Temporal Synchrony, Dynamic Bindings, and Shruti: a Representational but Non-classical Model of Reeexive Reasoning

1 Synchrony versus signatures As Lange & Dyer point out, there are a number of important structural similarities in shruti and robin. The two models, however, diier signiicantly in the way they solve the dynamic binding problem. In shruti, dynamic bindings are represented by the synchronous ring of nodes. Thus the binding (John=giver) is represented by the synchronous ring of node clusters representing`John' and`giver'. robin on the other hand associates a predetermined and unique pattern of activation called signature with each concept and represents dynamic bindings by propagating appropriate signature patterns over appropriate clusters of role nodes. Thus the binding (John=giver) is represented in robin by establishing the pattern of activation corresponding to the signature of`John' over the cluster of nodes representing the rolègiver'. Lange & Dyer argue that the signature approach is superior to the synchronous activation approach. Their argument can be summarized as follows: (i) Whereas the signature approach allows arbitrarily many entities to be co-active, the synchronous activation approach allows at most 10 distinct entities to participate in dynamic bindings. This severely limits the representational and processing power of shruti. (ii) Whereas \signatures" have semantic content, \synchrony" does not. This lack of semantic content would make learning more diicult in the synchronous activation approach. The arguments put forth by Lange & Dyer, however, are based on a serious misunderstanding of shruti's synchronous activation approach. Lange & Dyer make two crucial errors in understanding the synchronous activation approach. First, they misunderstand what is meant by an \entity". Second, they confuse the medium of representing and propagating information, namely, synchrony, with the content that is represented using this medium. The rst error leads Lange & Dyer to the erroneous conclusion that shruti will rapidly run out of phases during language processing. The second error convinces them that shruti's representation of bindings lacks semantic content, and therefore, it will be diicult for shruti to learn high-level knowledge.