Getting Fundamental Aspects of Language Understanding to Work in a Neuromorphic Framework

A Tutorial Presented By Christian Huyck

Natural language processing is incredibly complex, and even the best artificial systems are vastly inferior to a typical three year old. The best artificial systems are symbolic, but neural systems can be developed to process relatively simple language. Regular languages are a simple type of formal language that can carry one a long way in processing. Regular language processors can be readily implemented with simulated neurons using Cell Assemblies and Finite State Automata. Similarly words can be represented in a range of ways, and the typical semantic output of an NL system is a verb frame; both words and frames can be implemented in neurons. Combining words, regular grammar, and case frames yields a simple but relatively effective language system. More complex language processing (e.g. context free languages) can be processed when variable binding is included. I'll also briefly discuss my neural psycholinguistic parsing model, that is a step toward human level language processing.