Hearing loss as an inference problem: sensory precision, conversational repair, and partner adaptation

Conversational understanding is often approached as a problem of successful signal transmission. This paper develops a complementary account in which conversation is treated as a reciprocal process of inference and action: interlocutors must infer the latent causes of one another’s signals, estimate the reliability of the available evidence, and act to keep uncertainty sufficiently manageable for coordinated exchange. Hearing loss provides a theoretically revealing case for understanding conversation as reciprocal inference because it perturbs the generative process through which speech causes produce auditory observations, while also increasing the importance of contextual inference, precision control, conversational repair, and partner adaptation. We develop an active inference framework with three principal contributions. First, we distinguish the sensory perturbation produced by hearing loss from the listener’s generative model representation of that perturbation through reduced, redistributed, or potentially miscalibrated likelihood precision. Second, we differentiate compensatory precision control, delayed commitment, covert repair, and overt conversational repair, treating overt repair as an epistemic policy whose value depends on expected information gain, prior preferences, learned policy habits, and interpersonal consequences. Third, we extend the framework beyond the isolated listener by formalising how interlocutors infer one another’s states, select partner-sensitive policies, and update partner models through repeated interaction. The proposal therefore connects foundational questions with a clinically and socially consequential sensory condition. Rather than treating hearing loss solely as an individual perceptual deficit, it examines how altered sensory evidence changes the coupled inferential dynamics through which listeners and communication partners jointly maintain understanding. The framework generates empirically discriminating predictions and outlines how these can ultimately be evaluated through model comparison, behavioural and physiological measures, dyadic interaction paradigms, and longitudinal research. Overall, the proposal recasts hearing-impaired conversation as the joint regulation of uncertainty under asymmetric sensory and interpersonal constraints.