Category vs. Object Knowledge in Category-based Induction

A comparative investigation of integral- and separable-dimension stimulus-sorting behavior

Studies focusing on unsupervised categorization and sorting behavior have traditionally identified and empirically isolated a few select strategies frequently employed by observers, one-dimensional (SD), family resemblance (FR), and exclusive-or (XOR). To date, these investigations have mostly involved creating sets of multidimensional stimuli that directly contrast utilization of these strategies coupled with task or stimulus property manipulations to see their effects on categorization and sorting behavior. Currently, we extend on this methodological approach by having observers sort integral-dimension stimuli for two recently developed constrained unsupervised categorization tasks employing three-dimensional Boolean category structures. These structures instantiate six different sorting strategies, including the aforementioned SD, FR, and XOR strategies. Additionally, we connect the prevalence of the strategies observed across both tasks to previous investigations employing separable stimuli with the same tasks and underlying Boolean category structures. In comparison, our results indicate significant reductions in rule-like sorting behavior (SD or XOR) across multiple structures with the integral stimuli. Associated with this decrease in SD and XOR behavior was a corresponding increase in FR and FR-related sorting behavior. Finally, we assess how well generalized representational information theory and the simplicity model can account for the pattern of results across both experimental paradigms and stimulus sets. In general, a formal model derived from "GRIT" outperforms the simplicity model, accounting for nearly the entire range of unsupervised sorting behavior (SD, FR, XOR) observed across the separable and integral tasks.

Inductive generalization with familiar categories: developmental changes in children's reliance on perceptual similarity and kind information

Inductive generalization is ubiquitous in human cognition; however, the factors underpinning this ability early in development remain contested. The present study was designed to (1) test the predictions of the naïve theory and a similarity-based account and (2) examine the mechanism by which labels promote induction. In Experiment 1, 3- to 5-year-old children made inferences about highly familiar categories. The results were not fully consistent with either theoretical account. In contrast to the predictions of the naïve theory approach, the youngest children in the study did not ignore perceptually compelling lures in favor of category-match items; in contrast to the predictions of the similarity-based account, no group of participants favored perceptually compelling lures in the presence of dissimilar-looking category-match items. In Experiment 2 we investigated the mechanisms by which labels promote induction by examining the influence of different label types, namely category labels (e.g., the target and category-match both labeled as bird) and descriptor labels (e.g., the target and the perceptual lure both labeled as brown) on induction performance. In contrast to the predictions of the naïve theory approach, descriptor labels but not category labels affected induction in 3-year-old children. Consistent with the predictions of the similarity-based account, descriptor labels affected the performance of children in all age groups included in the study. The implications of these findings for the developmental account of induction are discussed.

A taxonomy of inductive problems

Inductive inferences about objects, features, categories, and relations have been studied for many years, but there are few attempts to chart the range of inductive problems that humans are able to solve. We present a taxonomy of inductive problems that helps to clarify the relationships between familiar inductive problems such as generalization, categorization, and identification, and that introduces new inductive problems for psychological investigation. Our taxonomy is founded on the idea that semantic knowledge is organized into systems of objects, features, categories, and relations, and we attempt to characterize all of the inductive problems that can arise when these systems are partially observed. Recent studies have begun to address some of the new problems in our taxonomy, and future work should aim to develop unified theories of inductive reasoning that explain how people solve all of the problems in the taxonomy.

Decision making under uncertain categorization

Two experiments investigated how category information is used in decision making under uncertainty and whether the framing of category information influences how it is used. Subjects were presented with vignettes in which the categorization of a critical item was ambiguous and were asked to choose among a set of actions with the goal of attaining the desired outcome for the main character in the story. The normative decision making strategy was to base the decision on all possible categories; however, research on a related topic, category-based induction, has found that people often only consider a single category when making predictions when categorization is uncertain. These experiments found that subjects tend to consider multiple categories when making decisions, but do so both when it is and is not appropriate, suggesting that use of multiple categories is not driven by an understanding of whether categories are relevant to the decision. Similarly, although a framing manipulation increased the rate of multiple-category use, it did so in situations in which multiple-category use both was and was not appropriate.

