An examination of the neural correlates and behavioural phenomena of category learning
Carpenter, Kathryn Louise
Thesis or dissertation
University of Exeter
This thesis investigates the neurobiological pathways that underpin learning of visual categories, and the behaviour associated with these neural systems. The work contains two strands. The first assesses the neural and behavioural predictions of the COmpetition between Verbal and Implicit Systems (COVIS) account of category learning. The second aims to examine the brain regions implicated in the prototype effect after transcranial Direct Current Stimulation (tDCS) to the left dorsolateral prefrontal cortex (DLPFC). COVIS predicts there are separate explicit and implicit category learning systems. According to COVIS, the explicit system optimally learns rule-based (RB) categories and relies upon the frontal lobes for working memory (WM) and executive functioning processes, and the medial temporal lobes (MTL) to store decision boundaries. In contrast, the implicit system employs the basal ganglia to procedurally learn information-integration (II) categories through stimulus-response associations. Experiment 1 found little evidence of separable implicit or explicit systems in an fMRI study that investigated category decision making processes during RB and II category learning using conditions matched in difficulty, category separation and number of relevant stimulus dimensions. Contrary to the predictions of COVIS, the MTL was more active during the II condition compared to the RB condition, an area that should be more engaged by the explicit system. There was also extensive neural activation overlap found between RB and II learning. Experiments 2 and 3 aimed to generalise these neural findings to activation during feedback processing in RB and II conditions. Experiment 2 was a behavioural study which showed that adding a feedback delay necessary for fMRI data analysis did not differentially impact RB or II learning. Experiment 3, including this feedback delay, found the same neural pattern of results as Experiment 1 offering further support that the MTL is more engaged in II learning than RB learning. There was also again considerable overlap in the regions involved in the two tasks. Taken together, Experiments 1 to 3 found no evidence for the neurally dissociable category learning systems predicted by COVIS. Experiments 4, 5 and 6 investigated the behavioural dissociation reported by Smith et al. (2014) that deferring feedback to the end of a six trial block selectively impairs II learning compared to a unidimensional RB condition. Experiment 4 replicated this result. However, when equating the number of dimensions relevant for RB and II learning in Experiment 5, both conditions were hindered by deferring feedback, with Experiment 6 confirming that conjunctive RB learning was impaired by deferred feedback compared to immediate feedback. I concluded that the dissociation reported by Smith et al. is attributed to the use of a unidimensional category as a comparison for II performance, and that when the number of relevant stimulus dimensions between conditions are controlled there is little evidence for the separable systems of COVIS. Experiment 7 used tDCS to investigate if RB or II learning was differentially affected by anodal stimulation to the left DLPFC. Although there was no significant difference in learning between category conditions, during anodal stimulation participants improved less across blocks than those receiving sham stimulation. While the results suggest that the effect of tDCS on RB and II learning may be more tangible during stimulation, the numerical pattern of the data warrants further research into the possibility that RB participants are more affected by tDCS than II participants after stimulation to the left DLPFC. Strand 2 of this thesis aimed to further previous work that suggests anodal stimulation to the DLPFC during a prototype distortion task induced a prototype effect (better responding to unseen prototype trials than other category exemplars derived from this prototype) that was not present in sham participants. Contrary to this past work, Experiments 8 and 9 found that anodal stimulation to the left DLPFC inhibited a prototype effect that was present in sham participants. Experiment 10 implemented a combined tDCS and fMRI task and found that anodal participants engaged the stimulated DLPFC and the MTL more than sham participants in measures of the prototype effect. Based on these findings, this thesis argues that anodal tDCS to the left DLPFC inhibits perceptual learning by disrupting error prediction processes. Anodal participants are also considered to use generalization more than sham participants when perceiving category exemplars, a process attributed to the MTL.
Economic and Social Research Council
Carpenter, K. L., Wills, A. J., Benattayallah, A., & Milton, F. (2016). A Comparison of the neural correlates that underlie rule‐based and information‐integration category learning. Human brain mapping, 37(10), 3557-3574.
McLaren, I. P. L., Carpenter, K., Civile, C., McLaren, R., Milton, F., Verbruggen, F., ... & Ku, Y. (2016). Categorisation and Perceptual Learning: Why tDCS to Left DLPFC Enhances Generalisation. Associative Learning and Cognition. Homage to Professor NJ Mackintosh. In Memoriam (1935-2015), 37-69.
PhD in Psychology