Memsource
Integrated Machine and Human Translation
Junko Kato
Memsource offers a translation solution for everyone. Whether you translate yourself or choose to outsource your translations, Memsource can help you make the process more efficient. Some of the world’ s largest translation agencies and enterprise translation departments rely on Memsource to optimize their translation productivity.
Memsource does not provide machine translation of its own but offers access to 3rd-party machine translation engines via an API. Users can choose machine translation among 20 MT engines. Memsource API offers connection to common machine translation engines (such as Google Translate and Microsoft Translator) as well as custom-trained machine translation engines. |
Kilgray Translation Technologies Ltd.
Introducing memoQ server/QTerm
Yo Miura
memoQ server is a full-fledged enterprise translation management system (TMS) that comprises a repository for your translation assets, a control center for your translation projects and a hub for collaborative translation projects. It is the TMS of choice for enterprises and translation companies all over the world that use it on a daily basis to manage and automate localization processes that enable their global presence. You can also select the option within memoQ to connect to the neural MT engines of Google and Microsoft.
With QTerm, companies and organizations can turn their terminology into a corporate asset that facilitates internal and external communication, increases brand awareness, improves the quality of technical communication and cuts the costs of misunderstanding. QTerm is a memoQ server add-on that offers powerful terminology functionality. It is a browser-based application linked directly to memoQ server. Although memoQ translator pro and memoQ server do offer an integrated term base functionality, QTerm is a more powerful solution, and can be customized for specific needs.
At our booth, we are presenting how memoQ server and QTerm can optimize translation and localization workflows, and as a result, help you reach large-scale savings in large and complicated projects – the ones that generate the lion’s share of your revenue. |
NICT
Advanced Translation Research and Development Promotion Center National Institute of Information and Communications Technology, Japan
Xiaolin Wang, Masao Utiyama, Eiichiro Sumita
Simultaneous interpretation is one type of real-time interpretation where interpreters perform translation without interrupting speakers. It is a very effective way to bridge language gaps. Neural machine translation, as a new technology emerged from the field of deep learning, has improved the quality of automated machine translation into a significantly higher level compared to statistical machine translation.
Here we demonstrate a fully automatic simultaneous interpretation system which is powered by fast neural machine translation engines. Translation quality and low latency are the two major requirements for simultaneous interpretation services. Our new system, compared to our previous system which was powered by statistical machine translation engines, exhibits significant improvement on translation quality while suffers very little on latency.
The framework of our system, for the trade-off between performance versus extensible, adopts a simple pipeline of six components: an incremental speech recognition engine, a sentence segmenter, a punctuation predictor, a machine translation engine, and a speech synthesizer. The user interface of our system is fully automatic. It receives audio signals constantly from its microphone. If no one is speaking, it will produce no output. If someone is speaking, it will speak out the translation, normally, in only a few seconds. |
Toshiba Digital Solutions Corporation
Toshiba Human-AI Collaboration Translation Platform
Satoshi Sonoo
Toshiba released its first machine translation system “The Hon-yaku” for a desktop PC over twenty years ago. More recently, we have launched an AI service RECAIUSTM (< rikai = understanding + us = people) that supports various activities by understanding the intent and conditions of people in audio and visual recordings. RECAIUSTM provides speech recognition and translation PaaS/SaaS featuring preprocessing of speech recognition results coupled with a hybrid engine between RBMT and SMT for spoken language translation.
We have also devised the concept of a human-AI collaboration translation platform that reuses human translation resources (previous human translations, otherwise known as translation memory) by integrating our AI technologies employed in “The Hon-yaku” and RECAIUSTM. Thus, documents are automatically preprocessed to make them amenable to machine translation. Another feature of this platform is that it offers multiple translation engines, including RBMT(s), SMT(s), and NMT(s) and utilizes models acquired from human post-editing feedback for better translation. |
Fujitsu
Hands-free Speech Translation System for Effective Communication between Patients and Medical Staff
Tomoki Nagase
Fujitsu has developed a speech translation system that can be applied to medical fields which enable to be used without touching the terminal.
In the medical fields, there are two essential requirements for speech translation system, the high quality translation engine that covers various medical expressions, and the hands-free operation interface.
As for translation accuracy, we have developed medical translation corpora collected by clinical trials at 6 hospitals. By adding these corpora into the existing medical corpora developed by NICT, we realized the accuracy of more than 80%.
And we realized the hands-free operation interface by our unique techniques that consist of the speaker-detection technique based on the direction of utterance and the accurate speech segment recognition technique based on accumulation of speech technology we developed for many years. |
Translated Srl. / Fondazione Bruno Kessler / TAUS / The University of Edinburgh
ModernMT: An Open-Source Machine Translation Platform for the Translation Industry
Ulrich Germann
ModernMT is a 3-year innovation action (2015–2017) funded by the European Union’ s Horizon 2020 research and innovation programme. This project has brought together leaders in the translation industry and academics at the forefront of machine translation research to develop a powerful, scaleable machine translation platform that enables language service providers to set up and self-host state-of-the-art machine translation technology with minimum effort.
ModernMT offers the following benefits:
• An easy, turn-key installation procedure in a native Linux environment, or in a virtual machine.
• A REST API for easy integration into the translation workflow, e.g., as a plug-in in computer-aided translation tools.
• Automatic model building from parallel corpora or translation memories.
• Models can be updated with new data at any time with very low latency, close to real time.
• Instant automatic domain adaptation of the translation engines to the input that is being translated. |
The SUMMA Consortium
SUMMA: Scalable Understanding of Multilingual Media
Ulrich Germann
SUMMA (Scalable Understanding of Multilingual Media) is a 3-year research and innovation project to develop an open-source software platform for monitoring multiple news streams in various languages, from live broadcast TV and audio to on-demand streaming to textual data.
