Machine translation is an essential component of translation automation. If used correctly, it leads to increased productivity, faster turnaround times, higher consistency, and improved accuracy.

Machine translation can be a useful tool in many domains and with different text typologies: from weather reports to tourism, from technical documentation to user-generated content, from websites to e-commerce.

But how would you choose the engine that best suits your projects? We are here to help. How? With a checklist, of course, that we created specifically for new users.

Are You a Freelance Translator?

• Do you have 1-3 language combinations?
• Do you have 1-3 specialisations?
• Do you translate at least 20,000 words every month?
• Do you see or expect an increase in your incoming volumes?
• Do you have specialized translation memories and glossaries at your disposal?
• Do you use a CAT tool, either a desktop version or a cloud version like Wordbee?

If you’ve replied YES to all the questions above, you might consider using a general, out-of-the box machine translation engine (like Google Translate, Microsoft Translator or DeepL) that can be easily implemented and that performs well or even beyond expectations in certain cases. After registration, you’ll be assigned an API key which you can then enter into your CAT tool configuration settings and start using the MT engine right away.

If your translation memories and glossaries are up to date, consistent and well “cleaned,” you can combine them with the machine translation system of choice. You can pre-translate the source text and then post-edit it (saving the post-edited segments in your translation memory); alternatively you use the engine “as-you-go” at segment level, deciding every time whether to keep, post-edit or reject the suggestions from the MT engine.

Are You a Language Service Provider (LSP) or a Translation Department?

• Do you have in-house skilled human resources? For example, software engineers and/or computational linguists.
• Do you translate at least 100,000 words every month in a specialised sector?
• Do you have a privileged client who could be eligible for a (dedicated) MT+PEMT workflow?
• Do you see or expect an increase in your incoming volumes?
• Do you have consistent, well-maintained specialized translation memories (amounting to no less than 100,000 segments) and glossaries?
• Do you have a pool of reliable and competent staff ‒ whether in-house or freelancers ‒ who could take care of post-editing (or are available for post-editing training)?

If your answer to all the questions above is YES, you might consider a specialized machine translation system — whether one already available or a customized engine — that allows you to leverage the language data in your possession. Many machine translation providers have specialized engines for a quick solution that can be accessed through a standard translation project management system. Keep in mind, though, that this solution can in some cases prove time-consuming and expensive, requiring a substantial financial investment and dedicated staff.

Some Cautionary Advice

Two more aspects come into this equation.

The first one is twofold, concerning intellectual property (IP) and confidentiality. As of May 25, 2018, compliance with the GDPR has become mandatory in Europe. Both freelance translators and LSPs must take this into consideration. 

• Do you process any content containing personal data?
• Do you process sensitive and/or confidential information?
• Do you have a procedure in place to anonymize the data in your possession?

From an ethical point of view, freelance translators and LSPs alike should inform their clients about the use of an MT engine (especially a public engine), or in any case, ask them about any specific requirements that might prevent them from using public MT engines or even MT tout court.

A general or specialized engine readily available may be a great solution, but ONLY if your texts do not contain sensitive or personal information that could end up in the engine’s training archive of data together with the content to be translated.

The second aspect is training. As an LSP, you should first train your in-house and freelance staff for post-editing. You could start with some general training and then provide your staff with special training on the engine(s) of choice, on project management for post-editing projects, and even on how to use RegEx and other more sophisticated functionalities.