Research Initiation: Induction of Casual Theories

研究人员将通过观察事件及其后果的例子来调查一般因果关系理论的习得。因果关系理论是关于使人相信一个行为具有必然后果的原则的信息。例如，人们倾向于选择一个在空间和时间上与结果接近的原因。这项工作建立在一个计算机模型的先前实现的基础上，该计算机模型使用因果关系理论作为明确的偏见，以促进因果关系理论的获得(即，推理规则编码了关于行为效果的知识，例如“玻璃物体在受到足够的力打击时会破碎”)。新的工作是研究从例子中获得这些一般原理的计算模型。学习程序的输入将由代表世界状态的命题的时间序列组成。输入表示将通过模拟(通过改变参数，例如数据中的噪声量)和通过手动编码儿童科学入门课文中的因果变化的示例来构建(以确保学习程序将能够解释各种常见的因果关系)。

The researcher will investigate the acquisition of general theory of causality by observing examples of events and their consequences. A theory of causality is information about the principles that lead one to believe that an action has a necessary consequence. For example, people tend to select a cause which is close in space and time to an effect. The work builds on prior implementation of a computer model which uses a theory of causality as an explicit bias to facilitate the acquisition of a theory of causation (i.e., inference rules encode knowledge of the effect of actions, such as "glass objects break when struck with sufficient force."). The new work is an investigation of computational models for acquiring these general principles from examples. Input to the learning program will consists of a temporal sequence of propositions representing the states of the world. The input representation will be constructed by simulation (to allow systematic experimentation by varying parameters such as the amount of noise in the data) and by hand coding examples of causal changes from children's introductory science texts (to insure that the learning program will be able to account for a wide variety of common causal relationships).