The incorporation of human memory models into artificial intelligence-based problem solving
University of New Brunswick
Artificial intelligence is in an ideal position to use research in other fields to improve itself. In this thesis, ideas from psychological and cognitive models of learning, as well as human memory models, have been hybridized to form a problem solving system. This was done by applying aspects of human learning and memory to a problem in natural language processing, the word sense disambiguation problem. The question being answered is "can human-like memory improve accuracy in a particular area of artificial intelligence, specifically word sense disambiguation?". This involved exploring the various models and presenting how portions of the models could fit together into a computer system. The model was implemented and the details for how it came together, what worked and what did not are presented. Finally, the system was tested, using an open source competition data set, and compared with other systems using the data set. The results ranked the system among the top five contestants. The results also show a potential for future accuracy improvements in determining word sense, as well as a chance to better model human memory and learning.