Periodic Reporting for period 1 - DELTA-LANG (The Delta of Language)
Berichtszeitraum: 2024-06-01 bis 2025-11-30
DELTA-LANG aims to understand the fundamental role of speech and language in thinking and behaviour. By grasping the association between language and thought, we aim to use speech as a marker of mental states, in particular for psychosis. We will leverage changes in speech to inform risk assessment in psychosis and predict imminent relapse at clinically relevant time scales. Psychosis affects approximately 3% of the population and often recurs after remission, with 80% of patients experiencing multiple relapses. It is an important clinical utility, but also acts as a showcase to explore changes in language to reflect actual mental states.
Our project tests the hypothesis that linguistic change – the delta of language – can serve as a personalized and interpretable predictor of state transitions. In this case, from remission to psychosis. To achieve this, DELTA-LANG integrates psychiatry, linguistics, neuroimaging, and e-health in four synergistic work packages:
1. Define theory-driven, generalizable language metrics across languages.
2. Link these metrics to brain connectivity using dense sampling during transitions from psychosis to remission.
3. Validate predictive power retrospectively in a large longitudinal cohort.
4. Translate findings into a remote monitoring system.
Expected impact includes fundamental knowledge on how language changes can be used as a window into mental state. Including earlier intervention windows, reduced hospitalizations, and a scalable methodology for cognitive phenotyping in real-world settings. Beyond psychiatry, the project will advance fundamental knowledge on how language relates to thought and brain function.
Our project tests the hypothesis that linguistic change – the delta of language – can serve as a personalized and interpretable predictor of state transitions. In this case, from remission to psychosis. To achieve this, DELTA-LANG integrates psychiatry, linguistics, neuroimaging, and e-health in four synergistic work packages:
1. Define theory-driven, generalizable language metrics across languages.
2. Link these metrics to brain connectivity using dense sampling during transitions from psychosis to remission.
3. Validate predictive power retrospectively in a large longitudinal cohort.
4. Translate findings into a remote monitoring system.
Expected impact includes fundamental knowledge on how language changes can be used as a window into mental state. Including earlier intervention windows, reduced hospitalizations, and a scalable methodology for cognitive phenotyping in real-world settings. Beyond psychiatry, the project will advance fundamental knowledge on how language relates to thought and brain function.
Work performed during the first reporting period concentrated on infrastructure, data preparation and pilot analyses. We finalized the harmonized speech protocol and secured ethics clearance at all sites. Data collection has started in multiple languages (Dutch, German, Italian, Turkish) for WP1, enabling initial tests of semantic similarity and co-reference resolution metrics. We implemented secure data storage and analysis pipelines and piloted NLP models (word embeddings and transformer-based models) on early samples to assess sensitivity to symptom variability. In WP2, preparatory work for dense sampling was completed: imaging sequences were optimized and piloted in healthy controls, and the study initiation package was registered. WP3 preparations included pipeline refining algorithms for longitudinal prediction using HAMLETT data. WP4 activities focused on design specifications for the remote monitoring system and GDPR-compliant architecture.
Results so far go beyond the state of the art by creating the first integrated framework combining linguistic theory, NLP, and neuroimaging for translating changes in language to mental and brain states. Unlike previous cross-sectional studies, DELTA-LANG emphasizes longitudinal, individual-level modeling and interpretable metrics grounded in semantic theory. Early achievements include: (i) a unified pipeline for cross-linguistic speech analysis; (ii) pilot validation of semantic distance and referential graph metrics; and (iii) secure infrastructure for remote speech collection. These advances set the stage for dense sampling and predictive modeling in the next period. The project is on track to deliver a paradigm shift in relapse prevention and scalable cognitive phenotyping.