Normative clinical language data and task specific effects

Language production in clinical populations varies not only by neurological condition but also by the type of discourse task used during assessment. This study introduces a detailed analysis of connected speech production tasks—Narrative, Procedural, and Picture-based—conducted in etiologically heterogeneous patient groups with neurological damage, namely, 1,394 individuals with Left and Right Hemisphere Damage (LHD, RHD), Traumatic Brain Injury (TBI), Mild Cognitive Impairment (MCI), dementia, and healthy controls. Drawing on over 290 linguistic features spanning phonology, morphology, syntax, semantics, lexicon, and readability, we conducted a large-scale analysis of spoken texts using mixed-effects models to isolate the effects of diagnosis, task, and their interaction. Results revealed that task type exerts a pervasive influence on linguistic output, significantly interacting with diagnostic group across nearly all linguistic domains. Narrative tasks elicited more complex syntax and aspectual morphology, while procedural and descriptive tasks prompted simpler grammatical structures and greater lexical diversity, respectively. To explore high-dimensional linguistic patterns visually, we applied Uniform Manifold Approximation and Projection (UMAP), which revealed clear clustering of observations according to task type. The spatial arrangement of clusters reflected a continuum of contextual constraint: from highly structured picture description tasks (e.g., Cookie Theft), through procedural instruction tasks (e.g., sandwich-making), to conversational speech and open-ended narratives (e.g., accounts of brain injury, recovery, or illness). This gradient likely corresponds to increasing cognitive–linguistic demands, with unstructured tasks requiring greater topic generation, discourse planning, and memory retrieval. Crucially, the interaction between task and diagnosis modulates how underlying impairments manifest, indicating that linguistic deficits are not uniformly expressed across tasks. This challenges the validity of task-agnostic assessment and analysis practices, as well as normative comparisons. We propose a task-specific framework for clinical language assessment, enabling more precise identification of linguistic impairments and informing targeted therapeutic interventions. Our findings underscore the necessity of accounting for task effects in both clinical evaluation and the development of computational tools for neurogenic language disorders.