Machine Translation Post-editing

Machine Translation (MT) solutions have improved enormously in recent years. Once difficult to deploy and maintain, and with varying, sometimes unusable quality, today’s solutions provide powerful and sophisticated algorithms, and advanced user functionality with much improved results.

What this means is that we can now pre-translate large volumes of content, using constantly improving machine translation programs. This can be done very quickly with a relatively short preparation time, and at relatively low cost. Content which previously may not have been translated can now be made available more cheaply, allowing brands and companies to engage more with their users and audiences. It also allows us to reduce time to market, which is particularly valuable for very large volumes.

Brazilian Portuguese is one of the languages for which machine translation tools provide relatively good results. This is due to its linguistic proximity both to English and the other Romance languages, and due to the availability of data, used to train the leading engines. Depending on the type of content, and using English as the source language most MT solutions can provide an output for Brazilian Portuguese that translators can work with. Very often we use Machine Translation solutions in conjunction with Translation Memories, so that only those phrases without a match need to be machine translated, and the linguists are also able to refer to the TM to identify approved translations and assure consistency.

Naturally, almost all machine translated content, needs to be fully reviewed, edited and quality-assured before publishing. Typically, we provide two types of post-edit.

Types of Post-editing

High or Full Post-editing means we edit the MT output to the point where it is indistinguishable from a human translation. The end result is a translation which is 100% accurate, grammatically correct and error-free, and consistent throughout in its use of terminology, style and tone, which may be defined by the client.

High or Full Post-editing is used in scenarios where quality is very important and where there is no tolerance for errors.

Low or Light Post-editing means we edit the MT output to the point where its meaning is clear, accurate and unambiguous, but the translation may still contain issues of grammar, spelling, and consistency; and we do not spend time on conveying the original style or tone.

Low or Light Post-editing is used in scenarios where understanding is more important than language quality, and where linguistic and stylistic issues may be tolerated.

The only time MT output would not be reviewed is when the content is destined for a forum where a lesser linguistic quality would not reflect badly on the brand, perhaps in an online user forum, or where content is made available on the explicit understanding that quality will not be good.

Otherwise all content is reviewed, edited and QAed by skilled and experienced post-editors.

Quality Expectations

The most important thing, therefore, when approaching Machine Translation, is to make sure that the desired quality level is agreed between both client stakeholders and language services provider and that the clients’ tolerance for language and stylistic issues is discussed in detail in advance, so that their preferences can be considered when defining the desired quality level and the corresponding customized solution.

Our Solution

TextTrans has invested heavily in training our resources and equipping them with the tools and the knowledge needed to become expert MT Post-editors, no matter what the content or the quality level required. We have an experienced and scalable team of linguists who can provide varying levels of post-editing across all of our major verticals. Several of our team are recognized experts in the field and share their experience in our customized training programs for our linguists.

Training Provided includes:

General Training around MT Topics and Technologies

Introduction to Machine Translation & post-editing,
Types of machine translation engines and tools
Different quality levels
Challenges of post-editing
Examples of post-editing levels
Differentiating review for post-editing

Client Specific Training

In this case we would bring the linguists through:

Sample content
Sample MT output
Desired quality level
Language assets and reference material
Challenges specific to the account
Differentiating review for post-editing

Building Teams

Linguists are selected to work on PE projects based on their experience in PE, also in the domain, and if they have an affinity with the content. From an assessment perspective, we ask the linguists to carry out tests for full and light PE, and provide them with feedback on this. We then often decide to carry out review of their work for an initial phase, to make sure they are hitting the required quality levels. Our post-editors are also assessed based on their productivity, and their ability to post-edit enough words at the requisite quality level.

All linguists are supported by our usual Quality Process and all of the usual Quality Assurance Tools, which verify things like inconsistent, missing or unmodified translations, spelling errors, mismatch in numbers or tags, deviations from key terminology and more.

To learn more about our Machine Translation and Post-editing solutions, please contact us here.

Machine Translation Post-editing