The limits of co-occurrence: Tools and theories in language research

The development of new tools such as those described in the articles of this special issue marks an important advance in discourse research. These tools include ways to test competing models and ways to track very large language databases. The former aid the field's ability to demonstrate theoretical progress, and the latter can perform computations beyond what a human expert could do in a lifetime. There is a difference between tools and theory, however, and in this article, I point out some of the ways in which large dimensional space methods such as Latent Semantic Analysis and Hyperspace Analog to Language fall short of being plausible theories about psychological reality. I examine in-principle failures and wrong-kind failures that arise in the systems and point out the limitations of systems based exclusively on co-occurrence.