Dan Jurafsky
Making natural language processing a teachable discipline — and using that perspective to read where the field's current evaluation practices fit, and don't fit, into its longer history.
The textbook Jurafsky co-wrote with James Martin, Speech and Language Processing, has been the entry point for graduate NLP students for over two decades — the third edition tracks how the field absorbed deep learning while keeping the linguistic and probabilistic foundations available to anyone who wants them. His broader work runs against the grain of modern NLP in a productive way: computational sociolinguistics, the language of food, courtroom dialogue analysis, the kinds of projects that treat language as a thing humans do in social settings rather than as a benchmark to be saturated. The wider lens gives his current commentary on LLM evaluation an unusual depth — he can locate any current methodological argument in the 50-year arc of language technology.