Phoneme restoration and empirical coverage of interactive activation and adaptive resonance models of human speech processing

Lexical Feedback in the Time-Invariant String Kernel (TISK) Model of Spoken Word Recognition

The Time-Invariant String Kernel (TISK) model of spoken word recognition (Hannagan, Magnuson & Grainger, 2013; You & Magnuson, 2018) is an interactive activation model with many similarities to TRACE (McClelland & Elman, 1986). However, by replacing most time-specific nodes in TRACE with time-invariant open-diphone nodes, TISK uses orders of magnitude fewer nodes and connections than TRACE. Although TISK performed remarkably similarly to TRACE in simulations reported by Hannagan et al., the original TISK implementation did not include lexical feedback, precluding simulation of top-down effects, and leaving open the possibility that adding feedback to TISK might fundamentally alter its performance. Here, we demonstrate that when lexical feedback is added to TISK, it gains the ability to simulate top-down effects without losing the ability to simulate the fundamental phenomena tested by Hannagan et al. Furthermore, with feedback, TISK demonstrates graceful degradation when noise is added to input, although parameters can be found that also promote (less) graceful degradation without feedback. We review arguments for and against feedback in cognitive architectures, and conclude that feedback provides a computationally efficient basis for robust constraint-based processing.

Is prediction nothing more than multi-scale pattern completion of the future?

While the notion of the brain as a prediction machine has been extremely influential and productive in cognitive science, there are competing accounts of how best to model and understand the predictive capabilities of brains. One prominent framework is of a "Bayesian brain" that explicitly generates predictions and uses resultant errors to guide adaptation. We suggest that the prediction-generation component of this framework may involve little more than a pattern completion process. We first describe pattern completion in the domain of visual perception, highlighting its temporal extension, and show how this can entail a form of prediction in time. Next, we describe the forward momentum of entrained dynamical systems as a model for the emergence of predictive processing in non-predictive systems. Then, we apply this reasoning to the domain of language, where explicitly predictive models are perhaps most popular. Here, we demonstrate how a connectionist model, TRACE, exhibits hallmarks of predictive processing without any representations of predictions or errors. Finally, we present a novel neural network model, inspired by reservoir computing models, that is entirely unsupervised and memoryless, but nonetheless exhibits prediction-like behavior in its pursuit of homeostasis. These explorations demonstrate that brain-like systems can get prediction "for free," without the need to posit formal logical representations with Bayesian probabilities or an inference machine that holds them in working memory.

Phoneme restoration and empirical coverage of Interactive Activation and Adaptive Resonance models of human speech processing

Magnuson [J. Acoust. Soc. Am. 137, 1481-1492 (2015)] makes claims for Interactive Activation (IA) models and against Adaptive Resonance Theory (ART) models of speech perception. Magnuson also presents simulations that claim to show that the TRACE model can simulate phonemic restoration, which was an explanatory target of the cARTWORD ART model. The theoretical analysis and review herein show that these claims are incorrect. More generally, the TRACE and cARTWORD models illustrate two diametrically opposed types of neural models of speech and language. The TRACE model embodies core assumptions with no analog in known brain processes. The cARTWORD model defines a hierarchy of cortical processing regions whose networks embody cells in laminar cortical circuits as part of the paradigm of laminar computing. cARTWORD further develops ART speech and language models that were introduced in the 1970s. It builds upon Item-Order-Rank working memories, which activate learned list chunks that unitize sequences to represent phonemes, syllables, and words. Psychophysical and neurophysiological data support Item-Order-Rank mechanisms and contradict TRACE representations of time, temporal order, silence, and top-down processing that exhibit many anomalous properties, including hallucinations of non-occurring future phonemes. Computer simulations of the TRACE model are presented that demonstrate these failures.