Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Exploratory Speech Metrics for Detecting Agrammatism in Aphasia using Natural Language Processing
Poster Session D, Thursday, October 1, 4:30 - 6:30 pm, Wangari Maathai
Sladjana Lukic1, Aoken Chen1, Adolfo García2; 1Florida State University, 2Universidad de San Andrés
Non-fluent aphasia (NFA) is characterized by profound expressive language impairments driven largely by deficits in grammatical encoding (agrammatism). Although connected speech provides a valid window into these impairments, current clinically interpretable approaches rely on laborious, expert-based manual coding using theory-based profile-aligned perceptual systems such as the Auditory-Perceptual Rating of Connected Speech in Aphasia (APROCSA), limiting clinical feasibility, delaying results, and allowing for clerical errors. While computational tools offer operational advantages, current options lack clinically meaningful grammatical markers, reducing interpretability and translational value. In this study, we aim to validate interpretable grammatical features via NLP algorithms within the validated Toolkit to Examine Lifelike Language (TELL) v.2.0 platform. We quantified grammatical metrics based on dependency relations between words, including core, non-core, and nominal categories from continuous free speech (Typical Day) and picture description (Picnic Scene). Core relations captured essential sentence participants required by the verb (e.g., nominal subjects, direct objects), whereas non-core and nominal relations reflected optional modifiers and noun-phrase structure. These dependency categories, widely used in NLP frameworks such as Universal Dependencies, provided objective indices of grammatical structure deployment during connected speech. In parallel, we derived non-dependency (non-syntactic) metrics. Ratios for each category were calculated relative to the total proportion of dependency and non-dependency features. Preliminary data from 20 participants (10 healthy controls, 10 individuals with aphasia) demonstrated significant group differences: individuals with aphasia produced fewer non-core dependency relations (p=.013) and greater reliance on non-dependency/non-syntactic measures (p=.005) compared to healthy controls. The picnic scene speech sample showed a trend similar to that of the Typical Day free-speech task, but did not reach significance in this preliminary cohort. These findings provide preliminary evidence that combined dependency-based syntactic and non-dependency metrics may sensitively capture profile-aligned agrammatism in aphasia, often missed by clinical tests.
Topic Areas: Disorders: Acquired, Syntax and Combinatorial Semantics