Schizophrenia is associated with a wide range of language alterations. Interestingly, the recent predictive coding account does not only provide a testable theory for the explanation of positive symptoms in schizophrenia, but is also one of the most promising theories of language processing. Experimental evidence of the predictive coding account of psychosis suggests that an overweighing of sensory likelihood may occur at lower levels (e.g., early sensory processing areas) due to increased dopamine activity causing aberrant salience and leading to delusions, while an overweighting of prior beliefs at higher levels, potentially caused by altered glutamatergic receptor signalling, may cause hallucinations. The interplay of prior information and sensory likelihood can be tested efficiently during language processing, due to its predictive nature. This study therefore investigates (1) how the use of prior knowledge relative to sensory information in a predictive language task changes when patients with schizophrenia transition from acute psychosis to psychotic remission, and (2) whether an overreliance on prior semantic knowledge induces conditioned hallucinations in a state of acute psychosis, but not in a state of psychotic remission.
Using a longitudinal approach, we assessed 20 patients with schizophrenia first during a state of active psychosis and then, after approximately three months, in a state of psychotic remission. Controls were assessed at two similarly delayed time points. All participants completed a predictive language paradigm in which they listened to sentences of varying predictability (e.g., high predictability: “Goethe was a famous German … poet”; low predictability: “Next to the window was a … hole”). The final word of each sentence (e.g., “poet” or “hole”) was degraded in clarity using a noise vocoding algorithm; degradations were available in four levels from fully unintelligible to fully intelligible. After listening to the full sentence, participants were asked to report the word they perceived, assessing conditioned hallucinations. We fitted a linear Bayesian regression model to estimate the prior weight which presents the relative strength of the prior over the sensory input. Using a repeated measurement model, we compared prior weights and conditioned hallucinations across time points and groups. Furthermore, we correlated those with positive symptoms at both time points.
The modelling results revealed that prior knowledge relative to sensory information was overweighed in patients compared to controls, but especially during the first time point, when patients were in a state of acute psychosis. The same relationship was detected for conditioned hallucinations. Furthermore, patients with stronger psychotic symptoms perceived more conditioned hallucinations, but this positive correlation was only present during the acute psychotic state and absent during the state of remission.
This study shows that prior knowledge was overweighted during a language perception task in schizophrenia patients during an acute state of psychosis, and less so during psychotic remission. Importantly, overweighting prior knowledge and stronger psychotic symptoms were linked to experiencing more conditioned hallucinations. Taken together, this study provides novel evidence of how alterations in predictive language processing change when schizophrenia patients change from an acute to a remitted state of psychosis and how this is contributes to the formation of hallucinations.