EFFECTS OF CAUSAL BACKGROUND KNOWLEDGE ON CATEGORIZATION

描述(申请人摘要)：19个实验研究 因果背景知识在确定特征中心性中的作用 人们的概念表征。主要假设是基于一个 最新的基于理论的观点，认为概念，如理论，具有 相互因果联系的特征。安提出了因果关系 一种状态假说，认为特征是其他的原因 功能应该比那些起到效果作用的功能更重要。这个 提案描述了三组旨在测试和改进这一点的研究 自动确定特征权重的因果状态模型 关于他们的因果状态。首先，将因果状态模型应用于 解释了许多现有的现象，证明了 背景知识。这些包括作为以下函数的基本电平移位 专业知识，自然种类和人工制品之间的差异，开发 儿童对待自然种类和人工制品的方式的趋势，类别 分类的可变性，以及基于类别的属性的类型 归纳法。第二，因果状态假设的计算模型是 通过改变预测会影响的因素来实施和测试 特征权重。这些因素包括因果关系之间的因果关系 要素、由目标要素导致的要素数量以及 从目标特征分支出来的因果链接。因此，该模型将 为复杂知识库中的特征权重预测提供了依据 它们具有强度不同的多个相互交织的因果联系。第三, 该模型被测试以解释临床医生的诊断过程 不仅要调查模型在样本复杂知识中的一般性 基础，以及对类别和功能知识的广泛使用程度 概率可能与因果状态偏差相互作用。建议数 实验依赖于两种方法：(1)使用人工类别的任务 直接操纵新特征的因果状态并收集参与者 因果和非因果特征的特征中心性评级和(2) 使用熟悉类别的任务测量参与者的现有知识 随后将与其关联的特征的因果状态 中心性评级。该模型的主要理论贡献是 严格定义基于理论的分类，可以应用于 真实的案例。此外，对概念核心的理解也体现在 人们的因果解释将阐明结构和习得 知识的总称。

DESCRIPTION (Applicant's Abstract): Nineteen experiments investigate the role of causal background knowledge in determining feature centrality in people's conceptual representations. The main hypothesis is based on a recent theory-based view which suggests that concepts, like theories, have features that are causally connected to each other. Ahn proposes the causal status hypothesis which states that features serving as causes for other features should be more essential than those serving as effects. The proposal describes three sets of studies designed to test and improve this causal status model which automatically determines weights of features based on their causal status. First, the causal status model is applied to account for numerous existing phenomena demonstrating the effect of background knowledge. These include the basic level shift as a function of expertise, differences between natural kinds and artifacts, developmental trends in the ways children treat natural kinds and artifacts, category variability on categorization, and the types of properties in category-based induction. Second, a computational model of the causal status hypothesis is implemented and tested by varying the factors which are predicted to affect feature weighing. These include causal strengths between causally related features, the number of features caused by a target feature, and the number of causal links branching out from a target feature. Thus, the model will provide a basis for predicting feature weighing in complex knowledge bases which have multiple interwoven causal links varying in strengths. Third, the model is tested to explain clinicians' diagnosis processes to investigate not only the model's generality in a sample complex knowledge base but also how extensive use of categories and knowledge on feature probabilities might interact with the causal status bias. The proposed experiments rely on two methods;(1) Tasks using artificial categories directly manipulate causal status of novel features and collect participants ratings on feature centrality for causal and non-causal features and (2) tasks using familiar categories measure participants' existing knowledge on causal status of features which will be subsequently correlated with their centrality ratings. The major theoretical contribution of the model is to rigorously define theory-based categorization which can be applied to real-life cases. In addition, an understanding of conceptual cores in terms of people's causal explanations will elucidate the structure and acquisition of knowledge in general.