Can predictive coding along the auditory pathway explain deviance detection and adaptation in one framework?

Why adaptation and deviance detection can be two sides of one process

The core idea is economical: the brain does not passively wait for sound and then react. It generates an expectation about what the next sound will be, and the neural response reflects the mismatch between expectation and input. When a sound repeats, the expectation becomes strong and the mismatch shrinks, so the response declines. That decline is what researchers call adaptation, or more specifically stimulus-specific adaptation when it is tuned to a particular sound rather than to sound in general. When a different sound appears, the expectation is violated and the mismatch grows, producing deviance detection.

Framed this way, adaptation is not a separate low-level fatigue effect that has to be explained by different machinery from deviance detection. Both are readouts of the same prediction-error signal: small when the input is expected, large when it is not. The review by Carbajal and Malmierca integrates evidence on stimulus-specific adaptation and mismatch negativity, a scalp-recorded response to unexpected sounds, and proposes a shared vocabulary built on this predictive-coding interpretation [1]. That shared vocabulary is the practical payoff: findings from cellular recordings and from human scalp recordings can be described in the same terms instead of as unrelated phenomena.

Does this happen only in the cortex, or further down the pathway?

A natural objection is that prediction sounds like a high-level cortical job, so adaptation and deviance detection should be cortical phenomena. The framework reviewed by Carbajal and Malmierca explicitly traces predictive coding from subcortical roots up to cortical deviance detection [1], meaning the comparison between expected and actual sound is not treated as cortex-only. That matters for the question, because if deviance detection begins below the cortex, a single framework has to accommodate both levels rather than assigning adaptation to one stage and deviance detection to another.

Related work extends the predictive-coding lens to subcortical auditory processing in a clinical context. In dyslexia, Järvikylä and colleagues describe atypical predictive coding specific to the left subcortical auditory pathway [3]. This is evidence that predictive coding is a useful description of subcortical auditory function, though the study is about dyslexia rather than about reconciling adaptation with deviance detection, so it supports the framework's reach more than its internal unification.

What the framework looks like when prediction misfires

The framework earns credibility when it can say something about cases where auditory perception goes awry. In tinnitus, Schilling and colleagues argue that intrinsic neural noise is generated and amplified along the auditory pathway as a compensatory mechanism to restore normal hearing, and that this noise can be misinterpreted as input and perceived as tinnitus [2]. They formalize this in a Bayesian brain framework, where the percept combines a prior prediction with the bottom-up signal, and propose predictive coding together with adaptive stochastic resonance, a bottom-up amplifying mechanism, as the two principles with the most explanatory power for phantom sounds [2].

Adaptation itself appears altered in tinnitus. Colak and colleagues report altered adaptation to repetitive auditory stimulation in tinnitus, measured with evoked potentials, and interpret it within a predictive-coding framework [5]. Read alongside the tinnitus account above, this is consistent with the idea that the same predictive machinery that normally suppresses responses to repetition is behaving differently when phantom perception arises. It is worth being precise about what this shows: these are observations of altered adaptation and a proposed mechanistic account, not a demonstration that faulty prediction causes tinnitus.

How far does the unified account actually reach?

The unification is a proposed interpretive framework, and the evidence behind it is uneven across the phenomena it covers. The strongest direct statement of the unified view is the review linking stimulus-specific adaptation and mismatch negativity under predictive coding [1]. Other studies apply predictive coding to auditory deviance detection in the context of schizophrenia-related cortical processing [4], or to subcortical processing in dyslexia [3], without testing whether adaptation and deviance detection share a single mechanism.

There is also a conceptual caution embedded in this literature. Järvikylä and colleagues note that adaptation models are a phenomenological description of stimulus-driven response reduction, in contrast to the predictive-coding account [3]. In plain terms, observing that a response shrinks with repetition does not by itself prove that prediction is the reason; adaptation can be described without invoking prediction at all. So the honest answer is that predictive coding can explain both adaptation and deviance detection within one framework, and there is a serious proposal and supporting evidence for doing so [1], but the framework's ability to fully replace purely descriptive adaptation accounts is still being worked out.

Sources used in this answer

Related research questions