What functional brain imaging reveals about the semantic representa- tion of object knowledge

We propose a generative classifier which models the hidden factors that underpin the neural representation of objects with a multivariate multiple linear regression model. Our results indicate that object features derived from a feature norming study or word cooccurrences in web corpus can capture some of these hidden factors and explain a significant portion of the systematic variance in the neural activity observed in an object-contemplation task. Furthermore, the resulting regression model is useful for decoding mental states, can generalize across people, and allows us to extrapolate the neural activity for unseen objects, which cannot be done with a discriminative classifier.