Introduced Google Neural Machine Translation integrated with Memsource for streamlining product manual translation tasks.

Background

WingArc1st Inc. boasts a top share of over 50% in Japan with a cumulative number of 21,000 companies using their report solution "SVF". They are also leading the BI market with their BI solutions "MotionBoard" and "Dr.Sum".

Not only in Japan, but also expanding the potential of data utilization with our own products and services, accelerating our global expansion to countries such as China, Singapore, and Australia.

The Document Solutions Department of the Product Management Department, which you requested this time, is responsible for maintaining the documents created by the development team and releasing product manuals in both web (HTML) and PDF formats.

We have been assisting with the translation of English manuals for overseas expansion since 2012, and in July 2017, we participated in the "Neural Machine Translation Seminar" and started the Google Neural Machine Translation implementation support project.

Challenges before implementation

Human resources for document translation are insufficient. Want to utilize Google Neural Machine Translation.

Previously, product manuals were translated by humans using the translation support system Memsource.

However, the average translation volume is high, ranging from 200,000 to 500,000 characters per month, making it difficult to secure human resources for translation. As a result, we have started considering the introduction of machine translation for the purpose of efficiency.

Specifically, this is a method of using machine translation for new translation parts that do not match with existing translation memories, and having translators make corrections to the translation.

In addition, when introducing machine translation, we also wanted to apply UI terminology and styles. However, our product manuals are not in the standard HTML format, but rather a customized HTML format created in-house. As a result, there were some parts that could not be effectively integrated with the standard features of Memsource for machine translation.

[Translation Assets]

・Translation Memory: 120,000 segments

・UI Glossary: A UI glossary exists for each product group

Introduction Process

After participating in the seminar in July 2017, we conducted meetings and specification design through meetings and emails, and implemented the following process from September.

・Memsource Plugin Development (for Google Neural Machine Translation Integration)

・Glossary Application

・Style Processing

"I participated in a seminar by Human Science Co., Ltd. and learned that it is possible to customize machine translation output. I also had a personal consultation. The key was whether we could solve the issues that arise when integrating Memsource and Google machine translation."

In addition, we were considering the possibility of applying our own terminology and style as an additional effect. We directly consulted with the developers and worked on refining the requirements while confirming the output results of machine translation through exchanging sample files. We also provided feedback on improvement requests even after starting the test operation. Those were also improved if technically feasible.

(WingArc1st Product Management Department Document Services Team Representative)