turkish-nlp-suite/TrGLUE

TrGLUE是第一个非翻译的土耳其语自然语言理解基准数据集，灵感来源于GLUE基准。它包括多个单句和句子对分类任务，如语言可接受性、情感分析、句子对语义等价性、语义相似性、问题对语义等价性、文本蕴含、问答/自然语言推理等。数据集是从土耳其语语料库中构建的，部分任务通过翻译和修订得到。数据集的大小从3.06K到369K不等，涵盖了多种自然语言处理任务。

TrGLUE is the first non-translate natural language understanding benchmark for Turkish, inspired by the original GLUE benchmark. It includes several single sentence and sentence pair classification tasks such as linguistic acceptability, sentiment analysis, semantic equivalence of sentence pairs, semantic similarity, semantic equivalence of question pairs, textual entailment, question answering/natural language inference, and more. The datasets are constructed from Turkish corpora, with some tasks derived through translation and revision. The dataset sizes range from 3.06K to 369K, covering a variety of natural language processing tasks.