Randomised Sentences and Word-Embeddings Generated During the National Corpus of Irish Project

While collecting high-quality language data for the National Corpus of Irish project between 2022 and 2024 it became clear to the Gaois research group that there was a broader need for data in Irish-language NLP tasks and applied linguistic research.

This paper details specifications of the shareable by-products that were generated from some or all of these data; a 76,578,449-word dataset of randomised sentences and two different types of 100-dimension word embeddings that were generated from the 150-million words of high-quality Irish that was collected during the Corpas Náisiúnta na Gaeilge project. All datasets are shared as zipped TXT files.

This paper also explains the considerations for creating resources of this sort, some of which pertain to minority languages and smaller language communities, others to languages that have fewer language technologies.