UA-CBT is a dataset inspired by Children's Book Test (
https://arxiv.org/abs/1511.02301) containing machine-generated (and human-corrected) stories with gaps, and multiple possible options for words to fill the gaps.
It's released as part of the Eval-UA-tion 1.0 Benchmark (paper:
https://aclanthology.org/2024.unlp-1.13/)
It differs from the original in the following ways:
The language is Ukrainian
The stories were LLM-generated, then… See the full description on the dataset page:
https://huggingface.co/datasets/shamotskyi/ua_cbt.
