MTrill project: Machine Translation impact on language learning

Incorporating External Annotation to improve Named Entity Translation in NMT Maciej Modrzejewski Thanh-Le Ha Alexander Waibel Institute for Anthropomatics and Robotics KIT - Karlsruhe Institute of Technology, Germany maciej.modrzejewski@student.kit.edu firstname.lastname@kit.edu Miriam Exel Bianka Buschbeck SAP SE, Walldorf, Germany firstname.lastname@sap.com Abstract The correct translation of named entities (NEs) still poses a challenge for conven- tional neural machine translation (NMT) systems. This study explores methods incorporating named entity recognition (NER) into NMT with the aim to improve named entity translation. It proposes an annotation method that integrates named entities and inside–outside–beginning (IOB) tagging into the neural network input with the use of source factors. Our experiments on English→German and English→ Chinese show that just by including different NE classes and IOB tagging, we can increase the BLEU score by around 1 point using the standard test set from WMT2019 and achieve up to 12% increase in NE translation rates over a strong baseline. 1 Introduction The translation of named entities (NE) is challeng- ing because new phrases appear on a daily basis and many named entities are domain specific, not to be found in bilingual dictionaries. Improving named entity translation is important to transla- tion systems and cross-language information re- trieval applications (Jiang et al., 2007). Conven- tional neural machine translation (NMT) systems are expected to translate NEs by learning complex linguistic aspects and ambiguous terms from the training corpus only. When faced with named en- tities, they are found to be occasionally distorting c⃝ 2020 The authors. This article is licensed under a Creative Commons 3.0 licence, no derivative works, attribution, CC- BY-ND. location, organization or person names and even sometimes ignoring low-frequency proper names altogether (Koehn and Knowles, 2017). This paper explores methods incorporating named entity recognition (NER) into NMT with the aim to improve NE translation. NER systems are often adopted as an early annotation step in many Natural Language Processing (NLP) pipelines for applications such as question an- swering and information retrieval. This work explores an annotation method that integrates named entities and inside–outside–beginning (IOB) (Ramshaw and Marcus, 1999) tagging into the neural network input with the use of source factors. In our experiments, we focus on three NE classes: organization, location and person, and use the state-of-the-art encoder-decoder Transformer network. We also investigate how the granularity of NE class labels influences NE translation quality and conclude that specific labels contribute to the NE translation improvement. Further, we execute an extensive evaluation of the MT output assessing the influence of our annotation method on NE translation. Our experiments on English→German and English→Chinese show that by just including different NE classes and IOB tagging, we can increase the BLEU score by around 1 point using the standard test set from WMT2019 and achieve up to 12% increase in NE translation rates over a strong baseline. 2 Related Work Several research groups propose translating named entities prior to the translation of the whole sen- tence by an external named entity translation model. Li et al., (2018a); Yan et al., (2018); …

Content truncated to 500 words