QUAK

韩英合成机器翻译质量预测数据 (韩英神经机器翻译的一个合成质量估计数据集，QUAK) 是指韩语句子和英语机器翻译句子，并且每个句子的机器翻译结果的质量是OK/这是一个机器翻译质量预测模型训练数据集，它预测与坏标签。由于用相应数据学习的机器翻译质量预测模型即使没有正确的句子也能预测翻译质量，因此它可以应用于没有正确句子的现实问题。OK/BAD tag对每个单词都有标记，以指示机器翻译结果中的哪些标记 (单词) 被错误翻译，对齐的源标记也被标记，因此即使不知道翻译语言的用户也可以知道哪些单词被错误翻译。有。

The Korean-English Synthetic Machine Translation Quality Prediction Dataset (QUAK, a synthetic quality estimation dataset for Korean-English neural machine translation) is a training dataset for machine translation quality prediction models. It contains Korean source sentences and their corresponding English machine translation outputs, with the overall quality of each translation annotated with OK/BAD labels. Machine translation quality prediction models trained on this dataset can predict translation quality even without access to the correct reference translation, making it applicable to real-world scenarios where no reference sentences are available. Additionally, the OK/BAD tags are assigned at the token (word) level to indicate which tokens in the machine translation output are incorrectly translated, and aligned source tokens are also marked. Therefore, even users who are unfamiliar with the target language (English) can determine which words were mistranslated.

搜集汇总

背景与挑战

背景概述

QUAK是一个韩英神经机器翻译的合成质量估计数据集，包含韩语句子和对应的英语机器翻译，每个句子标注了OK/BAD质量标签，并标记了错误翻译的单词及对齐的源标记。该数据集由Upstage和高丽大学于2022年发布，旨在训练机器翻译质量预测模型，适用于无正确句子的实际应用场景。