Dementia Language Models: a generalizable and controllable representation of cognitive impairment
We introduce Dementia Language Models (DLMs)--a generalizable and controllable representation of cognitive impairment through language--alongside an evaluation framework for establishing their validity and clinical grounding. DLMs created by large language models fine-tuned on a small clinical corpus successfully generated patient-like narratives across unseen tasks, received predicted MMSE…
Researchers have developed Dementia Language Models (DLMs), a versatile and controllable way to represent cognitive impairment through language. These models, created by fine-tuning large language models on a limited clinical corpus, can generate patient-like narratives across various tasks. When given unseen tasks, the DLMs produced narratives that received predicted MMSE scores in the impaired range, and neurologists identified them with accuracy comparable to real patient transcripts.
The internal representations of the DLMs, as well as their non-linguistic decision-making, also supported cognitive-state detection in new patient groups. Most importantly, the models showed controllability: moving the models from a healthy state towards a dementia-like state in weight space gradually worsened their language abilities and predicted MMSE scores, while also increasing the probability of dementia.
This controllable aspect of DLMs could prove beneficial in several areas, including clinician training, hypothesis generation, and scalable experimentation. The involvement of patients in these experiments would be limited to situations where it is truly necessary.
Written by urgent.news from bioRxiv's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
