关键词: 类别学习, 学习条件, 样例相似性, 元认知监控

Abstract: Recently, category learning has become one of the targets on metacognitive monitoring. Most previous studies focus on the potential clues of metacognitive monitoring only under no-rule learning condition, primarily using bird family pictures as materials. However, due to the large number of categories, it is difficult to manipulate similarity. Therefore, the influence of exemplar similarity on metacognitive monitoring cannot be explored. Further, less is known about the role of learning conditions in the metacognitive monitoring of category learning, especially rule learning condition. Accordingly, in order to reveal the effect of learning condition and exemplar similarity on metacognitive monitoring, the study aims to investigate potential clues of metacognitive monitoring and the interrelationships between clues by differences between classification accuracy and metacognitive judgement under diverse learning conditions and exemplar similarities. Materials are 32 animals, including 8 original and 24 deformed items which vary in five two-valued features. Before the experiment, a similarity assessment task was performed to determine three exemplar similarity level of low, middle and high in deformed animals. The experiment is a mixed group design with learning condition (rule vs. no-rule) and exemplar similarity (low vs. middle vs. high) as between-subject variables and with match type (good vs. bad) as a within-subject variable. A total of 278 participants (male=73; average age=20.23, SD=1.79) were randomly assigned in 6 groups. Each participant learned to classify original animals into one of two categories on learning phase and then take a test, containing the items of original animals (old items) and deformed animals with a certain exemplar similarity (new items). In particular, the participants from rule learning condition were informed about the classification rule at the beginning. However, the participants from no-rule learning condition learned classification through feedback. When classification accuracy in two consecutive blocks was reached 100% or the requirement was failed to achieve within 20 blocks, the learning phase would be followed by the prediction phase, where the participants need to make category learning judgements (CLJs) on two category labels. The transfer phase immediately followed the end of the prediction phase, presenting totally 16 trails including old items and news items in random order without feedback. And after each classification, the participants needed to make confident judgement on their answers. The mixed design ANOVA was performed on classification accuracy of new items, revealing a significant three-way interaction, which indicated the participants were more accurate in classifying good transfer items than bad transfer items in the high exemplar similarity group of rule condition whereas the participants relied on similarity of old items to make classificatory decision in no-rule condition. Moreover, compared with no-rule condition, classification accuracy of transfer items and CLJs were observably higher in rule condition, suggesting category learning judgement is more sensitive to the rule condition. As for confident judgement of new items, the result demonstrated the confident judgement increased with the improvement of exemplar similarity degree in no-rule condition. In summary, rule and exemplar similarity are the clues of metacognitive monitoring in category learning, and following the correct rules to classify is the optimal way of metacognitive judgment. More importantly, rule-based and similarity-based classification processes can appear in the same classification task for which high exemplar similarity would block the application of rule.

Key words: category learning, learning condition, exemplar similarity, metacognitive monitoring