Three use cases drive the project: (1) monitoring hundreds of news channels worldwide by BBC Monitoring to support the BBC’ s journalists, and as a service to external customers; (2) in-house monitoring of broadcast news to enable Deutsche Welle to keep track of its world-wide broadcasting, which is produced in regional newsrooms; and (3) to support data journalism.
The SUMMA platform is designed to ingest hundreds of news streams at once. Spoken language is automatically transcribed and translated; named entities are recognised and marked in the text; and incoming news items are clustered by topic and organised into storylines. All information is stored in a database, and the platform offers multiple views of the data, including storylines, chronological listings, and trending topics. |
Chizai Corporation
Machine Translation Services
Chie Shibata
Chizai Corporation has established itself as the leading company in patent translation. Chizai Corporation provides English to Japanese and Chinese to Japanese translation services via machine translation and post-editing as cost-effective and time-saving solution, as well as Medical Device Dictionary “MeLDy” English-Japanese-Chinese powered by machine translation. |
Nara Institute of Science and Technology (NAIST)
Active Annotator for MT
Akiva Miura
In SMT/NMT, large quantities of high-quality bilingual data are essential to achieve high translation accuracy. While in many cases large corpora can be collected, for example by crawling the web, in many domains or language pairs it is still necessary to create data by hand, either by hiring professionals or crowdsourcing.
In these cases, active learning, which selects data to annotate based on their potential benefit to the translation system, has been shown to be effective for improving SMT/NMT systems while keeping the required amount of annotation to a minimum.
We developed an open-source, web-based, intelligent annotation tool to assist human translators. This tool automates incremental selection of translation candidate, which improves the model’s coverage, and supports partial annotation with keyword highlighting instead of annotation for full sentences. It allows you either new parallel data construction or quick domain adaptation with less effort. |
Japan Science and Technology Agency
NMT for Scientific Papers Powered by JST’s S&T Databases
Toshiaki Nakazawa
JST has been constructing scientific and technological database which contains major English and Chinese journals with their Japanese translations of the abstracts. With these valuable resources, JST and Kyoto University has developed a high quality NMT system for scientific paper. For Japanese-English translation, we used about 10 million parallel sentences and for Japanese-Chinese, we used about 4 million parallel sentences, both were extracted from the parallel abstracts in the database. In the exhibition, we will demonstrate our NMT system. |
XTM International Ltd.
XTM as a perfect platform for post editing Machine Translation
Yumiko Metsugi
XTM is an enterprise scale translation management system. It includes a browser based CAT tool, a highly automated and flexible workflow module, centralized translation memory and terminology management, terminology extraction, costing, reporting, management level dash-boarding and APIs.
XTM Connect offers out of the box integration with third party CMSs or repositories. Many of the leading Machine Translation engines are integrated into XTM making it an ideal platform for post editing MT. In addition, you can send productivity and quality data to TAUS DQF to benchmark your performance and quality against industry averages or use it to help you choose the best MT engine.
XTM is a secure, robust system that is available on the public cloud, privately hosted or can be installed on your own servers. |
NHK Science & Technology Research Laboratories
A New Japanese News Service with Simplified Japanese Paraphrasing Generated by Machine Translation System
Hideki Tanaka
The foreign population in Japan has steadily increased to currently over 2.4 million people, comprising 1.9% of the total population. Japan’ s public broadcaster, NHK, presently provides these people with the news in five foreign languages (English, Spanish, Portuguese, Chinese, and Korean) as well as in simplified Japanese. We will demonstrate another service under study: an online news service with optional reading assistance. The main feature is simplified Japanese paraphrasing superimposed above the original news text.
This is automatically generated by an in-house phrase-based statistical machine translation system between Japanese and simplified Japanese.
The news service can also superimpose the phonetic reading of words for all kanji characters and glossaries. The superimposition of the reading assistance is controlled by the viewers and is used to aid their Japanese-language learning. We will also display our prototype production system of the new service. |
Toyohashi University of Technology (TUT)
TUT’s Societal Implementation Project for Multilingual Information Outbound
Hitoshi Isahara
Toyohashi University of Technology (TUT) started joint R&D project with several companies in Japan in order to empower global public relations from Japan by information technology (IT). Furthermore, TUT is concentrating on developing links with IT companies, language service providers (LSPs) and local governments to create shared databases for documents in industries and tourism for business use, and other purposes. |
Yaraku, Inc
CAT YarakuZen
Tsuyoshi Minato
YarakuZen is a light-weight CAT tool for non-professional translators. With Easy-to-Use Interface and Adaptive NMT, YarakuZen aims to change the way translation asses are created and managed today within the organizations. |
Institute of Scientific and Technical Information of China (ISTIC)
ISTIC-NMT, a Japanese-Chinese Neural Machine Translation System for Scientific and Technical Texts
Zhou Lei (Contacts), He Yanqing (Principle)
ISTIC-NMT is a novel machine translation system consisting of an NMT engine and a supplementary PBSMT engine for processing UNK words. The NMT system is an attention based encoder-decoder architecture with Bidirectional Long Short Term Memory networks on each side applying residual connections to alleviate the vanishing gradient problem. An attention model is applied to retain the most probable source word towards an UNK word. The PBSMT is trained to translate the UNK words and the translated words are filled into the previous translation.
Besides, a domain-adaptive approach is applied to classify the training data and each category is used to train a separate NMT subsystem. In practice, texts to be translated are classified in the same way. The translation system will choose a corresponding NMT subsystem to carry on the translation process. |