"Q&A at the "Machine Translation and Post-Editing Utilization" Seminar"

On April 21, 2020, we held a web seminar with the theme "Why not review your translation workflow? Utilizing machine translation and post-editing" and had many participants.

At the seminar, we talked about the cost-effectiveness of machine translation, post-editing, and customer case studies.

Here, we will introduce the questions and answers received at the seminar.

Q Why does inconsistency occur in machine translation?

A The specifications of the engine make it difficult to give a clear answer, but from experience, there are cases where the same expression is used before and after, or where the translation may be different depending on the collocation.

Q Are there any points to note for Pre-edit?

A The key is to create Japanese documents that are easy to machine-translate, such as making the sentences short, unifying terminology, and not omitting subjects.

Q What is the recommended translation engine?

A Genre (e.g. IT, medical, tourism), purpose (e.g. manual, email), and translation volume will determine the most suitable engine. While there are highly rated engines based on reviews, it is important to choose the most suitable engine for your company. It is recommended to select a few and verify based on factors such as the presence of translation memory, accumulation rate, and availability of terminology databases.

Q Pre-edit and post-edit, who will be responsible for them? Will it be done by your company or by the requesting company? Currently, our internal staff is handling the English translation, but we are considering outsourcing due to quality issues.

A In any case, it is important to establish rules such as setting quality requirements and preparing style guides in order to solve translation quality issues. At Human Science, we not only offer pre-edit and post-edit services, but also provide support when our clients carry out these tasks themselves. Please feel free to consult with us.

Q Customer training documents, but are there any basic procedures or points to keep in mind when creating translation dictionaries in multiple languages?

A The basic principles are the same for both Japanese and English, as well as for multiple languages. First, let's extract terms (product names, buttons) that should be unified so that users are not confused.

One common issue in multilingual dictionary management is inadequate maintenance. It is important to establish and implement operational rules for maintaining the latest version, as updates may not be consistently applied across all languages and can become fragmented.

Q Google Translate has greatly improved its performance, but how does it compare to paid translation software?

A Google's translation accuracy has certainly improved, but the performance of machine translation will vary depending on the field and availability of terminology for your translation target. We recommend verifying with actual translated documents before deciding which automatic translation software to introduce.

Q There was a discussion about changing the level of editing (full, light, etc.), but won't there be differences in the end result depending on the worker? It seems difficult to establish a common understanding of the level.

A As you said, even if you say "Please do a light edit", the interpretation may differ depending on the person and there may be a difference in the finished product.

Therefore, it is important to establish and share criteria in the guidelines for "what points to correct". By including examples of corrected sentences and examples of sentences that should not be corrected, the understanding of the person performing the work can be deepened and unity can be achieved. At Human Science, before requesting a large amount of work, we review the post-edited translated text in a small amount of text and provide feedback if there are any points that need to be corrected, in order to ensure quality consistency.

Q Is neural machine translation the main method used when requesting work from your company?

A Neural machine translation is not always required. Depending on the client's requirements, we often provide human translation in addition to machine translation. Even in the case of machine translation, we always take into account the client's requirements (cost, quality, delivery time, etc.) and propose appropriate machine translation software.

Q To fully utilize the effectiveness of neural translation, I heard that a considerable amount of information needs to be taught to the machine translation. Is it possible to start with small projects?

A When you need a dedicated engine for your company, as you say, a large amount of corpus is necessary to train it. However, with a general translation engine, it is possible to start with small projects based on what the engine has learned.

Q Can English translation data be stored in Trados translation memory if the manual is created in Word?

A Possible.

Free Download

Japanese to English Large-Scale Document Translation Service Information