Predicting Cognitive Function Using Transformer-Derived Speech Representations and Longitudinal Coherence Features

Topic: Brain imaging, connectivity and assessment

Dementia affects more than 55 million people worldwide, and its progressive decline is difficult to track using infrequent in-person assessments, which can miss subtle changes between visits and add to clinician burden. One widely used measure, the Mini-Mental State Examination (MMSE), is administered intermittently and remains subject to inconsistent scoring judgment, limiting early detection. Prior speech- based machine learning approaches have largely focused on cross-sectional classification rather than longitudinal cognitive forecasting. We introduce a longitudinal, patient-level framework that combines transformer-derived semantic speech representations with longitudinal speech-change features and clinical history to forecast a patient’s future MMSE score from their history of prior visits. To our knowledge, this is the first frame- work to unite transformer-derived speech encoding with longitudinal forecasting of cognitive severity, rather than single-visit classification alone. We evaluate this framework on longitudinal transcripts from the DementiaBank Pitt Corpus using a LightGBM gradient-boosted regression model, validated with patient-grouped cross-validation to prevent identity leakage between training and evaluation folds. The model fore- casts future MMSE scores with high accuracy and stability across folds ( R 2 = 0.840 ± 0.015, RMSE = 2.75, MAE = 2.02, Pearson r = 0.917), with transformer-derived speech representations contributing meaningful predictive signal alongside clinical history. Ablation analysis demonstrated that transformer-derived speech representations provided complementary predictive information beyond clinical variables. These results establish speech as a viable longitudinal digital biomarker of cognitive decline, offering a low-burden complement to intermittent clinical assessment that could enable earlier detection and more frequent monitoring, supporting better-timed care decisions for patients with dementia.

Key findings

• Ablation analysis demonstrated that transformer-derived speech representations provided complementary predictive information beyond clinical variables.
• These results establish speech as a viable longitudinal digital biomarker of cognitive decline, offering a low-burden complement to intermittent clinical assessment that could enable earlier detection and more frequent monitoring, supporting better-timed care decisions for patients with dementia.

Study details

• Study designcross-sectional
• Evidence strengthModerate

Read the original

• DOI: 10.64898/2026.08.04.26359730
• Original source →

Free copy unknown: no free-copy check has an answer for this record yet. That does not mean it is paywalled.

@article{devatha2026-f7db6824,
  author  = {Devatha, D. and Xiao, J.},
  title   = {{Predicting Cognitive Function Using Transformer-Derived Speech Representations and Longitudinal Coherence Features}},
  journal = {medRxiv},
  year    = {2026},
  doi     = {10.64898/2026.08.04.26359730},
  url     = {https://doi.org/10.64898/2026.08.04.26359730}
}

Related studies in Cognitive tests and assessment

• Identifying and Predicting Cognitive Decline Using Multi-Modal Sensor Data and Machine Learning Approach Research square · 2025-06-18 · usefulness score 16/100
• Dementia classification using two-channel electroencephalography features Scientific Reports · 2025-04-03 · usefulness score 43/100
• Using Digital Speech Assessments to Detect Early Signs of Cognitive Impairment Frontiers in digital health · 2021-10-27 · usefulness score 50/100
• Predicting clinical decline and conversion to Alzheimer's disease or dementia using novel Elecsys Aβ(1-42), pTau and tTau CSF immunoassays Scientific reports · 2019-12-13 · usefulness score 71/100
• Unveiling 30 years of research on speech biomarker of dementia using text mining Digital health · 2025-07-29 · usefulness score 59/100
• Tablet-Based Automatic Assessment for Early Detection of Alzheimer's Disease Using Speech Responses to Daily Life Questions Frontiers in digital health · 2021-03-17 · usefulness score 45/100

Similar studies

• Speaker-Wise Dementia Screening fromNaturalistic Speech with Lexical Entropy, Semantic Drift, and Multimodal Fusion 2026 · Digital health tools
• AD-BERT: Using pre-trained language model to predict the progression from mild cognitive impairment to Alzheimer's disease 2023 · Digital health tools
• Validation of the Remote Automated ki:e Speech Biomarker for Cognition in Mild Cognitive Impairment: Verification and Validation following DiME V3 Framework 2022 · Cognitive tests and assessment
• Alzheimer's Disease Detection via Large Pre-Trained Speech and Language Models: Taxonomy, Progress, and Open Challenges 2026 · Digital health tools
• Transformer-based deep neural network language models for Alzheimer's disease risk assessment from targeted speech 2021 · Digital health tools
• Evaluating Patient Representation Strategies for Early Prediction of Alzheimer’s Disease and Related Dementias from Longitudinal EHRs 2026 · Mild cognitive impairment

More on Cognitive tests and assessment

• A review that covers this ground Decreased Serum Brain-Derived Neurotrophic Factor (BDNF) Levels in Patients with Alzheimer's Disease (AD): A Systematic Review and Meta-Analysis International journal of molecular sciences · 2019-01-10 · usefulness score 74/100
• Newest in this subject A digital programme to prevent falls and improve wellbeing in people living with dementia in the community: Quantitative results from the KOKU-LITE feasibility study medRxiv · 2026-10-29 · Preprint — not yet peer-reviewed · usefulness score 16/100
• Newest in this subject Factors Influencing the Implementation of Computerized Cognitive Training in Chinese Community-Dwelling Elderly: A Qualitative Study Research Square · 2026-10-28 · Preprint — not yet peer-reviewed · usefulness score 16/100

See all studies in Cognitive tests and assessment →