Probabilistic decoding of EEG potentials evoked by word associations.
The N400 potential has been instrumental to our understanding of how language is encoded in the brain. Recorded through electroencephalography (EEG), it is most strongly evoked in response to the sequential presentation of two anomalous words, but it is absent for strongly associated ones. The spectrum of association strengths of word pairs in relation to the evoked N400 effect has not yet been modeled but, when successful, it could provide evidence for the structure of semantic memory. The aim of this project is to develop for the first time an algorithm that automatically detects the N400 and that generates an estimate for the association strength between the words that has generated it. The theory behind this algorithm also forms the basis for a second algorithm that will arrange a given set of N400 word pair responses into a graphical model called semantic network. We will use carefully composed sets of word pairs, with increasing association strength, to train and assess our detection algorithm, and to compare the semantic network estimated by the second algorithm with the available one derived from behavioral measurements (e.g. rating scales). The challenge of the second algorithm is to reconcile association strength estimation based on the N400 with semantic network inference issues.