Automatically Machine Translate Content and Other MT Improvements Are Here

Mike Giannakopoulos
January 31, 2017
3 min read

Today, we’re rolling out a number of improvements to how Machine Translation (MT) and translation autofill work in Transifex. Together, these changes will make your day-to-day localization experience much better. Let’s take a look at the updates.

Automatically Machine Translate content

Automatically machine translating content is already available for Transifex Live projects. Now, we’re bringing it to file-based projects. If you enable this feature, then whenever you upload new files or update existing ones, untranslated strings will be automatically machine translated after 100% Translation Memory matches are used (if TM Autofill is enabled).

Enable automatic Machine Translation in Transifex

This means you won’t have to manually run machine translation if you’re using it as part of your translation workflow; for example, in conjunction with human post-edit.

We’ve also released a new webhook event for when translation fillups are complete so you can get notified when the translations are done.

Note that Automatic Machine Translation for file-based projects is only available on the Pro plan and up.

Machine Translated strings in Translation Memory

While Machine Translation has gotten better over the years, it’s still not quite as good as human translators yet. To make sure your TM’s quality isn’t affected by this, only machine translated strings that have been marked as reviewed (or manually run for a specific string) will be added to the TM. This means your TM will only have translations approved by a human.

Know where translations came from

You can now easily distinguish between human-translated strings and machine-translated strings. In the Editor, machine translated strings will say either “MT: Google or “MT: Microsoft.” It will, however, still say “Translated by Username” if you manually trigger MT for a specific string in the Editor using the Lightning icon.

Transifex Machine Translation Translated by Window

And in the filters, you can filter for strings by Machine Translation source as well.

Translation origin filter

Improvements to how variables are handled

Strings often contain variables. So we made some improvements to how they’re handled when you use Machine Translation:

  • Variables and placeholders will be left untranslated, including for right-to-left languages.
  • If a validation check fails for a string, machine translation will skip that string and keep running for the remainder of the strings.

Support for Microsoft Machine Translation (Azure)

At the end of last year, Microsoft announced that Microsoft Translator subscriptions have been moved from Microsoft DataMarket to Azure. In order to keep using Microsoft Translator after April 30, 2017, you’ll need to a new subscription key purchased through Azure. In the Machine Translation settings, you’ll now see an option for Microsoft Translator Text API (latest), which you can use to input your new key.

Machine Translation Service

If you have an existing Microsoft DataMarket subscription from before December 31, 2016, you can still use it by picking the Microsoft Translator (deprecated) option. Just note that your Microsoft Translator subscription will be canceled automatically (by Microsoft) on April 30, 2017.

Manually trigger translation fillups

If you have Translation Memory Autofill (use 100% TM matches as translations whenever new content is uploaded) and/or Automatic Machine Translation enabled, you can now manually trigger them in your project settings to bring your resources to 100%.

Whenever you trigger this, the Translation Memory Autofill will run first, before Machine Translation is used. This is useful for when you add new languages to a project or when you enable these settings after you’ve uploaded or updated your content. If these settings are already enabled when you upload or update your content, then you don’t need to do anything.

We hope you enjoy these machine translation updates in Transifex.

Mike Giannakopoulos
FacebookgithubGoogle+Fill 88Twitter