fc84ad56f9f547eb89c72b9bac209312-AuthorFeedback.pdf

WethankR2andR3for acknowledging that our experiments show3 interesting results that open up questions for future research and spur potential modeling innovations. Moreover, recent work such as LAMA, T5 and GPT3 suggest that language models can be treated as knowledge23 bases. We acknowledge that generalization is weaker in this harder setting, and we confirmed that by performing28 additional experiments on the natural language split.