Psychosis-related disorganisation impairs the construction and verbal expression of relational cognitive maps from visual experience

The human brain constructs relational cognitive maps that support flexible inference across domains by organising fragmented information into structured representations. The recently proposed “Shallow Cognitive Map” hypothesis suggests that disruptions in cognition, language, and behaviour in psychosis may arise from imprecise and unstable internal representations of relational structure. However, direct empirical evidence that supports this hypothesis is limited.

Here, we tested whether psychosis-related cognitive disorganisation impairs the construction of cognitive maps. Cognitive disorganisation (CD), a core feature of psychosis, is associated with impairments in thought disorder and the organisation of relational structure. We also examined whether such impairments depend on whether relational structure is acquired from visual input which requires the abstraction of relational structure using perceptual experience, or linguistic input which provides an already structured representational scaffold.

Participants (N = one hundred and eleven) learned the two-dimensional relational structure between two variables: age group and time of day, from either visual or linguistic input. Such learning relies on the human brain’s ability to integrate relational information into cognitive maps that support inference about unobserved structure. Participants then completed similarity judgment tasks and reward-learning tasks probing latent structure inference. They were subsequently asked to provide both a summary of their learning strategies and to teach the task to others. Statistical analyses included correlation analyses, chi-square tests, and regression models.

Participants with higher CD showed reduced sensitivity to latent relational structure when it had to be inferred from visual learning (r= −0.33, p = 0.017), but not from linguistic learning (r = −0.08, p= 0.554, BF₀₁ = 6.45). In addition, CD predicted weaker spatial generalisation in reward learning, captured by the length-scale parameter (λ) of a Gaussian Process Regression model (r= −0.31, p = 0.027), indicating reduced generalisation across the inferred relational space. These findings suggest that CD selectively disrupts the construction of spatial representations when they must be internally generated from visual input.

Moreover, spatial structure in verbal speech was more likely to be referred to when participants taught the task to others compared with self-report learning strategies in the visual condition (χ² = 8.27, p = 0.016), suggesting that teaching promotes the externalisation of latent relational knowledge. This externalisation effect was driven by individuals with lower CD (OR = 8.11, p = 0.010) but absent in those with higher CD (OR = 1.59, p = 1.000), indicating a difficulty in articulating spatial relational structure in verbal communication, potentially reflecting impairments in the underlying representations. No comparable effects were observed following linguistic learning, suggesting that language-based input provides an alternative representational scaffold that reduces reliance on internally constructed maps.

These findings provide converging evidence from behavioural, computational, and linguistic measures for the shallow cognitive map hypothesis, revealing that cognitive disorganisation selectively disrupts the construction and expression of relational cognitive maps from visual experience. These findings suggest that disruptions in relational structure can be detected not only in task performance but also in the structure of verbal explanations, providing a potential avenue for language-based assessment of psychiatric symptoms.

References

1. , A Musa, S Khan, M Mujahid, M El-Gaby, The shallow cognitive map hypothesis: A hippocampal framework for thought disorder in schizophrenia. Schizophrenia 8, 34 (2022).
2. , MM Nour, Y Liu, M El-Gaby, RA McCutcheon, RJ Dolan, Cognitive maps and schizophrenia. Trends Cogn. Sci. 29, 184–200 (2025).
3. , X Wu, et al., Disorganization impairs cognitive maps built from visual inputs. bioRxiv pp. 2025–12 (2025).
4. , MM Nour, DC McNamee, Y Liu, RJ Dolan, Trajectories through semantic spaces in schizophrenia and the relationship to ripple bursts. Proc. Natl. Acad. Sci. 120, e2305290120 (2023).
5. , I Fradkin, RA Adams, N Siegelman, R Moran, RJ Dolan, Latent mechanisms of language disorganization relate to specific dimensions of psychopathology. Nat. Mental Heal. 2, 1486–1497 (2024).