Sorting out categories: Incremental learning of category structure

Two experiments examine how inferences might promote unsupervised and incremental category learning. Many categories have members related through overall similarity (e.g., a family resemblance structure) rather than by a defining feature. However, when people are asked to sort category members in a category construction task, they often do so by partitioning on a single feature. Starting from an earlier result showing that pairwise inferences increase family resemblance sorting (Lassaline & Murphy, 1996), we examine how these inferences lead to learning the family resemblance structure. Results show that the category structure is learned incrementally. The pairwise inferences influence participants’ weightings of feature pairs that were specifically asked about, which in turn affects their sorting. The sorting then allows further learning of the categorical structure. Thus, the inferences do not directly lead learners to the family resemblance structure, but they do provide a foundation to build on as the participants make additional judgments.     

On the nature of implicit categorization

Current categorization models disagree about whether people make a priori assumptions about the structure of unfamiliar categories. Data from two experiments provided strong evidence that people do not make such assumptions. These results rule out prototype models and many decision bound models of categorization. We review previously published neuropsychological results that favor the assumption that category learning relies on a procedural-memory-based system, rather than on an instance- based system (as is assumed by exemplar models). On the basis of these results, a new categorylearning model is proposed that makes no a priori assumptions about category structure and that relies on procedural learning and memory.     

Relational properties and memory-based category construction.

Although many experiments have investigated factors that constrain perceptual category construction, there have been no investigations of factors that constrain memory-based (MB) category construction. Six experiments examined the extent to which perceptual and MB sorting were influenced by correlated dimensions, family resemblance principles, and conceptual knowledge. Sensitivity to many types of relational information (e.g., correlated features, causal relations, interactive properties of objects, and family resemblance relations) was observed with perceptual sorting, but these properties were rarely used to organize information in MB sorting conditions. Instead, there was a clear preference to organize categories around single dimensions. Even when perfectly correlated features were causally related, Ss in memory conditions did not use correlations to construct categories. The strengths and limitations of MB analyses and categorizations are discussed.     

The impact of training sequence and between-category similarity on unsupervised induction

Studies of supervised categorization often show better learning when examples are presented in random alternation rather than massed by category, but such interleaving impairs learning in unsupervised tasks. The exemplar comparison hypothesis explains this result by assuming that people in unsupervised tasks discover generalizations about categories by comparing individual examples, and that interleaving increases the difficulty of such within-category comparisons. The category invention hypothesis explains the interleaving effect by assuming that people are more likely to merge or aggregate potentially separable categories when they are interleaved, and this initial failure to recognize separate categories then acts as an effective barrier to further learning. The present experiments show that the interleaving effect depends on the similarity or alignability of the presented categories. This result provides evidence in favour of the category invention hypothesis, which expects that highly dissimilar (nonalignable) categories will resist aggregation and hence will not be affected by interleaving. The nonmonotonic pattern of learning, and the interaction between sequence and similarity, observed in the alignable conditions of Experiment 3 were also consistent with category invention, but not with exemplar comparison. Implications are discussed for real-world learning, especially the relationship between exposure and learning and between supervised and unsupervised learning.     

A probabilistic account of exemplar and category generation

People are capable of imagining and generating new category exemplars and categories. This ability has not been addressed by previous models of categorization, most of which focus on classifying category exemplars rather than generating them. We develop a formal account of exemplar and category generation which proposes that category knowledge is represented by probability distributions over exemplars and categories, and that new exemplars and categories are generated by sampling from these distributions. This sampling account of generation is evaluated in two pairs of behavioral experiments. In the first pair of experiments, participants were asked to generate novel exemplars of a category. In the second pair of experiments, participants were asked to generate a novel category after observing exemplars from several related categories. The results suggest that generation is influenced by both structural and distributional properties of the observed categories, and we argue that our data are better explained by the sampling account than by several alternative approaches.     

Essentialist to some degree: beliefs about the structure of natural kind categories

Previous research has provided conflicting evidence regarding the hypothesis that people are essentialists. Much of the evidence in favor of essentialism is based on demonstrating that categories are thought to have absolute membership. Although the hypothesis is often framed as an absolute claim about all categories of a certain type (e.g., natural kinds), it has generally been tested by making relative comparisons with a select sample. The present study assesses judgments of absolute structure across a range of categories. A further condition for essentialism is that the criteria for category identity be seen as objective rather than conventional. The results of three experiments based on these considerations do not provide support for essentialist claims. Few categories were judged to have essentialist structure, in terms of either absolute membership or objective criteria. Results are discussed in light of an alternative to the essentialist hypothesis that emphasizes a pragmatic view of categories.     

