Most previous computational studies of psychiatric speech have concentrated on lexical-conceptual meaning, using word embeddings and semantic similarity measures to quantify how ideas relate to one another across a stretch of discourse. That work has produced striking results, including demonstrations that semantic coherence degrades in psychosis in ways that defy simple similarity accounts. Yet meaning has at least two dimensions. Lexical-conceptual meaning concerns what words and concepts denote and how they cluster semantically. Referential meaning, by contrast, concerns how a speaker anchors those concepts to particular individuals and situations, maintaining a stable web of entities across sentences. When someone says that a neighbor complained and then adds that she would not stop, the listener must resolve that she refers back to the neighbor. Failures of this binding process, long noted clinically in formal thought disorder, had never been systematically quantified with modern language models across diagnostic groups until now.

The research team tested a specific hypothesis: that coreference would be harder for the automatic model to resolve in the speech of clinical participants than in controls. They assembled three groups of participants who produced spoken narratives, comprising 36 healthy controls, 41 individuals with schizophrenia spectrum disorder and 42 with major depressive disorder. The coreference resolution model was applied to transcripts of this speech, and its performance, essentially how accurately it could link referring expressions to the correct antecedents, was compared across groups. The results confirmed the prediction. Model performance was significantly lower in both the schizophrenia and depression groups relative to healthy controls, indicating that the referential threads running through clinical discourse are genuinely harder to track, even for a machine system trained on vast amounts of typical language.

What elevates the study beyond a simple group difference is its topological approach. The researchers constructed speech graphs, network representations in which linguistic units become nodes and the relations between them become edges, separately for referential structure and for lexical-semantic structure. Graph theory then allowed them to characterize the topology of each kind of meaning and, crucially, to compare the two. In healthy speech, the referential graph and the semantic graph diverge in characteristic ways: the web of entities a speaker tracks and the web of concepts they deploy are related but structurally distinguishable systems. In the schizophrenia group, the researchers found a reduced structural divergence between these two graphs, suggesting that the normal differentiation between referential and lexical-semantic organization is blunted. Meaning, in effect, loses some of its articulated internal architecture in schizophrenia spectrum disorder.

The correlational analyses sharpened the clinical picture considerably. Within the schizophrenia group, poorer model performance on coreference was negatively associated with positive symptoms, meaning that participants with more severe positive symptoms, the hallucinations, delusions and disorganized thinking that historically define psychosis, showed the greatest referential disruption. This aligns with a long clinical tradition linking formal thought disorder to breakdowns in the cohesion of discourse, but it now provides a quantitative, model-based handle on the phenomenon. Within the depression group, by contrast, coreference performance was associated with verbal memory measures rather than with symptom severity in the positive-symptom domain. This dissociation is theoretically important. It implies that although both disorders show referential difficulty at the group level, the cognitive machinery underlying that difficulty may differ: in schizophrenia it tracks the psychotic process itself, while in depression it may reflect the well-documented memory and executive impairments that accompany the illness.

The technical machinery behind these conclusions deserves attention, because it illustrates how computational linguistics is becoming a precision instrument for psychopathology. Coreference resolution is one of the harder problems in natural language understanding. A model must recognize that a definite description, a pronoun, a name and even an elliptical phrase can all denote the same individual, and it must do so across sentence boundaries where ambiguity abounds. When such a model, calibrated on ordinary language, stumbles more often on clinical speech, the errors are not random noise; they signal that the discourse itself provides weaker or more inconsistent cues about entity identity. Combined with speech graph topology, which quantifies properties such as connectivity and structural divergence between representational layers, the method yields a multidimensional profile of how meaning is organized, or disorganized, in an individual speaker.

The study forms part of a broader research program in which language is treated not merely as a symptom carrier but as a measurable biomarker of brain function. The same Barcelona group and their collaborators have previously shown progressive changes in descriptive discourse in first-episode schizophrenia using computational semantics, and related work has connected reduced linguistic coherence in psychosis to altered large-scale cortical hierarchy. The new findings extend this agenda by profiling two dimensions of meaning against each other rather than in isolation. The authors describe this as linking lexical-conceptual and referential-semantic alterations in a more integrated foundational model of language in major psychopathology. In practical terms, it means future computational assays of speech could in principle distinguish referential breakdown from semantic drift, refining both diagnosis and the tracking of illness course.

The clinical implications are potentially far-reaching. Speech samples are cheap, noninvasive and increasingly easy to collect, and language-based digital biomarkers are already being developed for early detection and monitoring of psychosis risk, including within European projects such as TRUSTING, which supported this work alongside the German Research Foundation. If coreference resolution performance and speech graph topology can be computed reliably from a few minutes of natural speech, they could complement clinical interviews with objective, quantifiable measures. The differential associations reported here, positive symptoms in schizophrenia and verbal memory in depression, hint that such measures might eventually help discriminate between conditions that overlap superficially, or track distinct therapeutic targets within a single patient over time. Much validation remains to be done, including longitudinal studies and replication across languages and clinical settings, but the direction of travel is clear.

