How to streamline post-editing after machine translation?

As machine translation becomes more widespread, the demand for post-editing is also increasing. However, due to the differences in workability between human translation and post-editing, translators tend to avoid it.
However, by properly understanding the differences between human translation and post-editing work, and automating some of the post-editing work, it is also possible to streamline the post-editing process and reduce the burden on translators.

 

 

Table of Contents

1. Difference between human translation and post-edit

2. Reasons why translators dislike post-editing

3. Post-editing can be automated

4. Summary

 

 

1. Difference between human translation and post-edit

Post-edit is typically handled by translators with experience in manual translation, but the workability is different from manual translation.
What specifically is different?

 

 

In manual translation, 63% of the work is taken up by inputting (typing) the translated text.
In contrast, in post-editing, "checking/correcting fluency" accounts for 35%, and "checking/correcting accuracy" accounts for 22%.
Different know-how is required compared to manual translation.

 

2. Reasons Why Translators Dislike Post-editing and Characteristics of Machine Translated Text

Many translators feel resistant to learning post-editing due to the significant differences in workability compared to manual translation.
Furthermore, the post-editing rate is lower than that of manual translation, making it even more psychologically challenging.
One of the reasons translators avoid post-editing is because they are occupied with the mundane task of correcting terminology and style.
In order to promote machine translation, it is necessary for the translation departments of business companies and translation agencies to provide support to alleviate the burden on translators.

 

What is the difficulty of post-editing? Generally, the following issues are cited as problems with machine-translated text.

• ・Translation is done on a sentence-by-sentence basis. It cannot be translated with expressions or terms that depend on the context.
• ・There may be incorrect dependencies.
• - Terminology/Style is not consistent
• 　　English and Japanese service names are inconsistent
• 　　Inconsistent use of full-width/half-width English characters and symbols
• 　　Cannot insert spaces between full-width and half-width characters
• 　　Katakana words have inconsistent long vowels at the end, etc.

Wouldn't it reduce the burden on translators if even a part of these machine translation errors could be streamlined?

 

3. Post-editing can be automated

What can be done on the client side to improve the translator's work efficiency?
There are three main things that can be done.

 

①Clarification of Quality Requirements
Agree on quality requirements with the client and share them with the translator.

 

②Improving Accuracy and Fluency
Machine translation engines are updated daily. Regularly compare multiple engines and use the best one.

 

③ Terminology and style automatic correction
Let's automate as much as possible during post-editing.
Utilizing the terminology and style functions of the engine is also an effective method.

 

Here, we introduce MTrans for Memsource/Trados, which performs terminology and style automation on Memsource/Trados.

 



 

This is a text that has been translated by Google Translate.
Style: Full-width colon. Unnecessary spaces are present.
Term: A different translation is used from the glossary.
Both style and term need to be corrected with post-editing.

 



 

This is the result of applying machine translation with the automatic post-editing function of MTrans enabled.

• Style: Colon is half-width. Unnecessary spaces have been removed.
• Term: The translation of the term is being used.

Terms and styles are automatically applied.

 

How much has the workload been reduced specifically by automatic processing of styles and terminology?
Let's take a look at the amount of revisions and time for 109 sentences.

 

	Google	MTrans
Correction Rate	87% (95/109 sentences)	0% (0/109 sentences)
Number of Corrections	661	0
Revision Time	25 minutes 5 seconds	0 seconds
Translation Quality	Good (Translator could not focus on accuracy/fluent)	Excellent (translator was able to focus on accuracy/fluidity)

 

 

In the case of Google, out of 109 sentences, 87% required corrections, but MTrans had zero.
There was also a 25-minute difference in correction time.
Not only was the working time shortened, but the quality of the translation also improved as translators were able to focus on accuracy and fluency checks.

 

4. Summary

By utilizing style and terminology automation tools, not only can efficiency be improved, but it can also lead to an increase in quality.

 

MTrans (MTrans) for Memsource/Trados is a machine translation solution developed by our company for corporate clients. By implementing MTrans, you can use machine translation services from Google/DeepL/Microsoft without having to contract with each provider. It also includes features necessary for industrial translation, such as automatic application of terminology and style rules, which greatly reduces the workload for post-editing.

 

For more details, please see here.

 

MTrans for Trados
MTrans for Memsource