A new benchmark shows that NLI models and LLMs struggle to infer presuppositions in conditionals, and fine-tuning on other presupposition datasets does not transfer to this structure.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Let's CONFER: A Dataset for Evaluating Natural Language Inference Models on CONditional InFERence and Presupposition
A new benchmark shows that NLI models and LLMs struggle to infer presuppositions in conditionals, and fine-tuning on other presupposition datasets does not transfer to this structure.