Google AutoML Translate was launched last year with the intent of allowing users to translate global content via automated machine learning without needing to learn how to code. The platform's Translate API, which is now in version 3 beta, has some newly announced features which aim to simplify batch translation, while also providing users more control over specific terminology.
Google announced a new Glossary feature that allows users to define specific terminology (e.g. brand names, street names, product names) in order to ensure that they are maintained during translation. The announcement says that users can “Simply define your company-specific names and vocabulary in your source and target languages, then save the glossary file to your translation project. Those words and phrases will be included in your copy when you apply the glossary in your translation request.”
The Translation API is also gaining the ability to select between pre-trained models and custom model translations, which should help to streamline workflows. Additionally, v3 will now also allow for batch translation, which will allow users to translate larger volumes of content in one translation request. With a single request, users can now upload multiple files utilizing different models and have them translated into different languages. Google provided this workflow as an example:
... if you wanted to translate an English product description on your website into Spanish, Japanese, and Russian, you could use your custom AutoML model for Spanish and a pre-trained model for Japanese and Russian. You would simply upload your English HTML file to your Cloud Storage bucket and send a batch request pointing to your Spanish AutoML model and pre-trained models for Japanese and Russian. Translation v3 will then output your HTML to Cloud Storage in Spanish, Russian, and Japanese in three separate files.
The company also announced a new free tier for the service that allows users to test these capabilities for the first 500,000 characters per month.