M3T
收藏资源简介:
M3T是由AWS AI Labs创建的一个新型多模态文档级机器翻译基准数据集,专注于评估NMT系统在翻译半结构化文档时的能力。该数据集涵盖了广泛的文档类型和语言对,特别强调了视觉布局信息的重要性。创建过程中,数据集从多个公开数据源中抽样,并进行了详细的布局和阅读顺序标注。M3T的应用领域主要集中在提高机器翻译系统在处理真实世界复杂文档布局时的性能,旨在解决现有系统在视觉信息处理上的不足。
M3T is a novel multimodal document-level machine translation benchmark dataset developed by AWS AI Labs, which focuses on evaluating the capabilities of Neural Machine Translation (NMT) systems when translating semi-structured documents. This dataset covers a wide range of document types and language pairs, with particular emphasis on the importance of visual layout information. During its creation, the dataset was sampled from multiple publicly available data sources and annotated with detailed layout and reading order information. The core application scenario of M3T is to enhance the performance of machine translation systems when handling complex real-world document layouts, aiming to address the limitations of existing systems in processing visual information.

- 1M3T: A New Benchmark Dataset for Multi-Modal Document-Level Machine TranslationAWS AI Labs · 2024年



