We obtained NLI data for Ukrainian from various sources: Ukrainian legal acts, fiction, news, and manually. The data can be used to tune and test Ukrainian NLI models closer to real-life scenarious!
Labels information: 0 - entailment, 1 - neutral, 2 - contradiction.
@inproceedings{dementieva-etal-2025-cross,
title = "Cross-lingual Text Classification Transfer: The Case of {U}krainian",
author = "Dementieva, Daryna and… See the full description on the dataset page:
https://huggingface.co/datasets/ukr-detect/ukr-nli-dataset-seminatural.
