Reconstructing impaired language using generative AI for people with aphasia.

In our GLM analysis, we found that phonological substitutions (both real-word and non‐word), semantic paraphasias (related and unrelated substitutions), and neologisms (with known and unknown targets) each exerted a highly significant negative impact on reconstruction accuracy (all Wald χ 2 > 9.8, p ≤ .007).

also in 132,142 other papers