Cognition, Behavior, and Memory
Author: Jeremias Inchauspe | Email: inchauspe.jeremias@gmail.com
Jeremias Inchauspe1°, Camilo Avendaño1°2°, Lucas Sterpin1°, Franco Ferrante1°, Gonzalo Pérez1°, Edinson Muñoz2°, Pedro Chaná-Cuevas2°, Yamile Bocanegra3°, Lucía Amoruso1°, Adolfo García1°2°4°
1° Cognitive Neuroscience Center, Universidad de San Andrés, Buenos Aires, Argentina
2° Universidad de Santiago de Chile, Santiago, Chile
3° Grupo de Neurociencias de Antioquia, Facultad de Medicina, Universidad de Antioquia, Medellín, Colombia; Grupo Neuropsicología y Conducta (GRUNECO), Facultad de Medicina, Universidad de Antioquia, Medellín, Colombia.
4° Global Brain Health Institute, University of California, San Francisco, USA
Frontostriatal atrophy in Parkinson’s disease (PD) impairs episodic memory (EM) skills. Though critical to predict further cognitive and functional decline, EM assessments are prone to examiner bias and prioritize retrieval of predefined items rather than graded proximity between stimuli and responses. Here we overcome these limitations through a novel NLP-based approach. Seventy-four participants (34 with PD, 39 healthy controls) retold two texts emphasizing either bodily actions or internal states. We ran the original texts and each participant’s retellings through NLP algorithms to calculate (i) the retelling’s verbosity (word count); (ii) semantic distance, via the cosine similarity between the text-level embeddings of texts and retellings; and (iii) topological distance, via differences in text-level connectivity, repetitions, and global structural properties captured by speech graphs. Robust ANOVAs showed that patients said fewer words only in the bodily-action text while exhibiting larger semantic distance and topological distance across texts. These findings suggest that NLP metrics of EM can afford useful markers of PD, opening new pathways for clinical assessments.