Artificial Intelligence Enables a Paradigm Shift in Understanding Nonfluent Aphasia

Nonfluent aphasia is a neurological condition characterized by profound alterations in the quality and quantity of language production. Despite the succession of attempts to uncover the underlying mechanism of the disease, a comprehensive explanatory framework has been lacking. This theoretical impediment mainly stems from the absence of measuring tools sensitive to the key features of language. One pivotal aspect often overlooked in aphasia research is language informativeness. Crucially, the advent of Large Language Models powered by artificial intelligence has now made it possible to measure the previously unexplored facets of language. By leveraging these emergent capacities, we introduce the Information Restoration Paradigm as a novel, empirically testable framework that offers insights into the diverse symptoms of nonfluent aphasia. The framework posits that the word-level features of nonfluent speech are not defects but rather manifestations of a compensatory mechanism that aims to preserve the information content of language. Our approach underscores the role of informativeness as a cornerstone in synthesizing fragmented theories of the disease into a cohesive framework amenable to testing. Furthermore, we discuss the clinical implications of this paradigm for the description and treatment approaches to nonfluent aphasia.