All Recognition is Accomplished By Interacting Bottom-Up Sensory and Top-Down Context Bias in Occipital to Frontal Cortex Neural Networks

TL;DR

This paper explores how bottom-up sensory and top-down cognitive processes interact in neural networks to facilitate word recognition, demonstrating that priming effects are driven by anticipatory bias rather than threshold lowering, supported by experiments and a neural network model.

Contribution

It introduces a new recurrent neural network model that explains facilitation in word recognition as a bias effect driven by top-down and bottom-up interactions.

Findings

Priming facilitates recognition by bias, not threshold lowering.

Frontal and temporal cortex interactions are key to priming bias.

The model shows how anticipation increases recognition sensitivity within a time window.

Abstract

Recognition of every word is accomplished by close collaboration of bottom-up sub-word and word recognition neural networks with top-down cognitive word context expectations. The utility of this context appropriate collaboration is substantial savings in recognition time, accuracy and cortical neural processing resources. Repetition priming, the simplest form of context facilitation, has been studied extensively, but behavioral and cognitive neuroscience research has failed to produce a common shared model. Facilitation is attributed to temporary lowered word recognition thresholds. Recent fMRI evidence identifies frontal, prefrontal, left temporal cortex interactions as the source of this priming bias. Five experiments presented here clearly demonstrate that word recognition facilitation is a bias effect. Context-Biased Fast Accurate Recognition, a recurrent neural network model, shows how this anticipatory bias is accomplished by interactions among top-down conceptual cognitive networks and bottom-up lexical word recognition networks. Signal detection theory says that this facilitation bias is offset by the cost of miss-recognizing similar, but different words. However, the prime typically creates a temporary time-space recognition window within which probability of prime recurrence is substantially raised paradoxically transforming bias into sensitivity.