About

Language models now even appear to read between the lines, though sometimes wrongly. Our research asks whether this apparent understanding holds up: we test it, dissect the mechanisms behind it, and build models that understand rather than merely seem to.

RebelsNLU is an international research lab for Artificial Intelligence (AI) and Natural Language Processing (NLP) at the Japan Advanced Institute of Science and Technology (JAIST), founded in 2022 and led by Associate Professor Naoya Inoue. We started the lab to build machines that read between the lines, hence our name: Reading between the Lines for Natural Language Understanding. Within a year, however, large language models seemed to do just that, so our question shifted from how to build such machines to whether we can trust them.