'Constituent length' effects in fMRI do not provide evidence for abstract syntactic processing