Future Directions

Abstract

The experiments offered here represent only a first step toward a theory of comprehension. Further experimentation is needed to determine the validity of the hypotheses put forth. A necessary second step is to test the model with a significantly larger knowledge base and demonstrate that it can achieve some form of mastery, perhaps by having the program read an entire issue of the Wall Street Journal and then answer questions about the articles. As the current simulation is inadequate to deal with these computational demands, this would require a parallel machine. A knowledge base containing at least 1000 concepts would probably provide an adequate initial conceptual vocabulary from which to test the theory. This represents an approximate 10-fold increase of the current conceptual vocabulary size. Experiments with LeMICON indicate a vocabulary of this size would probably generate between 150,000 and 200,000 relations.