Barsalou Lawrence W
Department of Psychology, Emory University, Atlanta, GA 30322, USA.
Philos Trans R Soc Lond B Biol Sci. 2003 Jul 29;358(1435):1177-87. doi: 10.1098/rstb.2003.1319.
After reviewing six senses of abstraction, this article focuses on abstractions that take the form of summary representations. Three central properties of these abstractions are established: ( i ) type-token interpretation; (ii) structured representation; and (iii) dynamic realization. Traditional theories of representation handle interpretation and structure well but are not sufficiently dynamical. Conversely, connectionist theories are exquisitely dynamic but have problems with structure. Perceptual symbol systems offer an approach that implements all three properties naturally. Within this framework, a loose collection of property and relation simulators develops to represent abstractions. Type-token interpretation results from binding a property simulator to a region of a perceived or simulated category member. Structured representation results from binding a configuration of property and relation simulators to multiple regions in an integrated manner. Dynamic realization results from applying different subsets of property and relation simulators to category members on different occasions. From this standpoint, there are no permanent or complete abstractions of a category in memory. Instead, abstraction is the skill to construct temporary online interpretations of a category's members. Although an infinite number of abstractions are possible, attractors develop for habitual approaches to interpretation. This approach provides new ways of thinking about abstraction phenomena in categorization, inference, background knowledge and learning.
在回顾了六种抽象意义之后，本文聚焦于以概要表征形式出现的抽象概念。确立了这些抽象概念的三个核心属性：（i）类型-实例解释；（ii）结构化表征；以及（iii）动态实现。传统的表征理论很好地处理了解释和结构问题，但缺乏足够的动态性。相反，联结主义理论极具动态性，但在结构方面存在问题。感知符号系统提供了一种能自然实现所有这三个属性的方法。在这个框架内，会发展出一组松散的属性和关系模拟器集合来表征抽象概念。类型-实例解释源于将一个属性模拟器绑定到一个被感知或模拟的类别成员的某个区域。结构化表征源于将属性和关系模拟器的一种配置以一种综合的方式绑定到多个区域。动态实现源于在不同场合将属性和关系模拟器的不同子集应用于类别成员。从这个角度来看，记忆中不存在一个类别的永久或完整的抽象概念。相反，抽象是一种构建对类别成员的临时在线解释的技能。虽然可能存在无数种抽象概念，但会为习惯性的解释方式形成吸引子。这种方法为思考分类、推理、背景知识和学习中的抽象现象提供了新的思路。
