ABSTRACT
Connectionist networks have earned recognition in many domains that can be characterized as hard or impossible to explicitly formalize, e.g., driving cars emotion recognition and pronunciation. The computational level of the classical cognitive architecture describes what a system computes, and why. At this level, the two systems of interest compute the same function, i.e., their external behavior is indistinguishable. A physical symbol system (PSS) uses a representing world containing ‘symbols’, e.g., a and b. These symbols can be realized in a lower-level physical implementation, such as metric vectors. Since the representing world is governed by an equivalent grammar to the represented world, it has the capability to represent the external tokens in isolation or combined into a structure. In fact, the definition of a PSS hinges on the ability to generate representations for the complex expressions of the represented world in this concatenative way.
