Two Words Are Faster Than One: Neural Dynamics of Accelerated Meaning in Phrases

Résumé fourni par la source

Although we increasingly understand how the brain processes phrase-level meaning, the role of semantic content remains underexplored. Recent models grounded in componential experiential features offer a principled framework for linking conceptual representation to the neurobiology of semantic composition. Here we used MEG to test whether experiential semantic features are retrieved and composed during two-word phrase comprehension, and whether lexical and phrase-level representations show temporally overlapping or strictly serial activation. Participants (16 women, 12 men) read two-word phrases and their component words in a rapid parallel visual presentation (RPVP) paradigm, equating the temporal dimension of the two types of stimuli. The phrases spanned six linguistic relations. All words and phrases were independently rated on 65 experiential features across 14 experiential domains and representational similarity analysis (RSA) was used to relate neural and semantic structure. Rather than emerging after or in parallel with single-word meanings, phrase-level meanings arose earlier, outpacing single words in isolation. Phrasal contexts also elicited earlier activation of word-level meaning: a word such as ‘cake’ showed semantic activation at ∼312 ms in isolation, but at ∼236 ms in ‘green cake.’ Experiential feature representations of both component words were detectable during phrase processing, with partially overlapping temporal windows. The magnitude of composition-related effects varied systematically across phrasal relation types and experiential domains. Together, these findings show that experiential semantic features are integrated during composition in ways that vary across phrasal relations and that combining words into phrases speeds semantic access, accelerating both phrase- and word-level meaning relative to isolated words. Significance Statement This study provides temporally resolved evidence that the brain simultaneously represents multiple levels of meaning during language comprehension. Using MEG and representational similarity analysis grounded in experiential semantic features, we show that compared to single words, two-word phrases elicit earlier and more sustained alignment between neural activity patterns and semantic feature representations. Further, the meanings of both constituent words remain detectable in overlapping time windows during phrase processing. Exploratory analyses further suggest that residual phrase-level semantic alignment varies with phrasal relations and semantic content. This work advances understanding of how distributed experiential knowledge is recruited during real-time language processing and how relational structure and semantic content jointly shape meaning composition.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.