NeurIPS 2020

Language Through a Prism: A Spectral Approach for Multiscale Language Representations


Meta Review

After reading each other's reviews and discussing the author response, reviewers lean towards acceptance of this submission. The primary weaknesses are related to the lack of analysis in the fine-tuning setting and a resulting uncertainty about the use case of this technique in a practical sense. On the positive side, the analysis itself is informative and introduces a new angle to the work on multi-scale representations. I encourage authors to consider reviewer feedback for future revisions -- especially those from R1 about the assumptions implicit in the work.