Objective: This study aimed to investigate dynamics in the semantic memory searching process of Parkinson’s disease patients.
Background: Linguistic studies in Parkinson’s disease highlighted the role of specific dynamics in semantic memory that could lead to identifying differences between healthy controls and other neuropathologies. Artificial intelligence methodologies have been a fundamental tool for extracting key features, particularly the use of the Large Language Model (LLM) to do complex semantic analysis. These advanced methods allow access to semantic trajectories that distinguish and aid in early diagnosis and understanding of the disease’s cognitive profile.
Method: We recruit the data from patients with Parkinson’s disease (PD) and Healthy subjects (HS). Participants had to perform action verbal fluency tasks in 3 domains: hand actions, leg actions, verbs, and an object verbal fluency task as a control. Our primary approach uses an LLM (transformer-based natural language processing) method to explore differences in the dynamics of navigation through trajectories in the semantic space (i.e., embedding space). This way, we computed distances to centroids using multidimensional scaling (MDS) for words extracted separately from each task to assess clusterization. Additionally, we analyzed trajectory metrics, incorporating word ordering during search, to evaluate the temporal organization of semantic properties.
Results: PD patients showed more distance to the centroid than HS in object fluency, a longer and more noisy search trajectory in object fluency, and fluency in hand actions, showing differentiation between both groups.
Conclusion: Our findings indicate that Parkinson’s disease patients exhibit altered semantic memory search dynamics compared to healthy controls. Specifically, they demonstrate greater distance from semantic centroids and more erratic search trajectories. These results highlight the potential of LLM-based semantic trajectory analysis as a valuable tool for identifying cognitive changes in Parkinson’s disease.
To cite this abstract in AMA style:
F. Toro Hernández, R. Cabral-Carvalho, N. Mendes Pellegrino, G. Paris-Colombo, A. Bontempo, A. Sena, H. Salmazo-Silva, M. Carthery-Goulart. Using large language models to assess dynamics in semantic memory search of patients with Parkinson’s disease [abstract]. Mov Disord. 2025; 40 (suppl 1). https://www.mdsabstracts.org/abstract/using-large-language-models-to-assess-dynamics-in-semantic-memory-search-of-patients-with-parkinsons-disease/. Accessed October 5, 2025.« Back to 2025 International Congress
MDS Abstracts - https://www.mdsabstracts.org/abstract/using-large-language-models-to-assess-dynamics-in-semantic-memory-search-of-patients-with-parkinsons-disease/