Recent Posts

Peer review in NLP: reject-if-not-SOTA

10 minute read

Many reviewers at major NLP conferences tend to reject models that fail to beat state-of-the-art. It is a heuristic that is simple, convenient, and wrong.