PheMT

PheMT是由东北大学和RIKEN联合创建的数据集，专注于评估机器翻译系统在处理用户生成内容中的特定语言现象时的鲁棒性。该数据集包含1566条日英双语句对，覆盖了四种常见的语言现象：专有名词、缩略名词、口语表达和变体。创建过程中，研究人员通过众包方式对数据进行了标注和规范化处理，确保了数据的质量。PheMT数据集的应用领域主要集中在机器翻译系统的性能评估和改进，特别是在处理含有噪声的输入文本时，旨在解决现有机器翻译系统在处理非标准语言表达时的不足。

PheMT is a dataset jointly created by Northeastern University and RIKEN, focusing on evaluating the robustness of machine translation systems when handling specific linguistic phenomena in user-generated content. This dataset contains 1566 Japanese-English sentence pairs, covering four common linguistic phenomena: proper nouns, abbreviated nouns, colloquial expressions, and variants. During its development, researchers conducted annotation and standardization processing on the data via crowdsourcing to ensure data quality. The PheMT dataset is primarily applied to the performance evaluation and enhancement of machine translation systems, particularly when dealing with noisy input texts, aiming to address the limitations of existing machine translation systems in handling non-standard linguistic expressions.