task_categories:
- summarization
language:
- tr
tags:
- summarization
- xsum
The XSum dataset was designed for the English language and consists of news articles, their corresponding single-sentence summaries, and associated BBC article IDs. Due to the lack of high-quality summarization datasets in Turkish, a machine translation approach was employed to adapt the dataset for Turkish. The Meta MT model1, which was created by Meta AI, was used to translate the original English texts into Turkish. This model was chosen because it was trained on a diverse set of languages, including Turkish, making it suitable for our MT task. After the MT process, the resulting dataset, referred to as XSum-TR, contains 204,045 training samples and 11,334 test samples, making it one of the most extensive Turkish summarization datasets available.