Skip to content

Adversarial NLI (ANLI)

Natural Language Inference (NLI)English

Adversarial NLI (ANLI) is a Natural Language Inference (NLI)-focused dataset in English that provides 169,265 labeled examples distributed in JSON format.

About Adversarial NLI (ANLI)

Dataset is an NLI benchmark created via human-and-model-in-the-loop enabled training (HAMLET). Human was tasked to provide a hypothesis that fools the model into misclassifying the label.

Details

Task
Natural Language Inference (NLI)
Language
English
Format
JSON
Rows / instances
169,265
Creator
Nie et al.
Year
2020
Download Paper

Related Natural Language Inference (NLI) datasets

FAQ