Semantic working memory supports phrasal planning: Evidence from acute stroke.

This study examined the contribution of semantic and phonological working memory (WM) to lexical-semantic planning during language production for individuals tested acutely after left-hemisphere stroke. Participants completed tasks assessing phrase and sentence production, single-word retrieval, semantic and phonological WM, and single-word input processing. Using a constrained production paradigm that manipulated structure (phrase vs. sentence) and number of adjectival modifiers (one vs. two), we found that semantic WM capacity, indexed by performance on a category probe task, predicted accuracy in producing phrases containing multiple content words, even after controlling for single-word processing abilities. In contrast, phonological WM capacity, indexed by performance on a digit matching span task, did not relate to production accuracy. The relation to semantic WM was specific to the condition with the largest number of content words within a phrase (two-adjective noun phrase). These findings support the contention that lexical-semantic planning occurs phrase by phrase and is constrained by the capacity of a semantic maintenance buffer that supports activation and maintenance of multiple lexical-semantic representations prior to articulation. In contrast, phonological planning appears to operate within a narrower window that makes few demands on phonological WM. The results converge with prior neuropsychological evidence linking semantic WM to phrase-level planning and further imply that the phrase serves as the default scope of lexical-semantic encoding in production. Implications for models of WM and language production are discussed.