Finally, the predictive brain aspect of the AxS model adds another possible answer to the question ofwhysome neuroimaging studies might not reveal activity in ‘‘the motor system” during speech perception. Specifically, if the involvement of speech pro- duction systems in perception is predictive, it implies that involve- ment occurs at a time that precedes onsets that might be used in statistical models for analysis. For example, regression models used in nearly all fMRI experiments assume a ‘‘canonical” hemodynamic response function that is convolved with stimulus onset and dura- tion that has a fixed shape and lag of two seconds. However, if pre- diction is a key aspect of processing speech, these lags and shapes should readily vary as a function of context. This could have the effect of obscuring speech production related activity given that it would a priori occur earlier in time and thus not be well modelled.