mass-mean probing

Research shows neural language models encode grammaticality in internal representations

A new research paper explores whether neural language models (NLMs) can distinguish grammatical from ungrammatical sentences by examining their internal representations, rather than just probability assignments. Using a…