A comparative study on the formation of prototype categories

我们通过模仿自然类别的属性来创建人工类别，使用彩色矩形作为组成类别的特征。一是家族相似性关系；高度可变的样本由一个相似网络构成，每个类别的特征相互关联。另一种是多态规则；没有任何一个特征是区分这些类别所必需的。人类在学习分类方面表现出困难，然后根据他们获得的规则以全有或全无的方式对新刺激做出反应。鸽子很容易学会这些类别，并表现出与家族相似性程度一致的原型效应。因为定义为类别原型的刺激包含了在给定类别中最频繁出现的特征，所以在原型上的最佳辨别很容易被特征学习和范例学习解释。在接下来的实验中，我们训练鸽子辨别人脸的变形图像。每个类别中的训练范例是通过将单个独特的面孔M与每个原始面孔（a， B， C和D）及其50%的变体混合而创建的。为了检验范例学习的效果，我们对鸽子进行了额外的中间形态测试，这些中间形态在面孔形态维度上介于M和每个原始面孔之间。鸽子的概化梯度随正面类别M的形态比例增大而减小，与类别特征的漫画效应相似。范例学习和特征学习模型都无法预测结果。本研究中使用的分类任务可能使鸽子能够抽象出类别成员共有的面部特征，并根据共同成分的程度对新面孔做出反应。

We created artificial categories by mimicking properties of natural category, using colored rectangles as the features comprising the categories. One is family resemblance relationship ; the highly variable exemplars were structured by a similarity network with the features correlating to one another in each category. Another is polymorphous rule ; no single feature was essential for distinguishing the categories. Humans showed difficulty in learning the categories and then responded to novel stimuli in an all-or-none fashion based on the rules they acquired. Pigeons learned the categories with ease and then showed a prototype effect in accord with the degrees of family resemblance. Because the stimuli defined as the prototypes of the categories contained the features that most frequently occurred in the given category, the best discrimination at the prototypes was readily explained by feature learning and exemplar learning.In the next experiment, we trained pigeons to discriminate morphed images of human faces. The training exemplars in each category were created by blending a single unique face M with each of the original faces (A, B, C, and D) and their 50% morphs. To examine the effects of exemplar learning, we tested the pigeons with additional intermediate morphs that ranged between M and each of the original faces on a face-morph dimension. The pigeons' generalization gradient increased as a function of morph proportion of M of the positive category and decreased as a function of morph proportion of M of the negative category, a phenomena similar to the caricature effect of category features. Both the exemplar learning and feature learning models failed to predict the finding. The classification task used in this study might have enabled the pigeons to abstract the facial features shared by the category members and to respond to novel faces based on the degrees of common component.