Semisupervised category learning facilitates the development of automaticity

Abstract

In the human category of learning, learning is studied in a supervised, an unsupervised, or a semisupervised way. The rare human semisupervised category of learning studies all focus on early learning. However, the impact of the semisupervised category learning late in learning, when automaticity develops, is unknown. Therefore, in Experiment 1, all participants were first trained on the information-integration category structure for 2 days until they reached an expert level. Afterwards, half of the participants learned in a supervised way and the other half in a semisupervised way over two successive days. Both groups received an equal number of feedback trials. Finally, all participants took part in a test day where they were asked to respond as quickly as possible. Participants were significantly faster on this test in the semisupervised group than in the supervised group. This difference was not found on day 2, implying that the no-feedback trials in the semisupervised condition facilitated automaticity. Experiment 2 was designed to test whether the higher number of trials in the semisupervised condition of Experiment 1 caused the faster response times. We therefore created an almost supervised condition where participants almost always received feedback (95%) and an almost unsupervised condition where participants almost never received feedback (5%). All conditions now contained an equal number of trials to the semisupervised condition of Experiment 1. The results show that receiving feedback almost always or almost never led to slower response times than the semisupervised condition of Experiment 1. This confirms the advantage of semisupervised learning late in learning.

When unsupervised training benefits category learning

“…That is, if subjects relied on the correct dimension throughout the experiment, then the unsupervised statistical information on this dimension was available to them and it alone could support the simple refinement of the category boundary. By contrast, more mixed results were reported in tasks that investigated semi-supervised learning with two-dimensional stimuli ( McDonnell et al, 2012 , Rogers et al, 2010 , Vandist et al, 2009 , Vandist et al, 2019 ). This would be the case if subjects were unable to pay attention to both dimensions equally, as is needed in order to use the statistical information appropriately (especially in information-integration tasks).…”

“…By contrast, work that investigated whether subjects’ categorisation performance improved when intermixing supervised with unsupervised training trials in two-dimensional tasks did not find conclusive evidence: Unsupervised trials have been reported to have no effect on categorisation accuracy in tasks that require subjects to integrate information from only one (rule-based; McDonnell et al, 2012 ) or multiple stimulus dimensions (information integration; Vandist et al, 2009 ), or only to affect response speed ( Vandist et al, 2019 ), or only to benefit learning under time pressure ( Rogers et al, 2010 ), or only to boost generalisation performance in relational category learning if unsupervised and supervised stimuli are similar ( Patterson & Kurtz, 2018 ).…”

“…In addition, there are substantial differences in experimental designs (e.g., tasks, presentation times, response requirements) making it difficult to determine the source of the observed variability. In fact, the mixed results have been taken as evidence that semi-supervised learning may only be beneficial under limited conditions (e.g., under time pressure or late in learning; Vandist et al, 2019 ) and that perhaps supervised items enjoy a special status, which is why they appear to be weighed more strongly in learning ( Lake and McClelland, 2011 , McDonnell et al, 2012 , Vandist et al, 2009 , Zhu et al, 2010 ).…”

When unsupervised training benefits category learning

“…By contrast, work that investigated whether subjects' categorisation performance improved when intermixing supervised with unsupervised training trials in two-dimensional tasks did not find conclusive evidence: Unsupervised trials have been reported to have no effect on categorisation accuracy in tasks that require subjects' to integrate information from only one (rule-based) or multiple stimulus dimensions (information integration) [Vandist et al, 2009, McDonnell et al, 2012, or only to affect response speed [Vandist et al, 2019], or only to benefit learning under time pressure , or only to boost generalisation performance in relational category learning if unsupervised and supervised stimuli are similar [Patterson and Kurtz, 2018].…”

“…These results show that while the idea that humans should be able to boost supervised information by integrating it with unsupervised information appears convincing, the literature only provides sparse and conflicting evidence, corrupted by substantial differences in experimental designs (e.g., tasks, presentation times, response requirements). In fact, the mixed results have been taken as evidence that semi-supervised learning may only be beneficial under limited conditions (e.g., under time pressure or late in learning; Vandist et al [2019]) and that perhaps supervised items enjoy a special status, which is why they appear to be weighed more strongly in learning [McDonnell et al, 2012, Lake and McClelland, 2011, Vandist et al, 2009.…”