REASONING LIKE ARISTOTLE: DIAGNOSING AND IMPROVING LANGUAGE MODELS FOR LOGICAL REASONING THROUGH SYLLOGISMS
- syllo.ipynb contians examples of generating syllogism questions and prompt GPT-3 for answering generated questions.
- train_comp.py & roberta_comp.py & roberta_n_sample.py contain code for testing RoBERTa with fine-tuning.
- train_folio.py & train_folio_base.py & zero_folio.py contain code for testing RoBERTa on FOLIO with/without pre-training.