Search Abstracts | Symposia | Slide Sessions | Poster Sessions
From neural signals to behavior: taking into account inter-trial variability in EEG
Poster B21 in Poster Session B, Friday, October 25, 10:00 - 11:30 am, Great Hall 4
Raphaël Fargier1, Emeline Manka2, Giulio Massari1, Fanny Meunier1, Patricia Reynaud-Bouret2; 1Université Côte d'Azur, CNRS, BCL, France, 2Université Côte d’Azur, CNRS, LJAD, France
The speed and accuracy with which speakers produce words are the results of successive mental operations that are achieved by functionally connected assemblies of neurons. Yet, our ability to identify these operations in brain recordings and characterize their spatial and temporal dynamics remains a challenge. Modeling and statistical tools, like segmentation/clustering approaches, can translate brain signals into a sequence of neural patterns or micro-states that, in turn, may be linked to specific cognitive/linguistic processes. This approach has been used, for example, to locate attentional and semantic interference effects on word production or reveal neurocognitive changes across the lifespan. However, the available algorithms cannot be used to reliably characterize the dynamics of processes when speed of word retrieval varies from trial to trial. They are traditionally applied on averaged signals or event-related potentials (ERPs), meaning that we use epochs of same lengths to study processes that are achieved with different timings. Averaging trials can therefore lead to spurious mixing of different neurocognitive dynamics, which affect our ability to test hypotheses about activation flow in models of lexical access (e.g. seriality/cascading of processes). There is thus a need to scale the dynamics of neural processes at the level of individual trials of different lengths in order to improve our ability to map neural activity to cognitive processes and to behavior. In Study 1, we tested a new segmentation/clustering algorithm (based on Pascual-Marqui’s model (1995)), built to deal with epochs of different lengths. Our “multi-trials” (with varying durations) model takes multiple trials as input and runs one unique segmentation, giving one unique set of microstates to describe all trials. We used simulations to validate our algorithm and compared it to both models applied to independent single trials and applied to averaged signals. Compared to the two other models, our “multi-trials” model made significantly less errors on 1) the microstate sequence, 2) topographies of microstates, and 3) their time-distribution. In Study 2, we apply our algorithm to a new EEG dataset of a picture naming experiment specifically designed to generate variations of reaction times (RTs). We manipulated the visibility of the pictures to be named by adding visual noise, with the goal of testing whether longer RTs would relate to longer processes distributed across all cognitive stages or strictly restricted to early processes such as visual/conceptual recognition. For that, two different levels of “salt-and-pepper” visual noises were applied to our pictures (N=200) to create two experimental conditions. The two stimuli lists were matched on a set of psycholinguistic variables known to affect accuracy and speed, and were counterbalanced across participants. Data are currently being collected and we plan to apply our “multi-trials” model on all EEG trials. We discuss the interest of this new approach to tackle current debates in the field of language production (e.g. seriality/cascading), and more generally to improve our understanding of brain-behavior relationships.
Topic Areas: Language Production, Methods