Deficits in category learning in older adults: Rule-based versus clustering accounts.

Abstract

Memory research has long been one of the key areas of investigation for cognitive aging researchers but only in the last decade or so has categorization been used to understand age differences in cognition. Categorization tasks focus more heavily on the grouping and organization of items in memory, and often on the process of learning relationships through trial and error. Categorization studies allow researchers to more accurately characterize age differences in cognition: whether older adults show declines in the way in which they represent categories with simple rules or declines in representing categories by similarity to past examples. In the current study, young and older adults participated in a set of classic category learning problems, which allowed us to distinguish between three hypotheses: (a) rule-complexity: categories were represented exclusively with rules and older adults had differential difficulty when more complex rules were required, (b) rule-specific: categories could be represented either by rules or by similarity, and there were age deficits in using rules, and (c) clustering: similarity was mainly used and older adults constructed a less-detailed representation by lumping more items into fewer clusters. The ordinal levels of performance across different conditions argued against rule-complexity, as older adults showed greater deficits on less complex categories. The data also provided evidence against rule-specificity, as single-dimensional rules could not explain age declines. Instead, computational modeling of the data indicated that older adults utilized fewer conceptual clusters of items in memory than did young adults.

Why Higher Working Memory Capacity May Help You Learn: Sampling, Search, and Degrees of Approximation

“…6A). This is consistent with the recent suggestion, in the context of category learning in older adults, that Type II advantage is modulated by WMC (Rabi & Minda, 2016), though our present model does not speak to the observation that relative performance on Type II and Type IV problems may sometimes reverse (e.g., in older adults-see Badham, Sanborn, & Maylor, 2017;Rabi & Minda, 2016).…”

Classification of three-dimensional integral stimuli: Accounting for a replication and extension of Nosofsky and Palmeri (1996) with a dual discrimination invariance model.

“…Last, the DDIM includes parsimonious cognitive mechanisms and explanations for differences in behavior across these structures, whereas the ICM lacks such mechanisms and explanations. Moreover, it may be that GIST can account for some of the alternative findings cited earlier (Badham et al, 2017;Kurtz, Levering, et al, 2013;Minda et al, 2008;Rabi & Minda, 2016) regarding the wording of instructions, mapping of dimensions to instances, and the developmental population under consideration by construing each of the mentioned conditions as moderators of the degree of discrimination between exemplars at a low level or local level of analysis and discrimination between structures at a global level. For example, in the DDIM, with respect to the integral dimensions under study in this article, setting the threshold to its extreme value of 1 corresponds to an extremely low level of discriminability that results in Type IV being relatively easier to learn than Type II.…”

“…2 Related to the latter, Love and Markman (2003) also found that separable dimensions may not always be treated independent from each other (termed "relationally independent"), and this effect also can cause learning of Type IV to be easier than Type II. Finally, the ordering is also sensitive to the developmental population under consideration, with younger children (Minda et al, 2008) and older adults aged beyond 60 years old (Badham et al, 2017) and 65 years old (Rabi & Minda, 2016) displaying an increased difficulty learning the Type II relation compared to the Type IV relation. For a more comprehensive discussion of these differential effects on the learning of the 3 2 [4] structures with different types of stimulus dimensions, including both the separable and integral type of dimensions discussed in the current study, we refer the reader to Sanborn et al (2021).…”

REFRESH: A new approach to modeling dimensional biases in perceptual similarity and categorization.

“…The ordering of the SHJ problems can also be produced using a discrete binary likelihood, but the correspondence of the model to the canonical ordering is parameter dependent and the parameters that produce this ordering are often not those that produce the best match to the overall accuracy level of human performance ( Nosofsky, Gluck, et al, 1994 ; Sanborn et al, 2010 ), though it has been successful on occasion ( Badham et al, 2017 ). The RMC for continuous data is unlikely to be able to produce violations of the triangle inequality because the probability of being a member of a cluster is Gaussian which corresponds to a Euclidean distance metric, as we discuss in the Appendix .…”

“…A reanalysis of the data of Lewandowsky (2011) by Lloyd et al (2019) showed that higher WMC results in a larger advantage of Type II over Type IV. There is some evidence that the ordering can even reverse: comparisons of older and younger adults show that older adults, who have worse working memory generally, perform better on Type IV than Type II problems ( Badham et al, 2017 ; Rabi & Minda, 2016 ). We can match this result by decreasing the number of particles in REFRESH (see Figure 22 ): For one hundred particles Type II is easier than Type IV, but for one particle the ordering reverses.…”