Hybrid Connectionist Models: Temporary Bridges Over the Gap Between the Symbolic and the Subsymbolic
Connectionist networks, often known as neural networks or spreading-activation networks, have recently been the subject of a tremendous rebirth of interest, as researchers have begun to explore their advantages for cognitive models ranging from low-level sensory abilities to high-level reasoning. Connectionist models employ massively parallel networks of relatively simple processing elements that draw their inspiration from neurons and neurobiology, as opposed to traditional symbolic artificial intelligence (AI) models, which are generally based on serial Von Neumann architectures. This chapter will explore the benefits o f building models that combine three major levels of network architectures that can be considered connectionist in a very broad sense: distributed connectionist net works and localist connectionist networks, which fall under the strict definition o f connectionism assumed elsewhere in this book, and marker-passing networks, the most connectionist of the symbolic architectures. Each has different types of cognitive models for which they are best suited.