Google Machine Translation - Examples and Challenges for IT Manuals

In the previous blog, I wrote about Google Neural Machine Translation (GNMT) supporting Japanese.

If it's a personal website page in English that you want to know, it might be enough to translate it to the general meaning with GNMT. However, when it comes to practical translation, it's a different story. When the target audience is a product user or a customer who has purchased a service, a certain level of translation quality is naturally required.

Many readers may have the question, "While acknowledging the improvement in GNMT's quality, how much can we actually evaluate the quality?"

So, I examined examples and issues of translation quality in the main translation genres of IT, medical, and machinery at Human Science. The translations by GNMT are based on the time of writing this article.

IT Manual (English to Japanese)

Although it is translated correctly in terms of meaning, it is necessary to confirm whether it conforms to the glossary and style guide.
For example, if the translation of "deployment" is unified as "配備" in the glossary, "展開" needs to be corrected to "配備".
Regarding style, it is necessary to make corrections according to the style guide, such as whether there is a space between full-width and half-width characters ("API" and "の"), and whether parentheses should be full-width or half-width.
Also, it is necessary to confirm and correct whether the referenced section title is actually "アプリケーションプログラミングインターフェイスを使用した展開（API）".

It is necessary to check whether the translated text from Google Translate conforms to the style guide. Also, it is necessary to check if the example sentences match the section titles as they are reference sentences.
These manual checking processes are called PE (Post-edit).

Related Services

Neural Machine Translation Solution

Machine Translation Seminar

Translation Open Seminar Scheduled