In order to successfully guide generalization of knowledge, category representation needs to be both: flexible enough to account for new evidence and stable enough to resist harmful change. Here we present a set of experiments designed to test how items that violate our expectations (i.e., category exceptions) affect category representation. Specifically, we wanted to know whether learning a category exception can change category boundaries. Does learning about penguins changes the way we think about birds? Do features of penguins contribute to making decisions as to whether a novel item is a bird? Across two experiments we found evidence that exceptions can change category boundaries and thus significantly affect future generalization. We discuss implications these findings have for the extent models of category learning and memory.

While exceptions to a regularity might be rare, categories that have exceptions are not. Previous studies on learning categories that have exceptions suggested special status of exceptional items in memory (e.g. Palmeri & Nosofsky, 1995, Sakamoto and Love, 2004). However, this might be true only for a special kind of exceptions – those that call for forming complex binding structures, and could be learned only if they are fully memorized. In the two experiments in this study, we show that memory for exceptions is not better than memory for regular category members (Experiment 1). On the contrary, both children and adults had better memory for the features of regular items (Experiment 2). In addition, adults, but not 4- year-olds, showed better memory for the rule than for probabilistic features. The overall results challenge the idea of the special status of exceptions in memory.

In this paper we explored how people represent categories that include exceptions by examining contributions that features of regular and exception items make to determining category membership. We examined performance of 4-year-old children and adults and found significant developmental differences. While for 4-year-olds, deterministic features of regular items and exceptions contributed comparably to determining category membership, an asymmetry was found in adults. For adults, deterministic features of regular items contributed more to determining category membership than features of exceptions. The results are discussed in relation to the SUSTAIN clustering model of category learning (Love, Medin, & Gureckis, 2004).

Many hallmarks of human intelligence including language, reasoning, and planning require us to draw upon knowledge about the world in which concepts, denoted by words, are organized by meaningful, semantic links between them (e.g., juicy-apple-pear). The goal of the present research was to investigate how these organized semantic networks may emerge in development from simple but powerful mechanisms sensitive to statistical co- occurrence regularities of word use in language. Specifically, we tested whether a mechanistic account of how co-occurrence regularities shape semantic development accurately predicts how semantic organization changes with development. Using a sensitive, gaze-based measure of the semantic links organizing knowledge in children and adults, we observed that developmental changes in semantic organization were consistent with a key role for statistical co-occurrence regularities.

With development, we acquire rich body of knowledge aboutthe world in which concepts denoted by words (e.g., juicy,apple, and pear) are connected by meaningful, semantic links(e.g., apples and pears are similar, and can both be juicy). Onepotentially powerful driver of this development is sensitivity toregularities with which words co-occur in language.Specifically, language is rich regularities that can support: (1)Associative semantic links between words that directly co-occur together (e.g., juicy-apple), and (2) Taxonomic semanticlinks between words similar in meaning that share patterns ofdirect co-occurrence (e.g., apple and pear both co-occur withjuicy). Here, we investigated the development of abilities toform semantic links from these regularities. Results revealedthat both children and adults formed direct co-occurrence-based links, whereas only adults formed shared co-occurrencebased links. We discuss how these results may provide keyinsight into how semantic organization develops.

Research has long investigated how knowledge about the world is connected by meaningful, semantic links. Much of this research has focused on a specific type of semantic link known as a taxonomic link, which connects concepts belonging to the same semantic category. However, many inconsistencies have emerged regarding how and when this specific type of semantic link is formed. The goal of the present study was to investigate this contradicting research and provide an explanation for the inconsistent results. To do this, we examined the linguistic environment of stimuli sets from three studies that either supported or did not support protracted taxonomic development. Results provided evidence that the idea that semantic links between members of taxonomic categories in early development may be based on simple co-occurrences in language.

A wealth of statistical learning research has provided evidence that regularities in which items co-occur (referred to hereas syntagmatic) can be learned implicitly. However, it is not known whether higher-order relations can also be learnedimplicitly. Here we present two experiments that investigate whether regularities, where items do not co-occur but insteadshare co-occurrence with each other (referred to here as paradigmatic), can be learned implicitly. In Experiment 1, weused a traditional auditory statistical learning paradigm where participants passively listened to an auditory stream con-taining syntagmatic and paradigmatic regularities and found evidence only of syntagmatic learning. In Experiment 2, weinstructed participants to attend to items during the training session and found evidence of learning paradigmatic relationsin participants who demonstrated high-level of syntagmatic learning. The results are discussed in terms of the limits ofimplicit learning and the role of attentional mechanisms in learning higher-order statistical regularities.

Extensive statistical learning literature suggests that regularities between co-occurring items can be learned implicitly.However, little is known whether higher-order statistics, such as paradigmatic relations, can be learned implicitly. Paradig-matic relations link items that may not co-occur but share each others patterns of co-occurrence. For example, A and Bare paradigmatically related when they both co-occur with C (e.g., A-C, B-C). Therefore, paradigmatic relations could,in principle, be implicitly learned through co-occurrence regularities. Here, we modified a contextual cueing task, wheresome of the targets independently appeared within distractors that had the same spatial configuration. We found that onlyparticipants who noticed the repeating distractor patterns tend to learn the paradigmatic relations, while most participantswere able to learn the simple co-occurring regularities. Our findings imply that the ability to learn simple co-occurrenceregularities is not sufficient for forming paradigmatic relations and that explicit attention may be critical.

Language is rich in statistical regularities that capture meaningful, semantic links between words crucial for languagefluency. Words that can be combined to express meaningful ideas (e.g., drink-soda) reliably directly co-occur together,and words similar in meaning share patterns of co-occurrence (e.g. soda and milk share co-occurrence with drink). Here,we investigate whether children (4-year-olds) and adults can capitalize on these regularities to form new semantic linksbetween new and familiar words. Participants hear sentences in which new words directly co-occur or share co-occurrencewith familiar words. We then assess the formation of corresponding semantic links using an implicit, gaze-based measureand an explicit labeling measure. Results suggest that new semantic links form only from direct co-occurrence in chil-dren, and from both direct and shared co-occurrence in adults. This research is therefore uncovering the development ofmechanisms for building organized word knowledge from mere exposure to language.