How do schizotypal and autistic traits alter semantic prediction error processing?

Testing opposite predictive profiles in one task

Predictive processing accounts have proposed that autism spectrum disorder (ASD) and schizophrenia spectrum disorder (SSD) sit at opposite ends of one computational axis: ASD is described as overweighting low-level sensory evidence, SSD as overweighting high-level prior beliefs, with both producing altered prediction error processing. This study tested that continuum hypothesis in subclinical trait expression rather than in clinical groups, using language comprehension as the test domain.

The design matters because it separates the two weights the hypothesis is about. The precision of the semantic prior was manipulated through sentence entropy, meaning how uncertain the sentence context leaves the final word. The precision of sensory evidence was manipulated by degrading the sentence-final word with noise vocoding at 1, 3, 6 or 12 channels, where clarity rises with channel number. Each of 200 sentences was heard once, and four counterbalanced groups rotated the clarity levels across participants and predictability conditions.

How prior and sensory weights were estimated

Two measurement layers were combined. A hierarchical Bayesian belief-updating model estimated how much each participant's responses were driven by prior beliefs versus sensory evidence, giving a computational readout of the weighting imbalance. Electroencephalography recorded the N400, a mean amplitude measure in the 300–500 ms window after the sentence-final word that indexes semantic prediction error: a larger N400 difference between expected and unexpected words means a stronger error signal.

The N400 analysis used linear mixed-effect models of mean amplitude, with schizotypal traits measured by the Schizotypal Personality Questionnaire, autistic traits by the Autism Spectrum Quotient, and verbal ability by the Verbal Comprehension Index from the WAIS-IV. Participants were 55 right-handed native German speakers aged 18 to 35 with normal hearing. Because the two trait dimensions correlated strongly in this sample (R = 0.51), the study notes limited power to dissociate disorder-specific mechanisms.

Schizotypy shifted weight toward prior beliefs

The computational model found that increasing schizotypal traits were associated with significant overweighting of prior beliefs. This is the study's clearest mechanistic result: the shift was in the model parameter describing relative reliance on the semantic prior, not merely in a raw response measure.

The neural data converged on the same direction. Higher schizotypy was reflected in a reduced semantic prediction error signal, indexed by smaller N400 differences between high predictability sentences and both low predictability sentences and high predictability mismatch sentences. In other words, the gap between expected and unexpected endings narrowed as schizotypal traits rose. The paper reports that this reduction was driven specifically by impaired processing of predictable input, with N400 amplitudes becoming more negative for sentence-final words that matched highly and moderately predictable contexts.

Autistic traits showed no sensory overweighting

The prediction for autistic traits was the mirror image: overweighting of low-level sensory evidence. The computational model did not show a significant shift with increasing autistic traits, so the central sensory-weighting claim was not supported in this task.

The electrophysiology was not entirely flat. A similar but weaker and less distinct pattern emerged, with only the N400 difference between high predictability and high predictability mismatch sentences decreasing. Post-hoc analysis could not identify which condition drove that effect. The authors interpret this as a possible subtle overweighting of prior beliefs that may only appear under specific task conditions, rather than as evidence for sensory overweighting.

Why language may differ from perceptual tasks

The authors address the mismatch with prior ASD literature directly. Evidence for sensory overweighting in ASD has largely come from perceptual and social decision-making tasks and remains heterogeneous overall. One possibility they raise is that predictive-processing alterations in ASD are domain-specific, with language processing relatively intact.

They also note that this explanation sits awkwardly with reports of speech-processing difficulties in ASD and with their own attenuated N400 finding. Their preferred reading is that both trait dimensions share a tendency to overweight semantic prior beliefs, consistent with the substantial symptom overlap between ASD and SSD and with the strong trait correlation in this sample. They suggest psychotic symptoms such as hallucinations and delusions may differentiate the two spectra more clearly than language paradigms do.

Verbal ability and the N400 boundary

Verbal comprehension was examined as a possible confound and as an individual-difference variable. The computational model showed no significant shift, but higher verbal ability trended toward greater reliance on sensory information, reflected electrophysiologically in a larger N400 difference between predictable and unpredictable endings after highly predictable sentences. Importantly, schizotypy was not associated with verbal comprehension, so prior overweighting in schizotypy cannot be explained by lower verbal ability.

Two measurement boundaries qualify the N400 results. Trials were averaged by sentence predictability for correct responses, collapsing across degradation levels because correct trials per level were too few for reliable separate analysis, so the joint effect of clarity and predictability on the N400 was not resolved. The authors also report that the N400-window modulation extended into later periods, with high predictability sentences eliciting more positive mean amplitudes than low and mismatch sentences, a pattern inconsistent with classic P600 findings; they suggest the robust N400 may have overlapped with or masked later components.

What the trait comparison actually supports

The study's contribution is a within-paper comparison of two trait dimensions on the same task, the same computational model and the same neural index. On that shared footing, schizotypy produced a significant prior-weighting shift backed by a reduced N400 prediction error signal, while autistic traits produced no computational shift and only a weaker, less localized N400 reduction. The diametrically opposed continuum was therefore not reproduced in subclinical language processing.

The practical priority for readers is to keep the two levels of evidence apart. The computational overweighting result is specific to schizotypy; the autistic-trait N400 pattern is an electrophysiological hint without a matching model parameter and without an identified driving condition. Because the sample was 55 healthy adults with strongly correlated traits, the appropriate next test is whether these dissociations hold in larger samples and clinical cohorts, where the heterogeneity of ASD and SSD can be separated rather than averaged together.

Sources used in this answer