Computational cognition and the age of supercomputing: using high-performance computing and usenet to model memory, meaning and the mind

摘要

The Hyperspace Analogue to Language (HAL) model of memory learns what words mean using a simple learning algorithm on 320 million words of Usenet text. The meanings of words are represented in a 140,000 dimensional hyper-space. a variety of metrics are developed in the model have broad explanatory power that captures a range of cognitive phenomena in normal, aging, and disordered individuals. These meansing representations cut across traditional semantic, grammatical, syntactic boundaries making the model useful for a variety of applications such as information retrieval, ambiguity resolution, automatic categorization, and content filtering. The advantage of these models is that they use learning procedures that scale up to real world language problems and make explicit the processes by which systems learn.