Does signal reduction imply predictive coding in models of spoken word recognition?

Sahil Luthra,James S. Magnuson,Monica Y. C. Li,Christian Brodbeck,Heejo You

doi:10.3758/s13423-021-01924-x

Abstract

Pervasive behavioral and neural evidence for predictive processing has led to claims that language processing depends upon predictive coding. Formally, predictive coding is a computational mechanism where only deviations from top-down expectations are passed between levels of representation. In many cognitive neuroscience studies, a reduction of signal for expected inputs is taken as being diagnostic of predictive coding. In the present work, we show that despite not explicitly implementing prediction, the TRACE model of speech perception exhibits this putative hallmark of predictive coding, with reductions in total lexical activation, total lexical feedback, and total phoneme activation when the input conforms to expectations. These findings may indicate that interactive activation is functionally equivalent or approximant to predictive coding or that caution is warranted in interpreting neural signal reduction as diagnostic of predictive coding.

Highlights

Theories of spoken language processing posit that listeners continually engage in predictive processing
Electrophysiological data support the idea of prediction; listeners show differential neural responses to predicted words compared to unexpected words (e.g., Kutas & Hillyard, 1980, 1984), with research suggesting that these responses index a process of pre-activation
We provide a brief overview of the Gagnepain et al (2012) study in order to define clear empirical targets for subsequent simulations. These patterns are observed in the dynamics of the TRACE model

Summary

Introduction

Theories of spoken language processing posit that listeners continually engage in predictive processing. TRACE incorporates mechanisms that should strengthen predictable inputs: Excitatory feedback connections from higher layers enhance signals consistent with higher level representations (which could, e.g., activate lexically consistent phonemes in advance of direct bottom-up support), and lateral inhibition within layers further enhances dominant signals (what Blank & Davis, 2016, term “signal sharpening”) The goal of this investigation is to consider whether a reduction of signal for expected inputs should constitute evidence for predictive coding. If the pre-DP segment matched a source word and an unconsolidated novel word (as in the pre-DP segment mushroo-, since mushrood had not yet been consolidated into the lexicon), there was a higher prediction error (since only one lexical entry supported the prediction) and correspondingly higher STG activity. The authors argued that these findings constituted evidence for predictive coding

Methods

Results

Results and discussion