Neural dynamics of speech and language coding: developmental programs, perceptual grouping, and competition for short-term memory.

A computational theory of how an observer parses a speech stream into context-sensitive language representations is described. It is shown how temporal lists of events can be chunked into unitized representations, how perceptual groupings of past item sublists can be reorganized due to information carried by newly occurring items, and how item information and temporal order information are bound together into context-sensitive codes. These language units are emergent properties due to intercellular interactions among large numbers of nerve cells. The controlling neural networks can arise through simple rules of neuronal development: random growth of connections along spatial gradients, activity-dependent self-similar cell growth, and competition for conserved synaptic sites. Within these networks, a spatial frequency analysis of temporally evolving activity patterns leads to competitive masking of inappropriate list encodings in short term memory. The neurons obey membrane equations undergoing shunting recurrent on-center off-surround interactions. Several design principles are embodied by the networks, such as the sequence masking principle, the long-term memory invariance principle, and the principle of self-similar growth.