Experiments in Acquiring Knowledge from On-line Corpora

Abstract

This chapter presents an approach to acquiring knowledge from an on-line corpus of text automatically, based on the use of mutual information statistics. More specifically, it explores the potential for automatically constructing a two-tier model of semantic memory from on-line textual corpora as follows: Automatically construct the interrelationships between concepts in semantic memory (i.e., construct the relational tier of semantic memory). Automatically encode the background frame knowledge associated with the concepts in semantic memory (i.e., encode each concept’s associational knowledge). Dynamically change semantic memory as new texts are processed (i.e., evolve semantic memory in response to new input).