The Effects of Negative Premises on Inductive Reasoning: A Psychological Experiment and Computational Modeling Study-猫眼课题宝

否定前提对归纳推理的影响：心理实验和计算模型研究

DOI：

发表时间：

2006

期刊：

影响因子：

通讯作者：

中科院分区：

文献类型：

作者：

M. Nakagawa;Kayo Sakamoto

否定前提对归纳推理的影响：坂本佳代的心理实验与计算模型研究东京工业大学，东京都目黑区大冈山2-21-1，邮编152-8552（NAKAGAWA@Nm. J. Titech.Ac.Jp）东京工业大学，2-21-1 O-okayama，Meguro-ku，Tokyo，152-8552 JAPAN研究，如因果学习（例如，Buehner和Cheng，2005年）。然而，在与论证评估有关的归纳推理研究中，否定前提的影响很少被详细讨论。研究否定前提的效果无疑有助于我们理解归纳推理。例如，否定前提和肯定前提中的实体属于同一类别的情况对于基于类别的归纳理论来说显然是有问题的（Osherson等人，1990年），因为不可能从范畴观点区分否定前提和肯定前提。与基于类别归纳理论的相似性和覆盖模型不同，Sloman（1993）提出了一种基于简单感知器的基于特征的模型。根据Sloman的观点，范畴结构的知识对于论证的评价是不需要的。相反，他假设论证评估是基于对前提实体和结论实体之间的特征相似性的简单计算。因此，Sloman的基于特征的模型可能比基于类别的归纳理论更有效地应对负面前提。然而，Sloman的模型的心理有效性还有待于测试方面的处理负面的前提。因此，本研究检验了基于特征的归纳理论在处理这些基于类别的归纳理论存在问题的情况下的有效性。在模型构建方面，什么样的模型才能充分代表包括否定前提在内的基于特征的归纳的认知过程？除了Sloman（1993）的模型外，Sakamoto，Terai and Nakagawa（2005）也提出了一个基于特征的模型。虽然结构相似，但他们的模型扩展了学习算法，以科普负面前提。此外，他们的模型利用语料库分析结果来计算特征相似性，而不是Sloman模型中使用的心理评估结果。这意味着Sakamoto等人的模型比Sloman的模型能够模拟更多种类的实体（超过20，000个，而只有46个），因为语料库分析提供了大量单词的信息。然而，当难以归纳出合适的类别时，特征相似性比较所涉及的计算水平将远远超过简单感知器的计算能力。因此，这项研究提出了一个修改版本的坂本等人摘要各种学习理论强调的重要性，消极学习（例如，Bruner，1959; Hanson，1956）。然而，在归纳推理的理论中，很少详细讨论否定前提的影响（除了Osherson等人，1990年）。虽然Sakamoto et al.（2005）提出了一些可以科普否定前提的计算模型，并验证了它们的心理有效性，但他们没有考虑基于类别的归纳理论无效的情况，例如否定前提和肯定前提中的实体属于同一类别。本研究的目的是检验一个假设，即即使否定前提和肯定前提涉及相同的范畴实体，人们也可以通过比较特征相似性来估计论证结论的相似性。基于这一假设，提出了两个计算模型来模拟这种认知机制。虽然这两种模型都能够模拟心理实验的结果，但感知器模型却不能。最后，我们认为，这两个模型的数学等价性（从支持向量机的角度来看）表明，它们代表了一种很有前途的方法来建模负面前提的影响，从而充分处理神经网络上基于特征的归纳的复杂性。介绍本研究关注的是评估“参数”，如：牧羊犬产生吞噬细胞。马产生吞噬细胞。牧羊犬产生吞噬细胞。线以上的命题被称为“前提”，而下面的陈述是“结论”。论证的评估包括根据前提估计结论的可能性。Osherson，Smith，Wilkie，洛佩斯和Shafir（1990）把这种论证称为“范畴”论证，因为谓词（e.例如，在一个实施例中，“产生吞噬细胞”），并且结论归因于一个或多个实体（例如，例如，在一个实施例中，“牧羊犬”、“牧羊犬”）。前提也可以是否定的形式（e）。例如，在一个实施例中，“企鹅不产生吞噬细胞”）。由于经典的研究，如歧视学习（e。例如，在一个实施例中，Hanson，1956）和概念学习（例如，Bruner，1959），反例的重要性已被广泛认识，并已在最近的