A New Zero-Inflated Negative Binomial Methodology for Latent Category Identification

We introduce a new statistical procedure for the identification of unobserved categories that vary between individuals and in which objects may span multiple categories. This procedure can be used to analyze data from a proposed sorting task in which individuals may simultaneously assign objects to multiple piles. The results of a synthetic example and a consumer psychology study involving categories of restaurant brands illustrate how the application of the proposed methodology to the new sorting task can account for a variety of categorization phenomena including multiple category memberships and for heterogeneity through individual differences in the saliency of latent category structures.     

Judgments relative to patterns: how temporal sequence patterns affect judgments and memory

Six experiments studied relative frequency judgment and recall of sequentially presented items drawn from 2 distinct categories (i.e., city and animal). The experiments show that judged frequencies of categories of sequentially encountered stimuli are affected by certain properties of the sequence configuration. We found (a) a first-run effect whereby people overestimated the frequency of a given category when that category was the first repeated category to occur in the sequence and (b) a dissociation between judgments and recall; respondents may judge 1 event more likely than the other and yet recall more instances of the latter. Specifically, the distribution of recalled items does not correspond to the frequency estimates for the event categories, indicating that participants do not make frequency judgments by sampling their memory for individual items as implied by other accounts such as the availability heuristic (Tversky & Kahneman, 1973) and the availability process model (Hastie & Park, 1986). We interpret these findings as reflecting the operation of a judgment heuristic sensitive to sequential patterns and offer an account for the relationship between memory and judged frequencies of sequentially encountered stimuli.     

Induction and category coherence

In studies of category formation, subjects rarely construct family resemblance categories. Instead, they divide objects into categories using a single dimension. This is a puzzling result given the widely accepted view that natural categories are organized in terms of a family resemblance principle. The observation that natural categories support inductive inferences is used here to test the hypothesis that family resemblance categories would be constructed if stimuli were first used to generate inductive inferences. In two experiments, subjects answered either induction questions, which made interproperty relationships more salient, or frequency questions, which required information only about individual properties, before they performed a sorting task. Subjects were likely to produce family resemblance sorts if they had first answered induction questions but not if they had answered frequency questions.     

The perception of identity in simultaneously presented complex visual displays

Two experiments were conducted in order to examine the information processing in a visual matching task, using digit sequences of varying complexity as the stimuli. Traditionally, reaction times for “same” judgments do not fit into a single-process self-terminating feature testing model, while those for “different” judgments do. Bamber (1969) proposed a two-stage model to account for the data, and the results of these experiments support this type of model. Strong evidence implying that Bamber’s “identity reporter” has a limited capacity in terms of stimulus complexity was also found. This complexity seems to be deemed by stimulus discriminability and the number of “chunks” of information rather than by “bits” of information being transmitted (Miller, 1956).     

The generalized polymorphous concept account of graded structure in abstract categories

Abstract categories present with graded structure. The extent to which feature commonality between exemplars and category provides a satisfying account of this graded structure varies from one abstract category to the other (Hampton, 1981). We investigate whether the incorporation of features that exemplars share with external categories yields an improved account of abstract categories' graded structures. In doing so, we follow the suggestion that abstract categories are relational in nature (Goldstone, 1996; Wiemer-Hastings & Xu, 2005). The generalized polymorphous concept model, which incorporates both types of features, is found to improve the account of typicality and category membership in three of seven studied abstract categories. These three categories are found to be the most abstract, suggesting that it is appropriate to think of abstract categories as varying along a continuum of abstractness/interrelatedness rather than as a distinct type of category altogether.     

Decision processes in verifying category membership statements: Implications for models of semantic memory

Current models of semantic memory assume that natural categories are well-defined. Specific predictions of two such models, the Smith, Shoben, and Rips (1974a) two-stage feature comparison model and the Glass and Holyoak (1974/75) ordered search model, were tested and disconfirmed in Experiment I. We propose an alternative model postulating fuzzy categories represented as sets of characteristic properties. This model, combined with a Bayesian decision process, accounts for the results of three additional experiments, as well as for the major findings in the semantic memory literature. We argue that people verify category membership statements by assessing similarity relations between concepts rather than by using information which logically specifies the truth value of the sentence. Our data also imply that natural categories are fuzzy rather than well-defined.     

The role of similarity in deductive categorical inference

The category inclusion rule specifies that categories inherit the properties of their superordinates. For example, given thatall metals are pentavalent, it can be concluded thatall iron is pentavalent. Sloman (1998) showed that people do not fully endorse conclusions that follow from the category inclusion rule. He claims that people rely on the similarity between the premise and the conclusion categories (metals andiron), rather than applying the category inclusion rule. By allowing reasoners to rate their certainty for category relations (e.g.,iron is metal), as well as for conclusions, the present study shows that similarity has only an indirect effect on the certainty of conclusions: Reasoners are more certain that similar categories have a category inclusion relation, and this in turn affects the certainty of conclusions based on this relation.     

