Unstructured Categorization with Probabilistic Feedback: Learning Accuracy Versus Response Time

Abstract

AbstractContradictory data exist whether the category number affects the learning performance in rule-based and integration-information classification tasks. When an effect is observed, the performance is better for a lower number of categories. We aimed to investigate the effect of the category number on the performance in the unstructured category learning tasks with probabilistic feedback. We conducted four experiments. The stimuli consisted of dot motion sequences. We presented eight motion directions (0°–315° through 45°) with motion direction coherence of 75% (Experiments 1, 3, and 4) and 20% (Experiment 2). We used the probabilistic rule of 79% (Experiments 1–3) or 75% (Experiment 4) correct answers. Eight observers classified the eight stimuli into 8 categories (Experiments 1–2); 2 categories (Experiment 3); 4 categories (Experiment 4). The results show: 1.) a wide variety of strategies adopted by the observers; 2.) Accuracy and response time changed at a different rate during learning; 3.) The rate of improvement differed between the experiments; 4.) The response time is a better characteristic of incremental category learning. The findings imply that the learning performance depends predominantly on the complexity of the rule of stimulus–response associations and to a lesser extent task’s difficulty.KeywordsAccuracyLearningProbabilistic feedbackResponse timeUnstructured category learning