Normative clinical language data and task specific effects

Language production in clinical populations varies by both neurological condition and discourse task. This study analyzes connected speech from 1394 individuals with left or right hemisphere damage, Traumatic Brain Injury, Mild Cognitive Impairment, dementia, or no neurological diagnosis. Participants completed narrative, procedural, and picture-based tasks. Using Open Brain AI to extract linguistic features across phonology, morphology, syntax, semantics, lexicon, and readability, we applied mixed-effects models to examine effects of diagnosis, task, and their interaction. Results showed that task type strongly shapes linguistic output and interacts with diagnostic group across nearly all domains. Narrative tasks elicited more complex syntax and aspectual morphology, while procedural tasks prompted simpler grammatical structures and picture-based descriptions produced greater lexical diversity. Uniform Manifold Approximation and Projection further revealed clear clustering by task type, reflecting a continuum of contextual constraint ranging from structured picture descriptions and procedural instructions to free speech and open-ended narratives. This gradient illustrates a shift in linguistic scaffolding: highly structured tasks tightly constrain the target lexicon and syntax, whereas open-ended tasks require speakers to independently generate and organize narrative content. These findings show that clinical language tasks are not interchangeable. Treating them as equivalent may reduce diagnostic accuracy and weaken comparisons with healthy norms. We propose an Open Brain AI task-specific framework for clinical language assessment to improve identification of linguistic impairments and guide targeted intervention.