change in the number of variant dimensions. A new experimental condition is proposed utilizing three stimulus dimensions to hold the number of stimuli constant while introducing variance along a second irrelevant dimension. Data indicate that interference increases, suggesting that changes in the number of potential stimuli within a block are not necessary for Garner interference, in contrast to the accounts provided by extant quantitative models.     

Multidimensional visual statistical learning

Recent studies of visual statistical learning (VSL) have demonstrated that statistical regularities in sequences of visual stimuli can be automatically extracted, even without intent or awareness. Despite much work on this topic, however, several fundamental questions remain about the nature of VSL. In particular, previous experiments have not explored the underlying units over which VSL operates. In a sequence of colored shapes, for example, does VSL operate over each feature dimension independently, or over multidimensional objects in which color and shape are bound together? The studies reported here demonstrate that VSL can be both object-based and feature-based, in systematic ways based on how different feature dimensions covary. For example, when each shape covaried perfectly with a particular color, VSL was object-based: Observers expressed robust VSL for colored-shape sub-sequences at test but failed when the test items consisted of monochromatic shapes or color patches. When shape and color pairs were partially decoupled during learning, however, VSL operated over features: Observers expressed robust VSL when the feature dimensions were tested separately. These results suggest that VSL is object-based, but that sensitivity to feature correlations in multidimensional sequences (possibly another form of VSL) may in turn help define what counts as an object.     

Fear generalization in humans: Impact of verbal instructions

Fear generalization lies at the heart of many anxiety problems, but little is known about the factors that can influence this phenomenon. The present study investigated whether verbal instructions about specific stimulus features can influence conditioned fear generalization. All participants were fear conditioned to a yellow triangle, using an electric shock. Participants had received pre-experimental instructions saying that the shapes (or colours) of the stimuli were informative for the occurrence of shock (group Shape and group Colour, respectively). Next, generalization was tested to presentations of a blue triangle (same shape) as well as a yellow square (same colour). Fear reactions were measured through skin conductance and online ratings of shock-expectancy. The results showed strongest generalization to the same shape stimulus in group Shape, versus the same colour stimulus in group Colour. Hence, the same learning experience can have opposite effects in terms of fear generalization, depending on verbally transmitted information about the relative importance of individual stimulus features.     

Naïve and Robust: Class‐Conditional Independence in Human Classification Learning

Humans excel in categorization. Yet from a computational standpoint, learning a novel probabilistic classification task involves severe computational challenges. The present paper investigates one way to address these challenges: assuming class‐conditional independence of features. This feature independence assumption simplifies the inference problem, allows for informed inferences about novel feature combinations, and performs robustly across different statistical environments. We designed a new Bayesian classification learning model (the dependence‐independence structure and category learning model, DISC‐LM) that incorporates varying degrees of prior belief in class‐conditional independence, learns whether or not independence holds, and adapts its behavior accordingly. Theoretical results from two simulation studies demonstrate that classification behavior can appear to start simple, yet adapt effectively to unexpected task structures. Two experiments—designed using optimal experimental design principles—were conducted with human learners. Classification decisions of the majority of participants were best accounted for by a version of the model with very high initial prior belief in class‐conditional independence, before adapting to the true environmental structure. Class‐conditional independence may be a strong and useful default assumption in category learning tasks.     

Incidental learning of S-R contingencies in the masked prime task

Subliminal motor priming effects in the masked prime paradigm can only be obtained when primes are part of the task set. In 2 experiments, the authors investigated whether the relevant task set feature needs to be explicitly instructed or could be extracted automatically in an incidental learning paradigm. Primes and targets were symmetrical arrows, with target color, not shape, the response-relevant feature. Shape and color covaried for targets (e.g., <> always blue, >< always green), whereas primes were always black. Over time, a negative compatibility effect (NCE; response benefits when prime and target had different shapes) developed, indicating that primes affected the motor system. When target shape and color varied independently (control condition), no NCE occurred, in line with the assumption that the NCE reflects task set-dependent motor processes, not perceptual interactions.     

An inverse base rate effect with continuously valued stimuli

It is well known that people do not always make normative use of information about relative frequencies of categories when making categorical judgments. The "inverse base rate" effect (Medin & Edelson, 1988) is a typical example of this: Subjects violate normative reasoning principles by assigning certain ambiguous stimuli as belonging to the less frequent of two categories, rather than to the more common category. This effect has been explained as being due to the shifting of attention from shared stimulus features to distinctive features during learning. When stimuli are defined by values along continuous dimensions, rather than by the presence and absence of features, then attention could shift between dimensions or between values, or both. In three experiments, base rate differences were used to determine the way in which attention is shifted during learning about stimuli with continuously valued dimensions. Simulation modeling shows that the results are consistent with the movement of attention both between and within stimulus dimensions.     

