Correlation And Number Of Additional Dimensions Imapct On Direct And Incidental Unsupervised Category Learning

Each object in the real world has a large number of dimensions,and the number of experimental materials used in the early category experiments research is small,which makes the results difficult to be inferred to the real situation.After that,researchers increased the number of dimensions to examine multi-dimensional category learning.However,most of these studies are carried out under the family similarity structure,which can not prove whether the results are applicable to other category structures.Adding dimensions is also called additional dimensions.Additional dimensions will not only change the number of dimensions,but also lead to different category structures due to different correlations of additional dimensions.Therefore,manipulating the correlation and number of additional dimensions can investigate multi-dimensional category learning under different category structures.Previous studies on multi-dimensional category learning all focus on the impact on learning accuracy,but have not systematically studied the category representation.The focus of category learning research is to explore the category representation of learners,because the interpretation of representation can deeply understand the cognitive process and strategies in category learning.In addition to learning materials,real learning situations also include a variety of learning methods.Most of the previous studies on category learning were carried out under the supervised mode,which did not conform to the typical real learning environment.After all,the real world cannot always give feedback to the learner’s problems.Direct and incidental unsupervised learning is more widely existing in the real learning situation.Previous studies have shown that direct and incidental unsupervised category learning may have different cognitive processes,which are manifested by different classification strategies and category representations.So far,there are few studies on unsupervised category learning,and its theoretical explanation is still weak.The investigation on its internal mechanism and influencing factors are not deeply investigated,and further research is needed.On this basis,we designed two experiments to manipulate the correlation and number of additional dimensions to investigate the learning effect and category representation of different structured multi-dimensional categories in a more realistic learning mode(direct and incidental unsupervised learning).Experiment 1 explored the influence of the correlation and number of additional dimensions on direct unsupervised category learning.Experiment 2 explored the influence of the correlation and number of additional dimensions on incidental unsupervised category learning.Then Experiment 1 and Experiment 2 were compared and analyzed the differences in learning effect and category representation between direct and incidental unsupervised category learning.This study not only complements multi-dimensional category learning research with different structure in a more natural way,but also explores the category representation and influencing factors of unsupervised category learning.This is conducive to a deeper understanding of the cognitive process of human category learning in real situations,enriching the theoretical model of category learning,which has strong theoretical and practical significance.The results show that:(1)The learning effect of the category with additional dimension related to category attribution is better than that of the category with additional dimension unrelated to category attribution.Learners can form similarity-based representations in categories with additional dimensions related to category attribution,but not in categories with additional dimensions unrelated to category attribution.(2)In a category structure with a deterministic dimension,the number of additional dimensions impairs unsupervised category learning.(3)With the increase of the number of additional dimensions,fewer learners tend to rule-based representation,and some learners on direct unsupervised category learning turn from rule-based representation to similarity-based representation,while the number of similarity-based representation learners on incidental unsupervised category learning has no significant change.It indicates that the number of additional dimensions damages the learning of rule-based representation and does not affect the learning of similarity-based representation.(4)The learning effect of direct unsupervised category learning is better than that of incidental unsupervised category learning.Direct unsupervised category learning learners tend to rule-based representation;incidental unsupervised category learning learners tend to both rule-based representation and similarity-based representation,without obvious classification tendency.