Neurodegenerative dementia syndromes, such as Primary Progressive Aphasias (PPA), have traditionally been diagnosed based in part on verbal and nonverbal cognitive profiles. Debate continues about whether PPA is best subdivided into three variants and also regarding the most distinctive linguistic features for classifying PPA variants. In this study, we harnessed the capabilities of artificial intelligence (AI) and natural language processing (NLP) to first perform unsupervised classification of concise, connected speech samples from 78 PPA patients. Large Language Models discerned three distinct PPA clusters, with 88.5% agreement with independent clinical diagnoses. Patterns of cortical atrophy of three data-driven clusters corresponded to the localization in the clinical diagnostic criteria. We then used NLP to identify linguistic features that best dissociate the three PPA variants. Seventeen features emerged as most valuable for this purpose, including the observation that separating verbs into high and low-frequency types significantly improves classification accuracy. Using these linguistic features derived from the analysis of brief connected speech samples, we developed a classifier that achieved 97.9% accuracy in predicting PPA subtypes and healthy controls. Our findings provide pivotal insights for refining early-stage dementia diagnosis, deepening our understanding of the characteristics of these neurodegenerative phenotypes and the neurobiology of language processing, and enhancing diagnostic evaluation accuracy.