The study also carries conceptual weight for theories of language in the mind and brain. Referential capacity, the ability to maintain a coherent cast of entities across discourse, has been proposed by some theorists, including Hinzen, to be a distinctive vulnerability point in schizophrenia, connected to the very architecture of human grammar and thought. The observation that referential and semantic graphs normally diverge structurally but converge in schizophrenia gives that proposal an empirical, topological form. Meanwhile, the depression findings caution against a single unified account of linguistic disruption in psychiatry: referential difficulty is not a psychosis-specific signature when it also appears, with different cognitive correlates, in mood disorder. Disentangling these threads is precisely what the new integrated framework is designed to do.

For now, the study stands as a demonstration that the fine grain of meaning, down to the humble pronoun and its antecedent, can be measured, modeled and related to symptoms and cognition in major mental illness. As language models grow more capable and clinical speech datasets grow larger, the boundary between computational linguistics and psychiatry is dissolving, and with it may come a new generation of tools that read the structure of thought directly from the structure of speech. The participants in the FOR2107 consortium whose narratives made this analysis possible contributed to a result that is as conceptually elegant as it is clinically promising: meaning, in its referential dimension, has a measurable topology, and in psychosis that topology changes.

Subject of Research: Computational analysis of coreference and speech graph topology in schizophrenia and major depressive disorder

Article Title: Co-reference and the topology of meaning in schizophrenia and depression

Article References: Co-reference and the topology of meaning in schizophrenia and depression. (n.d.). https://doi.org/10.1038/s41537-026-00802-6

Image Credits: AI Generated

DOI: 10.1038/s41537-026-00802-6

Keywords: schizophrenia, major depressive disorder, coreference resolution, natural language processing, speech graphs, computational psychopathology, formal thought disorder, positive symptoms, verbal memory, semantic topology, language biomarkers, digital psychiatry

The research examined 3,943 participants who took part in the Age 46 Biomedical Sweep of the 1970 British Cohort Study. Just over half, 53.3 percent, were female, and all participants included in the analysis had the relevant activity, behavior, cognitive and background data. The cohort follows people born in Britain in 1970, allowing researchers to study health and behavior in a population reaching midlife at the same period. The investigators focused on four sedentary behaviors: reading, internet use, playing video games and watching television. They classified the first three as more cognitively active activities and television viewing as a more cognitively passive activity. This distinction was not intended to imply that every individual engages with media in the same way. Reading can range from challenging nonfiction to light material, internet use can involve learning or passive scrolling, and games vary widely in complexity. The categories were broad behavioral measures, but they offered a way to test whether different kinds of sitting were linked to cognition rather than treating all sedentary time as one uniform exposure.

To measure movement, participants wore activPAL3 accelerometers, devices designed to distinguish body positions and record physical activity. The researchers also collected self-reported estimates of time spent in the specified sedentary activities. Cognitive performance was assessed using three tests that sample different mental abilities. An auditory verbal learning test involving 10 words evaluates verbal learning and memory by examining how well a person can acquire and recall information. A category verbal fluency test asks participants to name animals, providing a measure influenced by language retrieval, semantic memory and executive control. A two-letter cancellation task measures aspects of attention and processing speed, requiring participants to identify target items efficiently among distractors. Scores from the assessments were standardized as z-scores and combined into a global cognition score, the study’s primary outcome. Standardization places results from different tests on a common scale, allowing them to contribute to a composite measure despite having different scoring systems.

The statistical analysis used linear regression models to estimate how an additional hour per day of each activity was associated with cognitive scores. The models adjusted for sex assigned at birth, moderate-to-vigorous physical activity, occupational physical activity and educational level. These adjustments are important because cognitive scores and daily routines can be related to factors other than the activity under investigation. Education, for example, may be associated with both the kinds of leisure activities people choose and their performance on cognitive tests. Occupational activity and exercise could also influence the amount of time available for sitting and may be related to health. The researchers additionally conducted sex-stratified analyses, examining associations separately by sex. Because the study was cross-sectional, however, the measurements of behavior and cognition represented the same broad period. Regression can identify patterns of association after accounting for measured covariates, but it cannot establish which behavior came first or whether an unmeasured factor affected both.