The Effects of Negative Premises on Inductive Reasoning: A Psychological Experiment and Computational Modeling Study Kayo Sakamoto (SAKAMOTO@Nm.Hum.Titech.Ac.Jp) Tokyo Institute of Technology, 2-21-1 O-okayama, Meguro-ku, Tokyo, 152-8552 JAPAN Masanori Nakagawa (NAKAGAWA@Nm.Hum.Titech.Ac.Jp) Tokyo Institute of Technology, 2-21-1 O-okayama, Meguro-ku, Tokyo, 152-8552 JAPAN studies, such as causal learning (e.g., Buehner and Cheng, 2005). However, the effects of negative premises have rarely been discussed in any detail in the context of inductive reasoning studies concerned with the evaluation of arguments. Investigating the effects of negative premises can undoubtedly contribute to our understanding of inductive reasoning. For instance, cases where the entities in both negative and positive premises belong to the same category are clearly problematic for the category-based induction theory (Osherson et al., 1990) because it is impossible to distinguish between negative premises and positive premises from the categorical viewpoint. In contrast to the similarity and coverage model based on category-based induction theory, Sloman (1993) has proposed a feature- based model based on a simple perceptron. According to Sloman, knowledge of category structure is not required for the evaluation of arguments. Rather, he assumes that argument evaluation is based on a simple computation of feature similarities between the entities of the premises and the conclusion. Thus, Sloman’s feature-based model may be more effective at coping with negative premises than category-based induction theory. However, the psychological validity of Sloman’s model has yet to be tested in terms of processing negative premises. Accordingly, this study examines the validity of feature- based induction theory to handle these cases that are so problematic for category-based induction theory. In terms of model construction, what kind of model can adequately represent the cognitive process of feature-based induction, including negative premises? In addition to Sloman’s (1993) model, Sakamoto, Terai and Nakagawa (2005) have also proposed a feature-based model. While structurally similar, their model extends the learning algorithm in order to cope with negative premises. Moreover, their model utilizes corpus-analysis results to compute feature similarities, rather than the results of psychological evaluations used in Sloman’s model. This means that Sakamoto et al’s model is capable of simulating a far greater variety of entities than Sloman’s model (over 20,000 compared to just 46), because the corpus analysis provides information for an enormous quantity of words. However, when induction of the appropriate category is difficult, then the level of computation involved in the feature similarity comparisons will far exceed the computational capacity of simple perceptrons. This study therefore proposes a modified version of the Sakamoto et al Abstract Various learning theories stress the importance of negative learning (e.g., Bruner, 1959; Hanson, 1956). However, the effects of negative premises have rarely been discussed in any detail within theories of inductive reasoning (with the exception of Osherson et al., 1990). Although Sakamoto et al. (2005) have proposed some computational models that can cope with negative premises and verified their psychological validity, they did not consider cases where category-based induction theory is ineffective, such as when the entities in both negative and positive premises belong to the same category. The present study was conducted to test the hypothesis that, even when negative and positive premises involve same-category entities, people can estimate the likeliness of an argument conclusion by comparing feature similarities. Based on this hypothesis, two computational models are proposed to simulate this cognitive mechanism. While both these models were able to simulate the results obtained from the psychological experiment, a perceptron model could not. Finally, we argue that the mathematical equivalence (from Support Vector Machines perspective) of these two models suggests that they represent a promising approach to modeling the effects of negative premises, and, thus, to fully handling the complexities of feature-based induction on neural networks. Introduction This study is concerned with evaluating “arguments”, such as: Collies produce phagocytes. Horses produce phagocytes. Shepherds produce phagocytes. The propositions above the line are referred to as “premises” while the statement below is the “conclusion”. The evaluation of an argument involves estimating the likelihood of the conclusion based on the premises. Osherson, Smith, Wilkie, Lopez, and Shafir (1990) refer to this kind of argument as a “categorical” argument, because the predicate (e. g., “produce phagocytes”) in the premises and conclusion is attributed to one or more entities (e. g., “Collies”, “Shepherds”). The premises can also be negative in form (e. g., “Penguins do not produce phagocytes ”). Since classic studies, such as discrimination learning (e. g., Hanson, 1956) and concept learning (e.g., Bruner,1959), the importance of negative examples has been widely recognized, and has been demonstrated in more recent

The Effects of Negative Premises on Inductive Reasoning: A Psychological Experiment and Computational Modeling Study

否定前提对归纳推理的影响：心理实验和计算模型研究

文献摘要

被引文献

相似文献