HebArabNlpProject/shoshan-data
收藏资源简介:
Shoshan希伯来语词形还原数据集是一个用于自然语言处理任务的数据集,专门针对希伯来语的词形还原(lemmatization)和形态学分析。它为每个表面标记(token)在上下文中提供了一个内容词元(lemma),用于训练和评估Shoshan词形还原器。数据集包含多个文件:train.csv、dev.csv和test.csv用于域内数据(来自Knesset和Wikipedia,基于IAHLT UD树库),共约191k、11k和11k行;ood.csv等文件用于域外基准测试,约5k行;oov.csv用于处理未见词元,共100行。数据列包括form(形式)、lemma(词元)、pos(词性)、sentence(句子)、source(来源)和sent_id(句子ID)。数据集基于CC BY 4.0许可证,来源于IAHLT希伯来语UD树库和公共希伯来语词典。
The Shoshan Hebrew Lemmatization Dataset is a dedicated natural language processing dataset focused on Hebrew lemmatization and morphological analysis. It provides a corresponding content lemma for each surface Token in contextual scenarios, which is designed for training and evaluating the Shoshan lemmatizer. The dataset includes multiple files: train.csv, dev.csv and test.csv for in-domain data (sourced from Knesset materials and Wikipedia, based on the IAHLT UD Treebank), with approximately 191k, 11k and 11k rows respectively; ood.csv and other related files for out-of-domain benchmarking, containing around 5k rows; and oov.csv for handling unseen lemmas, which has 100 rows in total. The dataset columns consist of form, lemma, pos (part-of-speech), sentence, source and sent_id (sentence ID). This dataset is licensed under CC BY 4.0, and its data originates from the IAHLT Hebrew UD Treebank and public Hebrew lexical dictionaries.




