Neural Machine Translation for User-Generated Content

用户生成内容的神经机器翻译

• 批准号: 20K19879
• 负责人: MARIE BENJAMIN
• 金额: $ 2.66万
• 依托单位: National Institute of Information and Communications Technology
• 依托单位国家: 日本
• 项目类别: Grant-in-Aid for Early-Career Scientists
• 财政年份: 2020
• 资助国家: 日本
• 起止时间: 2020-04-01 至 2022-03-31
• 项目状态: 已结题
• 来源: https://kaken.nii.ac.jp/en/grant/KAKENHI-PROJECT-20K19879/
• 关键词: machine translation; COVID-19; user-generated content; language model; Asian languages; user-generated text; deep learning; unsupervised learning; social media

The main achievement for the second year of this research is a new method to extend monolingual data in a low-resource domain and style (e.g., tweets on the topic of COVID-19) to generate larger data for training NMT. For instance, given a a small set of Japanese tweets (e.g., 1000 tweets) about the COVID-19 crisis, that is too small to train NMT, this method artificially extends it to million tweets on the same topic and makes it useful to train better NMT to translate tweets. Using this artificial data to train NMT yields improvements of NMT that becomes better at translating texts even for domain and style for which very few data is available. Experiments have been successfully conducted in various domains and styles (medical, IT, news, tweets, online discussions), and languages (French, German, Japanese). This work has also been extended for "personalizing" NMT, i.e., adapt NMT so it translates texts written by a specific person while preserving the characteristics of writing of this person.

本研究第二年的主要成果是一种在低资源领域和风格（例如，关于COVID-19主题的推文）以生成用于训练NMT的更大数据。例如，给定一小部分日本推文（例如，1000条推文），这对于训练NMT来说太小了，这种方法人为地将其扩展到同一主题的数百万条推文，并使其有助于训练更好的NMT来翻译推文。 使用这种人工数据来训练NMT会产生NMT的改进，即使对于数据很少的领域和风格，NMT也能更好地翻译文本。实验已经成功地进行了各种领域和风格（医疗，IT，新闻，推文，在线讨论）和语言（法语，德语，日语）。这项工作也被扩展到“个性化”NMT，即，调整NMT，使其翻译由特定人撰写的文本，同时保留此人的写作特征。

项目成果

Altering Parallel Data into User-Generated Texts with Zero-Shot Neural Machine Translation

通过零样本神经机器翻译将并行数据更改为用户生成的文本

• 发表时间: 2021
• 作者: Marie Benjamin;Fujita Atsushi
• 通讯作者: Fujita Atsushi

Scientific Credibility of Machine Translation Research: A Meta-Evaluation of 769 Papers

• DOI: 10.18653/v1/2021.acl-long.566
• 发表时间: 2021-06
• 期刊: ArXiv
• 作者: Benjamin Marie;Atsushi Fujita;Raphaël Rubino

Extremely low-resource neural machine translation for Asian languages

• DOI: 10.1007/s10590-020-09258-6
• 发表时间: 2021-02-10
• 期刊: MACHINE TRANSLATION
• 影响因子: 1.9
• 作者: Rubino, Raphael;Marie, Benjamin;Sumita, Eiichiro
• 通讯作者: Sumita, Eiichiro

Tagged Back-translation Revisited: Why Does It Really Work?

• DOI: 10.18653/v1/2020.acl-main.532
• 发表时间: 2020-07
• 期刊: Current topics in membranes
• 作者: Benjamin Marie;Raphaël Rubino;Atsushi Fujita

Synthesizing Parallel Data of User-Generated Texts with Zero-Shot Neural Machine Translation

• DOI: 10.1162/tacl_a_00341
• 发表时间: 2020-11
• 期刊: Transactions of the Association for Computational Linguistics
• 影响因子: 10.9
• 作者: Benjamin Marie;Atsushi Fujita