Linguistic Generalizations are not Rules: Impacts on Evaluation of LMs