The Limitations Of Human Category Learning From The Perspective Of The XOR Task

Category learning is an important field of cognitive psychology. Category learning of objects with multiple attributes or features is an important way to understand human classification ability. XOR (exclusive-or) is a kind of logic operation and the XOR style task formed by such operation mode can be used in category learning. To successfully do XOR classification, one must make judgment by the conjunction of more than one feature dimensions. The task therefore provides an excellent research paradigm for exploring the mechanisms of human categorization and the limits of their classification ability.The purpose of this study used the XOR classification task is to explore the limits of human classification ability and associated cognitive mechanisms during such classification.The study involved six experiments. Experiment 1 investigated meaningless image classification with (three) similar feature dimensions. The results showed that subjects can learn the task, and most of them used the strategy connected with the three feature dimensions for category judgments. In experiment 2, we investigated the classification of geometric shapes with (three) dissimilar feature dimensions. The results showed that subjects can learn the task, mostly by directly memorizing one category of pictures. Experiment 3 examined the classification of artificial nature category shapes with (three) similar feature dimensions. Subjects can learn the task, mostly by using reference pictures or connecting with the three feature dimensions. Experiment 4 was to explore the classification of artificial nature category shapes with (four) similar feature dimensions. The results showed that although human beings can learn this task, individual differences were large, and it was difficult to complete the task for some of them. All subjects who have learned the task adopted a reference picture strategy, and found the right rules of feature change. Experiment 5 explored the classification of artificial nature category shapes with (four) similar feature dimensions by the way of whole representation. The results showed that it was very difficult to classify pictures by rote memory and the use of the rules was critical for good performance. Experiment 6 investigated the classification of multi-feature natural shapes and showed that subjects can learn the task, suggesting that the difficult in Exp.5 was not due to that they had to remember a large number of random pictures. This set of six experiments lead to some preliminary conclusions. Firstly, the human can complete XOR category learning task with three feature dimensions, but they showed some limitations for four feature dimensions. Secondly, the category learning process is not explained by a single category learning theory. Thirdly, whether the stimulus can be regarded as a whole and be classified based on whole representation mainly depends on the perceptual attribute integration of feature dimensions of experimental materials. Fourthly, the main factor that influences the learning effect of multi-feature XOR category learning is the number of feature dimensions but not the type of experimental materials. Fifthly, the use of rules is central to successful XOR category learning.