Using language models and latent semantic analysis to characterise the N400 m neural response

In this paper we build a computational model to predict N400 m response which is the neuromagnetic version of the N400. Stimuli consisted of 180 sentences presented to 22 listeners. There were 90 examples of "constraining context" sentences, and 90 examples of "non-constraining context" sentences. Besides the manually-annotated context predictor, we investigate 4 additional predictors based on Latent Semantic Analysis, a 4-gram language model, an incremental parser, and a novel pairwise- priming language model based on the IBM Model 1 translation model. Statistical analysis shows that all the predictors are significant. Moreover, we show that predictors based on the 4-gram language model and the pairwisepriming language model are highly correlated with the manual annotation of contextual plausibility, suggesting that these predictors are capable of playing the same role as the manual annotations in prediction of the N400 m response. We also show that the proposed predictors can be grouped into two clusters of significant predictors, suggesting that each cluster is capturing a different characteristic of the N400 m response.