This is a good point and the level of so-called task specific "inductive bias" in models is an active point of discussion, but I don't think it is fair to add all of our evolution to the model inductive bias because most of evolution was not towards giving better understanding of language to the model, it was towards better understanding of language in humans.