Differential brain responses when applying criterion attribute versus family resemblance rule learning

Subsystems of category learning have been identified on the basis of general domains of content (e.g., tools, faces). The present study examined categories from the standpoint of internal structure and determined brain topography associated with expressing two fundamentally different category rule structures (criterion attribute, CA, and family resemblance, FR). CA category learning involves processing stimuli by isolated features and classifying by properties held by all members. FR learning involves processing stimuli by integral wholes and classifying on overall similarity among members without sharing identical features. fMRI BOLD response to CA and FR categorization was measured with pseudowords as stimuli. Category knowledge for both tasks was mastered prior to brain imaging. Areas of activation emerged unique to the structure of each category and followed from the nature of the rule abstraction procedure. CA categorization was implemented by strong target monitoring and expectation (medial parietal), rule maintenance in working memory, feature selection processes (inferior frontal), and a sensitivity to high frequency components of the stimulus such as isolated features (anterior temporal). FR categorization, consistent with its multi-featural nature, involved word-level processing (left extrastriate) that evoked articulatory rehearsal (medial cerebellar). The data suggest category structure is an important determinant of brain response during categorization. For instance, anterior temporal structures may help attune visual processing systems to high frequency components to support the learning of criterial, highly predictive rules.     

Typhoid Pollyanna: The effect of category valence on retrieval order of positive and negative category members

Previous research suggests that people process pleasant information more efficiently than unpleasant information. This phenomenon has been illustrated in a variety of contexts and paradigms, and is frequently referred to as the Pollyanna principle. One important aspect of the Pollyanna principle is that people tend to retrieve pleasant members of a category from semantic memory prior to unpleasant members of the same category. However, we propose that the retrieval advantage of favourably viewed category members holds only for positive categories and that prototypicality of category members might be a better predictor of ease of retrieval than favourability. These possibilities were tested in two studies. In Study 1, favourability effects on retrieval order were observed for positive categories but not for neutral or negative categories. In Study 2, prototypical category members were retrieved before less prototypical category members for both positive and negative categories. These results suggest that the Pollyanna principle might be the result of a confound between favourability and prototypicality for members of favourable categories.     

Expanding the search for a linear separability constraint on category learning

Formal models of categorization make different predictions about the theoretical importance of linear separability. Prior research, most of which has failed to find support for a linear separability constraint on category learning, has been conducted using tasks that involve learning two categories with a small number of members. The present experiment used four categories with three or nine patterns per category that were either linearly separable or not linearly separable. With overall category structure equivalent across category types, the linearly separable categories were found to be easier to learn than the not linearly separable categories. An analysis of individual participants' data showed that there were more participants operating under a linear separability constraint when learning large categories than when learning small ones. Formal modeling showed that an exemplar model could not account for many of these data. These results are taken to support the existence of multiple processes in categorization.     

类别学习的阻碍效应与双机制理论

通过对类别学习中的阻碍效应进行系统研究,尝试性的提出了类别学习的双机制理论。三个实验分别考察：当样例特征随机呈现时;定义特征固定呈现在样例首位时;以及刺激材料为知觉图形材料时,类别学习中的阻碍效应。三个实验的研究结果都发现：在类别学习中存在一定程度的阻碍效应,支持类别学习同时存在联结学习机制和认知学习机制的双机制观点。     

Novel Labels Increase Category Coherence,But Only When People Have the Goal to Coordinate

From infancy, we recognize that labels denote category membership and help us to identify the critical features that objects within a category share. Labels not only reflect how we categorize, but also allow us to communicate and share categories with others. Given the special status of labels as markers of category membership, do novel labels (i.e., non‐words) affect the way in which adults select dimensions for categorization in unsupervised settings? Additionally, is the purpose of this effect primarily coordinative (i.e., do labels promote shared understanding of how we categorize objects)? To address this, we conducted two experiments in which participants individually categorized images of mountains with or without novel labels, and with or without a goal of coordination, within a non‐communicative paradigm. People who sorted items with novel labels had more similar categories than people who sorted without labels only when they were told that their categories should make sense to other people, and not otherwise. We argue that sorters' goals determine whether novel labels promote the development of socially coherent categories.     

Effect of causal structure on category construction

In four experiments, the question of how the causal structure of features affects the creation of new categories was examined. Features of exemplars to be sorted were related in a single causal chain (causal chain), were caused by the same factor (common cause), or caused the same effect (common effect). The results showed that people are more likely to rely on common-cause or common-effect background knowledge than on causal-chain background knowledge in category construction. Such preferences suggest that the common-cause or the common-effect structures are considered more natural conceptual structures.