CONFER is a curated dataset of 18,000 sentence pairs in natural language inference (NLI) format, developed to investigate how language models handle presuppositions in specific types of conditional ...