HiTZ/ALIA_DocMT_Evaluation
收藏资源简介:
ALIA_DocMT_Evaluation 是一个具有挑战性的文档级机器翻译(Doc-MT)基准数据集,包含高质量的人工翻译参考数据。请注意,对于BOPV和Parliament领域,片段是从真实源数据的测试分割中采样,目前正在手动翻译成英语。数据集涵盖多个领域:Berria(来自巴斯克报纸Berria的新闻领域)、Parliament(来自巴斯克议会的政治和法律对话领域)、BOPV(来自巴斯克自治区官方公报的行政和法律领域)、Instructions(用于任务评估的多样化指令提示,如问答、摘要等)、EUSparallel(从euscrawl语料库过滤的通用网络领域)、Aldizkariak(来自巴斯克期刊的科学和学术领域)以及Podcasts(从巴斯克播客转录中提取的非正式对话领域)。该数据集旨在用于评估文档级机器翻译模型和多语言NLP研究。
ALIA_DocMT_Evaluation is a challenging document-level machine translation (Doc-MT) benchmark featuring high-quality, human-translated reference data. Please note that for the BOPV and Parliament domains, the segments are sampled from the test splits of the real source data and are currently in the process of being manually translated into English. The dataset covers multiple domains: Berria (news domain extracted from the Basque newspaper Berria), Parliament (political and legal dialogue domain from the Basque Parliament), BOPV (administrative and legal domain from the Official Gazette of the Basque Country), Instructions (diverse instructional prompts, e.g., Q&A, summarization, for task evaluation), EUSparallel (general web domain filtered from the euscrawl corpus), Aldizkariak (scientific and academic domain sourced from Basque journals), and Podcasts (informal conversational domain extracted from Basque podcast transcripts). It is intended for evaluation of document-level machine translation models and multilingual NLP research.




