Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Syntactic Network Analysis in Aphasia: Preliminary Findings from Narrative Discourse
Poster Session D, Saturday, September 13, 5:00 - 6:30 pm, Field House
This poster is part of the Sandbox Series.
Catherine Pham1, Peng Zhang1, Nichol Castro2, Jiyeon Lee1; 1Purdue University, 2University of Buffalo
Persons with aphasia (PWA) typically produce speech that is less syntactically complex than neurotypical speakers. Although traditional discourse metrics (e.g., mean length of utterance, clausal density, verb-to-noun ratio) capture some aspects of syntactic ability, they often overlook higher-order structural properties of language. Emerging work has highlighted the utility of network science approaches in quantifying lexical-semantic organization in aphasia, but its application to syntax remains largely unexplored. This project models syntactic organization in aphasia using tools from network science. We explore the extent to which network metrics capture syntactic differences between aphasia subgroups (fluent vs nonfluent PWA). We also examine how method of network construction (dependency versus co-occurrence) impacts the sensitivity of these metrics in capturing syntactic deficits. 104 PWA described the picnic image from the Western Aphasia Battery-Revised (WAB-R). All PWA were adult monolingual English speakers with no reported history of acquired neurological conditions, other than left hemisphere stroke, and were at least 3 months post onset of stroke at time of testing. PWA were categorized as fluent (N = 55) or nonfluent (N = 49) based on WAB-R fluency subscores (nonfluent PWA ≤5 and fluent PWA >5). Speech samples were transcribed and cleaned to exclude disfluencies (e.g., filled pauses, repetitions, self-corrections). For each group (fluent versus nonfluent PWA), we constructed two types of syntactic networks: dependency networks, where nodes represent lexical items and edges reflect syntactic dependency relations between items, and co-occurrence networks, where edges reflect linear adjacency between items. All networks were constructed using lemmatized forms to reduce redundancy. Preliminary results revealed that both co-occurrence and dependency network analyses captured hallmark reductions in lexical diversity (nodes) and syntactic connectivity (edges) in nonfluent PWA. Across network types, nonfluent PWA consistently exhibited higher modularity (co-occurrence: Mnonfluent = 0.151, Mfluent = 0.142; dependency: Mnonfluent = 0.209, Mfluent = 0.201) and average shortest path length (co-occurrence: Mnonfluent = 2.62, Mfluent = 2.58; dependency: Mnonfluent = 2.73, Mfluent = 2.71), indicating more segregated and less efficient syntactic structures. Fluent PWA demonstrated greater small-worldness (co-occurrence: Mfluent = 29.07, Mnonfluent = 21.60; dependency: Mfluent = 10.79, Mnonfluent = 10.46), reflecting better integration of local and global syntactic connections. Network type influenced sensitivity to group differences. Co-occurrence networks revealed distinctions in clustering (Mfluent = 0.46, Mnonfuent = 0.42) and small-worldness (Mfluent = 29.07, Mnonfluent = 21.60), suggesting greater sensitivity to discourse-level disruptions. Dependency networks also captured similar group differences, though with smaller effects, except for an increase in clustering among nonfluent PWA (Mnonfluent = 0.26, Mfluent = 0.24), potentially reflecting reliance on repetitive, formulaic structures. These findings suggest that co-occurrence networks are more sensitive to discourse-level disruptions, while dependency networks may highlight core syntactic simplifications common in nonfluent aphasia. Ongoing work is expanding these analyses to include a larger sample of PWA as well as incorporating disfluencies into the networks to explore how disruptions in fluency impact network structure. Additionally, future analyses plan to examine individual-level syntactic networks to probe the relationship between network metrics, traditional measures of discourse, and standardized aphasia language assessments.
Topic Areas: Disorders: Acquired, Language Production