特征概率影响多维和少维类别的分类学习和特征学习

The Influence of Probabilities of Features on the Classification Learning and Feature Learning of High-Dimension Categories and Low-Dimension Ones

探讨了特征概率对多维和少维类别的分类学习和特征学习的效果及策略的影响。结果表明高特征概率条件下,多维比少维类别的分类学习更容易,而且学到更多的特征知识,多维条件下人们更倾向于整体性加工策略,而少维条件下人们倾向于分析性加工策略。低特征概率条件下,多维比少维类别的分类学习和特征学习都困难,且两种条件下人们都倾向于采取分析性加工策略。

Researchers often use artificial categories instead of natural ones in psychology experiments in order to avoid familiarity, but these two kinds of categories are different in many ways, especially in category dimensionality. Natural categories have a lot of dimensions, but the artificial ones often have fewer dimensions. Since category materials influence how people learn categories and their representations, it is important to investigate how category dimensionality affects classification learning and feature learning. Hoffman and Murphy （2006） investigated this issue and found people learned high-dimension category no more slowly than low-dimension ones and learned more feature knowledge, which was surprising and contrary to what current category theories predicted. We analyzed the materials they used in their experiments and found that the probabilities of the features were high and different features could be integrated to generate holistic perceptions, which might be the reason behind the results. We thought probabilities of features played an important role in classification learning and feature learning of high-dimension categories. Our hypothesis was that classification learning and feature learning of high-dimension categories would not be advanced if probabilities of features were low and dimensions were poorly integrated. We explored how category dimensionality and probabilities of features affect category learning. 80 college students participated in the experiment voluntarily and we analyzed the data using SPSS 11. 0. The resuits showed that if the probabilities of features were high, the classification learning and feature learning of high-dimension category were better than those of the low-dimension one. However, if the probabilities of features were low, the classification learning and feature learning of low-dimension category were better than those of the high-dimension one. From this study we can draw the conclusion that probabilities of features and relations between dimensions play an important role in category learning, which is often ignored in current category theories, and that people tend to grasp the whole mode and holistic perception in high-dimension classification learning and feature learning.