Abstract :
[en] Working memory (WM), the ability to hold and manipulate information over a short period of time, is not an isolated cognitive function but appears to be influenced by verbal knowledge stored in the language system. This is indicated by the presence of several psycholinguistic effects in WM, with verbal items associated with richer lexico-semantic representations leading to higher recall performance in WM tasks, such as words vs. nonwords (lexicality effect), high vs. low frequency words (word frequency effect), related vs. unrelated words (semantic relatedness effect) and concrete vs. abstract words (concreteness effect). The present work aims to achieve a better understanding of the nature and origin of the interactions between WM and language knowledge. In Study 1, we assessed the degree of automaticity of these interactions. We showed that most psycholinguistic effects were observed when manipulated in a running span procedure preventing participants from implementing strategic processes during encoding, except for the word concreteness dimension. In Study 2, we assessed a strong version of the redintegration hypothesis of WM, which states that psycholinguistic effects stem from reconstruction processes occurring at the moment of recall. Contrary to the model’s predictions, we observed that the lexicality effect was still preserved when assessed in a combined running span – probe recognition procedure that prevented slow and controlled reconstruction processes. In Study 3, we assessed the role of linguistic cortices in the maintenance of verbal memoranda using a neuroimaging and multivariate neural decoding approach. We observed that neural patterns in linguistic cortices elicited during a WM task encode the type of stimuli (words vs. nonwords) being maintained in WM. In Study 4, we assessed a psycholinguistic effect that has not yet been explored in the WM domain, the lexical cohort effect. We observed that the lexical cohort effect, involving differences in speed of lexical access rather than in richness of linguistic representations, did not impact WM performance. In Study 5, we observed that semantic knowledge can protect memoranda against the deleterious effect of interference, but only when the semantic characteristics allowed participants to regroup items via their shared representation. Finally, Study 6 showed that semantic knowledge not only enhances the number of items that participants are able to recall, but also influences serial order recall performance. Our results show that language knowledge impacts verbal working memory in a fast and automatic manner, supports WM already during the maintenance stage and influences different aspects of WM such as the processing of item information, the processing of serial order information and resistance to interference. These findings provide support for theoretical models grounding WM in the linguistic system. The theoretical plausibility of linguistic accounts of WM is further demonstrated through a series of formal computational simulations.