Attentional processing and the independence of color and orientation

Judgments of the color and shape of a stimulus specified by a cue indicating its location demonstrate stochastic independence. Evidence for the independence of color and orientation was obtained when the cue followed stimulus offset immediately, the cue followed stimulus offset by 2 s, and when either a central or peripheral precue focused attention on the cued stimulus at the time of its presentation. It appears that an object's color and orientation are represented independently even following the attentional focusing thought to support feature integration.     

A confluence of contexts: asymmetric versus global failures of selective attention to stroop dimensions.

In 6 experiments probing selective attention through Stroop classification, 4 factors of context were manipulated: (a) psychophysical context, the distinctiveness of values along the color and word dimensions; (b) set size context, the number of stimulus values tested; (c) production context, the mode used to respond; and (d) covariate context, the correlation between the dimensions. The psychophysical and production contexts mainly caused an asymmetry in selective attention failure between colors and words, whereas the set size and covariate contexts contributed primarily to the average or global magnitudes of attentional disruption across dimensions. The results suggest that (a) Stroop dimensions are perceptually separable, (b) J.R. Stroop's (1935) classic findings arose from his particular combination of contexts, and (c) stimulus uncertainty and dimensional imbalance are the primary sources of task and congruity effects in the Stroop paradigm.     

A Dissociation Between Attention and Selection

It is widely assumed that the allocation of spatial attention results in the "selection" of attended objects or regions of space. That is, once a stimulus is attended, all its feature dimensions are processed irrespective of their relevance to behavioral goals. This assumption is based in part on experiments showing significant interference for attended stimuli when the response to an irrelevant dimension conflicts with the response to the relevant dimension (e.g., the Stroop effect). Here we show that such interference is not due to attending per se. In two spatial cuing experiments, we found that it was possible to restrict processing of attended stimuli to task-relevant dimensions. This new evidence supports two novel conclusions: (a) Selection involves more than the focusing of attention per se; and (b) task expectations play a key role in determining the depth of processing of the elementary feature dimensions of attended stimuli.     

The importance of shape in early lexical learning

We ask if certain dimensions of perceptual similarity are weighted more heavily than others in determining word extension. The specific dimensions examined were shape, size, and texture. In four experiments, subjects were asked either to extend a novel count noun to new instances or, in a nonword classification task, to put together objects that go together. The subjects were 2-year-olds, 3-year-olds, and adults. The results of all four experiments indicate that 2- and 3-year-olds and adults all weight shape more heavily than they do size or texture. This observed emphasis on shape, however, depends on the age of the subject and the task. First, there is a developmental trend. The shape bias increases in strength and generality from 2 to 3 years of age and more markedly from early childhood to adulthood. Second, in young children, the shape bias is much stronger in word extension than in nonword classification tasks. These results suggest that the development of the shape bias originates in language learning—it reflects a fact about language—and does not stem from general perceptual processes.     

Latencies to name one of three stimulus dimensions: A study of probability effects and dimension integrality

Following each of 400 stimuli that varied according to three binary dimensions (location, color, and shape), subjects named the value of one particular dimension as quickly as possible. Each stimulus appeared on the “left” and “right” side of a screen (location dimension), and was a “red” or “green” triangle (color dimension) with apex oriented “up” or “down” (shape dimension). The alternatives of each dimension varied according to different probability distributions (90/10, 70/30, or 50/50), and three schedule conditions were contrived so that each distribution occurred once for each dimension. Nine groups of 10 subjects each defined the factorial of 3 (response conditions: identify location, color, or shape) by 3 (schedule conditions: 90 left/70 red/50 down, 70 left/50 red/90 down, or 50 left/90 red/70 down). Latencies to identify the shape dimension were influenced reliably by the probabilities of both response-relevant and response-irrelevant stimuli. Shape probabilities did not affect latencies to identify location of color, but these latencies were significantly influenced by both color and location probabilities. Results are discussed with reference to the locus of the stimulus probability effect and dimensional integrality.     

On the relationship between memory and perception: sequential dependencies in recognition memory testing

Many models of recognition are derived from models originally applied to perception tasks, which assume that decisions from trial to trial are independent. While the independence assumption is violated for many perception tasks, we present the results of several experiments intended to relate memory and perception by exploring sequential dependencies in recognition. The findings from these experiments disconfirm the independence assumption for recognition memory. In addition, the pattern of sequential dependencies observed in recognition differs from that observed for many perception tasks. This suggests that sequential dependencies arise from mnemonic or perceptual processes and not from decision processes that should be common to memory and perception tasks.     

Electrophysiological evidence for temporal overlap among contingent mental processes.

A series of studies assessed perceptual-motor transmission of stimulus information by measuring lateralization of movement-related brain potentials in a choice reaction task with no-go trials. When stimuli varied in shape and size, lateralized potentials on no-go trials suggested that easily recognized shape information was used to initiate motor preparation and that this preparation was aborted when size analysis signified that the response should be withheld. This indicates that movement preparation can begin once partial perceptual information about a stimulus becomes available, contrary to an assumption of fully discrete models of information processing. By contrast, when stimuli varied only in size, no evidence for preliminary response preparation was obtained, contrary to an assumption of fully continuous models but consistent with asynchronous discrete coding models (Miller, 1982, 1988).     