Pseudocontingencies derived from categorically organized memory representations

Pseudocontingencies (PCs) allow for inferences about the contingency between two variables X and Y when the conditions for genuine contingency assessment are not met. Even when joint observations X i and Y i about the same reference objects i are not available or are detached in time or space, the correlation r(X i ,Y i ) is readily inferred from base rates. Inferred correlations are positive (negative) if X and Y base rates are skewed in the same (different) directions. Such PC inferences afford useful proxies for actually existing contingencies. While previous studies have focused on PCs due to environmental base rates, the present research highlights memory organization as a natural source of PC effects. When information about two attributes X and Y is represented in a hierarchically organized categorical memory code, as category-wise base rates p(X) and p(Y), the reconstruction of item-level information from category base rates will naturally produce PC effects. Three experiments support this contention. When the yes base rates of two respondents in four questionnaire subscales (categories) were correlated, recalled and predicted item-level responses were correlated in the same direction, even when the original responses to specific items within categories were correlated in the opposite direction.

Category inference as a function of correlational structure, category discriminability, and number of available cues

Two experiments investigated category inference when categories were composed of correlated or uncorrelated dimensions and the categories overlapped minimally or moderately. When the categories minimally overlapped, the dimensions were strongly correlated with the category label. Following a classification learning phase, subsequent transfer required the selection of either a category label or a feature when one, two, or three features were missing. Experiments 1 and 2 differed primarily in the number of learning blocks prior to transfer. In each experiment, the inference of the category label or category feature was influenced by both dimensional and category correlations, as well as their interaction. The number of cues available at test impacted performance more when the dimensional correlations were zero and category overlap was high. However, a minimal number of cues were sufficient to produce high levels of inference when the dimensions were highly correlated; additional cues had a positive but reduced impact, even when overlap was high. Subjects were generally more accurate in inferring the category label than a category feature regardless of dimensional correlation, category overlap, or number of cues available at test. Whether the category label functioned as a special feature or not was critically dependent upon these embedded correlations, with feature inference driven more strongly by dimensional correlations.

An integrated account of generalization across objects and features

Humans routinely make inductive generalizations about unobserved features of objects. Previous accounts of inductive reasoning often focus on inferences about a single object or feature: accounts of causal reasoning often focus on a single object with one or more unobserved features, and accounts of property induction often focus on a single feature that is unobserved for one or more objects. We explore problems where people must make inferences about multiple objects and features, and propose that people solve these problems by integrating knowledge about features with knowledge about objects. We evaluate three computational methods for integrating multiple systems of knowledge: the output combination approach combines the outputs produced by these systems, the distribution combination approach combines the probability distributions captured by these systems, and the structure combination approach combines a graph structure over features with a graph structure over objects. Three experiments explore problems where participants make inferences that draw on causal relationships between features and taxonomic relationships between animals, and we find that the structure combination approach provides the best account of our data.

The uncertain status of Bayesian accounts of reasoning

Bayesian accounts are currently popular in the field of inductive reasoning. This commentary briefly reviews the limitations of one such account, the Rational Model (Anderson 1991b), in explaining how inferences are made about objects whose category membership is uncertain. These shortcomings are symptomatic of what Jones & Love (J&L) refer to as “fundamentalist” Bayesian approaches.

Speeded induction under uncertainty: the influence of multiple categories and feature conjunctions

When people are uncertain about the category membership of an item (e.g., Is it a dog or a dingo?), research shows that they tend to rely only on the dominant or most likely category when making inductions (e.g., How likely is it to befriend me?). An exception has been reported using speeded induction judgments where participants appeared to use information from multiple categories to make inductions (Verde, Murphy, & Ross, 2005). In two speeded induction studies, we found that participants tended to rely on the frequency with which features co-occurred when making feature predictions, independently of category membership. This pattern held whether categories were considered implicitly (Experiment 1) or explicitly (Experiment 2) prior to feature induction. The results converge with other recent work suggesting that people often rely on feature conjunction information, rather than category boundaries, when making inductions under uncertainty.