3AM
收藏资源简介:
3AM是一个包含26,000对英汉平行句子和相应图像的多模态机器翻译数据集,由澳门大学计算机与信息科学系NLP2CT实验室创建。该数据集设计用于包含更多的歧义和更广泛的标题与图像种类,通过使用词义消歧模型从视觉与语言数据集中选择歧义数据,形成更具挑战性的数据集。3AM旨在通过提供丰富的视觉信息,帮助模型更好地理解视觉内容,从而提高翻译质量。该数据集适用于多模态学习和机器翻译领域的研究,特别是解决视觉信息在翻译中的有效利用问题。
3AM is a multimodal machine translation dataset containing 26,000 pairs of English-Chinese parallel sentences and their corresponding images, created by the NLP2CT Laboratory, Department of Computer and Information Science, University of Macau. This dataset is designed to include more ambiguities and a broader range of captions and image types. It selects ambiguous samples from visual-language datasets via word sense disambiguation models to build a more challenging benchmark dataset. The goal of 3AM is to assist models in better comprehending visual content by supplying abundant visual information, thus enhancing translation quality. This dataset is applicable to research in the domains of multimodal learning and machine translation, especially for solving the problem of effective utilization of visual information in translation.




