Behavior-oriented Acquisition of Semantic Concepts

本研究探讨了人类思维中的机制，这些机制参与了从通过学习对象或事件的例子获得的不相关的死记硬背知识到灵活的概念系统的转变，通过这种概念系统，我们可以将没有经历过的东西想象为例子的重组。虽然许多研究者认为，组成的象征性概念应该是基于感觉运动的，但如何实现这一点仍然不清楚。本研究从建构主义的角度重新审视“结构对齐”假说，弥补了这一差距。提出的子符号学习模型发出的“结构对齐”的行为概念的基础上给定的感觉运动相似性的机器人的行为模式，并可以识别和重新生成以前看不见的行为模式的重组已知的模式。

This study examines the mechanisms in the human mind that are involved in the shift from unrelated rote knowledge acquired by learning examples of objects or events into a flexible conceptual system by which we can conceive something not experienced as a recombination of the examples. Although many researchers have suggested that compositional symbolic concepts should be sensorimotor grounded, how this may be accomplished remains unclear. This study bridges the gap by reconsidering “structural alignment" hypothesis from a constructivist view. Proposed sub-symbolic learning model emits “structural alignment" of behavioral concepts based on given sensorimotor similarity of robot's behavioral patterns and can recognize and re-generate previously unseen behavioral patterns in terms of the recombination of known patterns.