On the design of a concurrent object-oriented spreading activation architecture

The notion of spreading activation over a semantic network is one of the most frequently used approaches to several important problems in artificial intelligence. We first argue that spreading activation subsumes not only marker-passing systems, but also local connectionist and recent ´hybrid´ models. We then suggest that spreading activation reduces to the passing of messages between concurrent objects. We argue therefore in favor of the concurrent and object-oriented nature of a general architecture for spreading activation. Concurrency is required for massive parallelism; object-orientedness, for the mechanisms it readily provides for semantic modeling. We then overview an object-oriented framework for spreading activation, which we have implemented in both Smalltalk and ABCL/1. The fundamental characteristic of this implementation is that it is programmable, and thus can accommodate most of the existing spreading activation models.<>