RA-Wiki-UA is a benchmark-targeted Ukrainian image-text augmentation corpus
for Visual Word Sense Disambiguation. It was introduced in From Sparse to
Sense-Grounded: Wikipedia Training for Ukrainian Visual-WSD.
The dataset was created to reduce the visual and linguistic domain gap between
generic image-text training data and Ukrainian Visual-WSD. Rather than sampling
Wikipedia uniformly, its construction begins from the 381 gold… See the full description on the dataset page:
https://huggingface.co/datasets/yuriilaba/ra-wiki-ua.