Classification errors and response times over multiple distributed sessions as a function of category structure

Learning difficulty orderings for categorical stimuli have long provided an empirical foundation for concept learning and categorization research. The conventional approach seeks to determine learning difficulty orderings in terms of mean classification accuracy. However, it is relatively rare that the stability of such orderings is tested over a period of extended learning. Further, research rarely explores dependent variables beyond classification accuracy that may also indicate relative learning difficulty, such as classification response times (RTs). Using a family of category structures defined over three binary dimensions and four positive examples that is well-known for its robust learning difficulty ordering, we report the results of two experiments that test the stability of the ordering (in terms of both errors and RTs) over multiple category learning sessions. The experimental stimuli consisted of instantiations of each of the six category structures in the family. These take the form of categories consisting of four “flasks” that vary along the binary features of size (large or small), shape (circular or triangular), and color (black or white). Experiment 1 shows that when participants are randomly presented instances of all six types, the difficulty ordering remains stable across all three sessions. This stability is present in terms of mean accuracy (errors) as well as mean RTs. In Experiment 2, participants were repeatedly exposed to category instances of a single type. In terms of errors, the ordering is revealed in the first session and disappears in later sessions. The opposite trend is observed for classification RTs: The ordering is not present in the first session but is revealed in later sessions. This suggests that even when individuals reach a relative degree of expertise in terms of reduced errors, the original degree of difficulty continues to influence processing. We interpret these results in the context of the concept learning and perceptual expertise literatures.     

Top-down task sets for combined features: Behavioral and electrophysiological evidence for two stages in attentional object selection

We studied whether visual search for targets defined by a combination of features from different dimensions is guided by separately represented target features or by an integrated representation of the target objects. In Experiment 1, participants searched for target singleton bars that were defined by a specific combination of color (red or blue) and size (small or large). The target arrays were preceded by cue arrays that contained a spatially uninformative color/size singleton. Behavioral spatial-cueing effects indicative of attentional capture were triggered only by cues that matched both target-defining features, but not by partially target-matching cues, suggesting that attention was guided by integrated object representations. However, the presence of reliable N2pc components for partially matching cues demonstrated that these cues did capture attention, in line with independent feature-based guidance of attention. This dissociation between the electrophysiological and behavioral markers of attentional capture was confirmed in Experiment 2, in which targets were defined by a color/size disjunction. Our results suggest that the attentional selection of targets that are defined by a combination of features is a two-stage process: Attention is initially captured by all target-matching features, but is then rapidly withdrawn from nontarget objects that share some but not all features with the current target.     

Attention and learning processes in the identification and categorization of integral stimuli

The relationship between subjects' identification and categorization learning of integral-dimension stimuli was studied within the framework of an exemplar-based generalization model. The model was used to predict subjects' learning in six different categorization conditions on the basis of data obtained in a single identification learning condition. A crucial assumption in the model is that because of selective attention to component dimensions, similarity relations may change in systematic ways across different experimental contexts. The theoretical analysis provided evidence that, at least under unspeeded conditions, selective attention may play a critical role in determining the identification-categorization relationship for integral stimuli. Evidence was also provided that similarity among exemplars decreased as a function of identification learning. Various alternative classification models, including prototype, multiple-prototype, average distance, and "value-on-dimensions" models, were unable to account for the results.     

Independent encoding of colors and shapes from two stimuli

Observers were presented with brief exposures of pairs of colored objects (letters) and asked to report both the color and the shape of each object. Several observers showed strikingly clear evidence of nearly perfect stochastic independence between reports of the four features (two colors and two shapes). For instance, the probability that the shape of a given object could be reported seemed independent of (a) whether the color of the object could be reported and (b) whether features of the other object could be reported. Such stochastic independence is predicted by many parallel-processing models (e.g., Bundesen, 1990). However, the results are difficult to reconcile with simple serial models in which the encoding of one object is completed before the encoding of another object is begun.     

Learning in small connectionist networks does not generalize to large networks

Summary A number of recent models of human information processing have been based on connectionist architectures. Such models are designed to illustrate specific psychological principles and are usually implemented in small networks. The assumption implicit in the work is that principles illustrated in small networks can be applied easily to brain-size networks by scaling up as required. To consider the scaling question, we used a back-propagation algorithm to compare learning in both large and small networks and found that learning depended on the size of the network. In small networks, increasing (the rate-of-learning parameter) beyond 1 increased the rate of learning; in large networks, the same manipulation reduced the rate of learning. The example illustrates the difficulty of generalizing across network size and calls into question the assumption that principles illustrated in small networks can be applied to the brain by expansion of the network.Formerly at Queen's University at Kingston