Predictive coding
Does the brain predict when a story character will come back?
Open access · cc by · source: Europe PMC
When a sentence structure made a character's return more predictable, the brain's response to that return was dampened in dorsal-stream regions, as predictive coding expects.
Study at a glance
- Design
- Human experiment — Within-subject 2 x 2 fMRI design embedded in spoken German stories: voice of the context sentence (active vs passive) crossed with verb causality (high vs low), measured at the later re-mention of the same character.
- N
- N=20 · 20 analysed right-handed native German speakers (22 scanned, 2 excluded for movement); a separate online pretest of the stories collected ratings from 177 people.
- Population
- Young adult monolingual German speakers recruited at the University of Marburg
- Outcome
- BOLD response to the re-mentioned referent across the whole brain
Structured fields used in claim comparison tables when every cited study has a complete layer.
Key findings
Listeners answered 90% of comprehension questions correctly, showing they were attending. Characters introduced with a passive sentence produced weaker, below-baseline responses at re-mention in regions including the right inferior parietal lobule, frontal gyri, cingulate cortex and cerebellum, consistent with predicted input being suppressed. Causality had separate effects: higher responses after high-causality events in left occipital, fusiform and right insula regions, but the reverse in the middle temporal gyrus. An interaction appeared only in the supplementary motor area, in the opposite direction to an attention-based account.
Methodology
Participants listened to 20 two-minute German stories in the scanner. Inside each story, a character was introduced either in an active sentence or in a passive one (passives make the character likely to be mentioned again), and with either a highly causal verb like 'push' or a low-causality one like 'hold in high esteem'. The team measured brain responses at the moment the same character was mentioned again two sentences later, keeping the actual words of that re-mention identical across conditions.
Limitations
The sample is small and the whole-brain threshold used a voxel level of p = 0.005 with cluster correction, which is relatively lenient. Naturalistic speech could not be fully matched for prosody; passive sentences were longer and subtle pitch differences may have contributed to the signal, as the authors acknowledge. There was no behavioural measure of prediction itself, so the 'when' versus 'what' interpretation is inferred from brain patterns, and some causality findings (e.g., in visual-stream areas) were explained only after the fact.
How this study connects
Role on claims
Each row is a claim on a concept or method page where this paper supports, challenges, or qualifies the statement. Roles are hand-checked — not a model guess.
Not yet placed on a claim. This paper has study layers, but no concept page yet cites it as support, challenge, or qualifier.
Related papers in this topic
Same topic cluster — not a recommendation engine.