Exploring Your Machine Translation Options
For companies looking to translate a high volume of content machine translation can be a key tool in your translation toolbox. Machine translation is now well-developed enough to flexibly fit into many workflows. But it may not be the right tool for every project.
In this post, we’ll give you a few pointers to help you decide if machine translation (MT) is right for your project, and then discuss some of the steps that you can customize in the machine translation process.
When to use machine translation
MT technologies allow you to batch-translate large amounts of text quickly, and automate parts of the process, but using it means making small tradeoffs in quality. We recommend using MT when:
- You have very large volumes of content or short turn-around times
- When quality is not a requirement
- As a placeholder for instant updates while higher-quality translation is being carried out
Machine translation technologies are less useful when quality is important. Don’t use MT when:
- Accuracy matters: when translating sales, marketing, legal, or safety content
- When errors could pose liability issues
- For branding, when you want an authentic voice
How to adapt MT to your specific needs
The first step of any MT process is to run the content through terminology enforcement. This is the step where you leverage prior translation memory, as well as glossaries and termbases, to define the parameters of your content. Then you run your content through an enterprise-level machine translation engine. These are usually Google Auto ML or MS Hub.
The combination of these two steps will give you a more custom MT process. By tweaking your terminology, termbases, and translation parameters, you will end up with a more accurate translation that better fits the style of your brand.
At this point, you will have a raw output from the machine translation. As the technology behind machine translation is not perfect yet, this “rough draft” will likely still have a few errors, mistranslations, or fluency issues. In some cases, however, this text might be good enough for your needs.
We recommend that even if you don’t need particularly accurate text, you should have at least some human post-editing. There are two basic levels, light editing and full editing, that you can use depending on how accurate you need the translation to be.
How is post-editing MT different from regular translation?
Unlike translation and localization, which need craft and skill to produce a coherent, beautiful text, post-editing is more about creating a functional text—one that accurately conveys meaning while keeping you to your schedule. For this reason, traditional translators or linguists may not always be the best people to post-edit outputs. When producing a post-edited text, it’s helpful to adjust your expectations: the text will not be quite as good as if as a translator had created it, but it will be good enough. So for these faster, functional jobs, it can be a good idea to use subject matter experts or people who have been trained in the process to review the output texts.
In post-editing, reviewers correct mistranslation and make small edits for fluency, in order to get the raw output to a publishable stage. The translation network Taus recommends minimally editing the raw output text when producing a “good enough” translation and focusing only on terminology or flow errors to create a “human-approximate” translation. For highly technical outputs, it can be useful to have a subject matter expert review the raw output, since they may be able to understand the technical content better.
By working with your existing technology and data, Venga can help your company build a custom machine translation process that fits your needs. Find out how on our website, or contact us for a consultation.