On the computational basis of learning and cognition: Arguments from LSA

The principal goal of this chapter has been to suggest that high-dimensional vector space computations based on empirical associations among very large numbers of components could be a close model of a fundamental computational basis of most learning in both verbal and perceptual domains.

Abstract

This chapter discusses the computational basis of learning and cognition. To deal with a continuously changing environment, living things have three choices: (1) evolve unvarying processes that usually succeed, (2) evolve genetically fixed effector, perceptual, and computational functions that are contingent on the environment, and (3) learn adaptive functions during their lifetimes. The theme of this chapter is the relation between (2) and (3): the nature of evolutionarily determined computational processes that support learning. The principal goal of this chapter has been to suggest that high-dimensional vector space computations based on empirical associations among very large numbers of components could be a close model of a fundamental computational basis of most learning in both verbal and perceptual domains. More powerful representational effects can be brought about by linear inductive combinations of the elements of very large vocabularies than has often been realized. Success of one such model to demonstrate many natural properties of language commonly assumed to be essentially more complex, nonlinear, and/or unlearned, along with evidence and argument that similar computations may serve similar roles in object recognition, are taken to reaffirm the possibility that a single underlying associational mechanism lies behind many more special and complex appearing cognitive phenomena.