MT Summit XVI JTF Workshop

Masao Utiyama is a research manager of the National Institute of Information and Communications Technology, Japan. His main research field is machine translation. He leads the development of "Minna No Jido Hon'yaku @ TexTra"

Mike Dillinger, PhD, is manager of taxonomies and MT at LinkedIn. Before that, he was manager and computational linguist at eBay, an independent consultant for Fortune 500 companies, and director of linguistics at both Spoken Translation and Global Words. He has two MT-related patents.

Manuel Herranz, CEO of Pangeanic, established Pangeanic in 2005. Collaboration with Valencia’s Polytechnic research group and the Computer Science Institute led to the creation of the PangeaMT platform, becoming the first LSP in the world to implement open source Moses successfully in a commercial environment.

(you can read a longer bio here)

Abstract

Neural networks need samples to learn the patterns that make an MT system, a picture catalogue system or even a summarization and chat system. Amazingly, the volume of data is typically less than the data required for a statistical machine translation system since it is the network that works out the relations and structure, thus obtaining more "fluency".

Neural networks produce human-like quality in practically all language pairs, and the outperform SMT in distant language pairs for Japanese/Arabic/Russian/Turkish <-> English as recent studies demonstrate.

However, there are known problems with NMT, such as terminology control, handling tags and fast re-training, which were solved in SMT a long time ago.

My presentation will deal with how NMT is likely to progress to incorporate online learning and achieve even higher levels of customization, such as the work Pangeanic will conduct for the EU in the iADAATPA project.

Dimitar Shterionov, head MT researcher at KantanMT, leads KantanLabs, a research group committed to advancing language technology. Shterionov holds a PhD in computer science from KU Leuven Belgium. He has worked on design and development of artificial intelligence software for learning and reasoning with uncertain data.

(you can read a longer bio here)

Abstract

With increasing translation demands due to rapid globalisation, Machine Translation (MT) has become a de-facto tool for a productive translation industry. With the emergence of Neural Machine Translation (NMT) in mid 2016, came promises for even better MT quality and for increased productivity. However, getting the best out of NMT requires a lot of work. One of the main challenges faced by those working with NMT engines is controllability.

Efficiently tackling this challenge is a key to successfully integrating NMT in a real-life translation production line. In this talk three ways to control the translation flow of NMT engines will be presented: data pre- and post- processing mechanisms; incremental training and engine adaptation; and enforcement of non-translatable assets.

This presentation will look at the effect of these procedures on the NMT output, their pros and cons, and will also discuss use-cases to show how an NMT engine can be customised towards specific translation requirements. All these controls will be discussed within the scope of an industry established MT platform – KantanMT.

Presenting these control mechanisms aims to show the necessary human-machine symbiosis that leads to an efficient NMT engine and therefore, to a successful productive translation workflow.

After several years working in the space development field switched his focus to localization and translation, running his own company from 1991. His company PostEdit.Tokyo provides consulting to translation firms looking to introduce MT. Since 2011 has served as the editor of the JTF Journal, the monthly magazine of the Japan Translation Federation.

(you can read a longer bio here)

Abstract

There are three important elements to benefit from the introduction of machine translation in the field of industrial translation: to bring out active cooperation of translators, to apply to fields suitable for machine translation, and to apply the quality assurance process corresponding to post edit.

機械翻訳の導入によって産業翻訳の分野で利益を得るためには、以下の三点が重要であることを示します：翻訳者の積極的な協力を引き出すこと、機械翻訳に適した分野に適用すること、そしてポストエディットに対応する品質保証プロセスを適用すること。

Gen Sato has been in Localization from 1999. He started as a translator and then experienced various roles including a reviewer, a vendor manager, and a language team manager. In 2009, he joined SDL Japan as a sales reps of translation software. He became a sales manager in 2016 and since then he has been proactively involved in Localization industry events to share his views toward its future.

(you can read a longer bio here)

Abstract

Although expectation for machine translation is very high, it is unrealistic to think machine translation can replace human translation completely. In this presentation, I will define pros and cons of machine translation and examine what would be the best scenario for both clients and translation industry.

機械翻訳への期待が高まっていますが、これまで翻訳業界が行ってきた作業を全て機械翻訳に置き換えるということはできません。この発表では機械翻訳が得意なことと不得意なことを定義して、翻訳の発注側と受注側の両方にとって機械翻訳を効率的に使用する方法を考察します。