Differentiating nonfluent/agrammatic and logopenic primary progressive aphasia in Catalan-Spanish bilinguals by applying multilingual multimodal machine learning to connected speech

The study addresses the differentiation of two types of primary progressive aphasia (PPA) – nonfluent/agrammatic and logopenic variants – in 34 Catalan-Spanish bilingual patients. The research team employed an automated machine learning approach that analyzes speech timing parameters and linguistic features from recordings of patients describing a picture in both languages. The system was trained using four feature sets: speech timing measures, word-level parameters, linguistic features, and image-text congruence scores. The best-performing classifier achieved F1 scores of 93 percent in the non-dominant language and 92 percent in the dominant language. Classification performance did not significantly differ between the two languages. This automated approach takes only 1–2 minutes and represents a step toward addressing inequities in PPA diagnosis for bilingual patients outside English-speaking populations.