Distinct post-sentence neural patterns representing lexical items vs. sentence integration

While comprehension marks the starting point in daily communication, the process is only fulfilled when suitable responses or inferences are followed. Listeners retain sentence information after initial comprehension. Although comprehension during listening has been widely studied, comparatively little is understood about how and where the brain retains linguistic information beyond the end of a sentence (EOS). A key question is whether the brain retains not only a holistic, sentence-level representation but also independent traces of individual lexical items-and, if so, how and where these dissociable signals are encoded in the brain. By analyzing the high gamma envelope in electrocorticography (ECoG) data from 15 patients with epilepsy, we directly investigated how neural signals encode and retain information about individual lexical items as well as the integrated sentence representation after the EOS. To this end, we employed a question-and-answer paradigm in which participants heard one of four sentences ("Is it alive?", "Is it not alive?", "Is it a part of body?" or "Is it not a part of the body?"), followed by a response prompt. To answer correctly, subjects must retain the relevant linguistic information, so we could trace retained neural representations in post-question periods, that respond either to each lexical item independently-content ("alive" vs. "part of the body") and negation ("positive" vs. "negative")-or to sentence-specific representations integrating both lexical items. Label-based encoding models were fit to predict neural responses from each label, and encoding strength was quantified by the correlation between predicted and observed signals. We found that channels selectively encoding lexical information were distributed across widespread cortical areas. In contrast, sentence-specific encoding was highly localized in the left posterior superior temporal gyrus (pSTG). Furthermore, by applying the same encoding model to neural signals recorded during the subsequent response-preparation period, we found that both lexical-item and integrated sentence information can persist significantly while participants prepared their responses. These findings provide direct evidence for the distinct spatial organization of lexical and sentence-level representations in the human brain after the end of a sentence.