Reading showed the clearest association among the sedentary activities studied. In adjusted pooled models, each additional hour per day of reading was associated with a 0.042 standard-deviation higher global cognition score, with a standard error of 0.008 and a 95 percent confidence interval from 0.026 to 0.058. The false discovery rate-adjusted probability value was below 0.001. Internet use was also positively associated with global cognition, although the estimated relationship was smaller: each additional hour per day corresponded to a 0.017 standard-deviation higher score, with a standard error of 0.006 and a 95 percent confidence interval from 0.005 to 0.029. Its adjusted probability value was 0.030. These estimates describe differences between people with different reported activity patterns; they do not predict how much an individual’s cognition would change after deliberately adding an hour of reading or internet use. The size of an association can also be influenced by measurement accuracy, the kinds of content people choose and the social or educational circumstances surrounding those activities.

The two other behaviors did not produce statistically significant associations with global cognition. Television viewing had an estimated coefficient of minus 0.006 standard deviations per additional hour per day, with a standard error of 0.005 and a 95 percent confidence interval ranging from minus 0.017 to 0.004. The false discovery rate-adjusted probability value was 0.297, meaning the analysis did not provide strong evidence of a relationship in either direction. Video game playing had an estimated positive coefficient of 0.015, with a standard error of 0.009 and a confidence interval from minus 0.002 to 0.032; its adjusted probability value was 0.128. The result therefore did not meet the study’s threshold for a statistically significant association. A non-significant result is not proof that an activity has no cognitive relevance. It may reflect limited time spent on the behavior, substantial variation in the types of games or imprecise self-reporting. It does mean that, within these data and models, the researchers could not distinguish the observed estimates reliably from chance variation.

Physical activity showed a separate and important pattern. Moderate-to-vigorous physical activity was positively associated with global cognition after adjustment, with an estimated coefficient of 0.069 standard deviations for each additional hour per day. The standard error was 0.021, the 95 percent confidence interval extended from 0.027 to 0.110, and the false discovery rate-adjusted probability value was 0.009. This result supports the study’s broader interpretation that replacing a sedentary lifestyle with more movement may be relevant to cognitive health. Yet the analysis does not establish that exercise directly improved the participants’ test scores. People who engage in more vigorous activity may differ in sleep, health, income, social contact, stress, diet or other characteristics that were not fully captured by the available adjustments. The study also did not conclude that cognitively engaging sitting compensates for insufficient physical activity. Rather, its conclusion presents two potentially complementary ideas: accumulating moderate-to-vigorous activity remains a priority, while the composition of unavoidable sitting may also be worth considering.

The biological explanation for the findings remains open. Reading commonly requires sustained attention, visual processing, language comprehension, working memory and integration of information across sentences or pages. Depending on the task, internet use can involve searching, evaluating sources, navigating interfaces, learning and making choices. Such repeated mental operations could be markers of cognitive engagement, but the study did not measure the neural activity or learning processes produced by any particular book, website or online session. People with stronger cognitive abilities may also be more likely to choose reading or intellectually demanding online activities, creating the possibility of reverse causation. Early differences in education, occupational opportunities or lifelong reading habits could influence both activity choices and later test performance. The single assessment of behavior likewise cannot show whether long-term exposure matters more than current routines. Longitudinal studies, repeated cognitive testing and more detailed measures of content and context will be needed to determine whether changing sedentary behaviors alters cognitive trajectories.

For now, the results offer a qualified message for middle-aged adults and for researchers designing public-health advice. Sitting should not be treated as a single category when its cognitive consequences are being studied, but neither should the findings be converted into a prescription that reading or internet use guarantees sharper thinking. The evidence supports maintaining or increasing moderate-to-vigorous physical activity and suggests that cognitively active pursuits may be preferable ways to occupy some sedentary time. Future research could test whether associations persist over years, differ across cognitive domains or vary according to the intensity and purpose of digital use. It could also examine whether short periods of movement interrupting sitting provide benefits beyond total daily activity. Because the current analysis used observational, cross-sectional data, its strongest contribution is to identify a pattern for further testing: among adults in midlife, reading and internet use were linked with better global cognitive scores, while television and video games were not. The pattern is intriguing, but experiments and longitudinal evidence are required before it can be considered causal.

Subject of Research: Associations between sedentary activities, physical activity, and cognitive function in middle-aged adults

Article Title: Cognitively engaging sedentary activities are associated with better cognitive function in middle-aged adults: a cross-sectional study using the 1970 British Cohort Study

Article References: Qadi, L. A., Saucier, D., Rayner, S., Pitre, J., Savoie, M., Stadler, J., Jbilou, J., & O’Brien, M. W. (2026). Cognitively engaging sedentary activities are associated with better cognitive function in middle-aged adults: a cross-sectional study using the 1970 British Cohort Study. Journal of Activity, Sedentary and Sleep Behaviors. https://doi.org/10.1186/s44167-026-00110-5

Image Credits: AI Generated

DOI: 10.1186/s44167-026-00110-5

Keywords: Cognitive function, Middle age, Sedentary behavior, Reading, Internet use, Physical activity, Television, Video games, Verbal memory, Executive function, Cognitively, engaging