Semantic decomposition and marker passing in an artificial representation of meaning
The research area of Distributed Artificial Intelligence aims at building intelligent agent systems. Multi-Agent Systems have been applied successfully in many domains, from an intermodal planning domain to cascading security thread simulations. But still, agents struggle with the meaning of concepts used in language. Intelligence needs language to form thoughts. Thus, the challenge addressed in this thesis is to provide a computable representation of meaning and evaluate its usefulness. Based on the theory of a mental lexicon and the thesis that meaning is a combination of symbolic and connectionist parts, I investigate the use of the theory of Natural Semantic Metalanguage (NSM) to build an artificial representation of meaning. I show that the use of NSM for creating a semantic graph out of different information sources can be utilized as a basis for Marker Passing algorithms. The Marker Passing algorithm encodes symbolic meaning to guide the reasoning over the connectionist semantic graph. Through the combination of a semantic graph and symbolic Marker Passing, I can combine connectionist and symbolic approaches to AI research to create my artificial representation of meaning. To test my approach, I build a semantic distance measure, a word sense disambiguation algorithm and a sentence similarity measure which all go head to head with the state-of-the-art. I apply those approaches to two use cases: A semantic service match marking and a context-dependent heuristics. I evaluate my heuristic by utilizing them in AI problem-solving component which uses AI planning guided by my heuristic.