The Influence Of Prototype's Similarity And Feature's Distribution On Category Dimensionality Effect

The study of Hoffman 2006 finds that learning in high-dimensional categories is no slower than in low dimensional ones; instead, subjects can learn more features from high-dimensional categories than from low-dimensional ones. The results show the occurrence of "category dimensionality effect"-categories with more dimensions enable individuals to learn more.Surprisingly, there is evidence in the experimental literature that it might be very difficult to learn categories with many features or dimensions. For example, Shepard 1961 found that in categories with more dimensions,learning became more difficult. Hoffman's study was different from the. original category theory and models. In this research we will investigate what will influence the occurrence of the category dimensionality effect.Previous studies regarded the category dimension as an important factor that affects category learning, and category feature is the form of category dimension, so we suppose that category feature would influence the results of category learning. In this research, the category feature will change in three conditions. First, we will vary the similarity of the stimulus features to category prototype; second,we will vary the distribute of the feature; third, we will change the nature of the features.Based on Hofffman's study, the research includes three experiments, and each experiment has two sections. In the first section, we will test the category dimensionality effect. In the second section, we will explore the influence of changes of the feature on category dimensionality effect. In experiment 1, we change the similarity of the stimulus features to category prototype, and test if the results will affect category dimensionality effect. In experiment 2, we change the distribution of the feature, to find out if the results will affect category dimensionality effect. In experiment 3, we will describe the category dimensionality effect in natural category. Dependent variables in all three experiments are learning accuracy, single-feature accuracy and the number of the learned dimension. The experiments get the following results:(1)In the condition that stimulus features are 4/5 similar with category prototype, the experimental results show the existence of the dimensionality effect. In the condition of stimulus features with 3/5 similarity to category prototype, the experimental results show no presence of the dimensionality effect. Therefore, the similarity of stimulus features to prototype is a factor to affect dimensionality effect.(2) In the condition that exceptional features were unevenly distributed between the stimulus, the dimensionality effect of category learning is also found. This shows that changes of the exceptional feature distribution will not affect the dimensionality effect.(3)In the natural category learning, the existence of the dimensionality effect also exists. The result from natural category learning is better than that from experiment category learning. It indicates that dimensionality effect complies with the laws of natural category learning.In short, all three experiments in this study have proved that dimensionality effect does not widespread, but only occur when the similarity of the stimulus features to category prototype is high. Dimensionality effect is in line with people's natural categories of learning, so the research on it is of practical significance.