Translating with Google Adaptive (Beta)

You have the option to utilize Google Adaptive as a machine translation (MT) provider within Smartling. 

With the help of Smartling's MT profile feature, you can safely store your Google credentials within Smartling. Once you have configured an MT profile for Google Adaptive, you can assign this profile to any machine translation workflow step. 

Interested in using Google Adaptive? Please contact your Customer Success Manager to enable access and assist with setup.

What is Google Adaptive?

Adaptive translations use LLMs combined with small datasets to provide high-quality translations, often on par with AutoML Translation custom models. With Google Adaptive, you don't train or maintain any models. Compared to custom models, adaptive translation works well for producing translations that are similar in style, tone, and voice, using your provided translation examples as a guide. 

Requirements

You have two options for using Google Adaptive: Bring Your Own Key or Use a Smartling Key. 

Bring Your Own Key

To use adaptive translation, you will need a Google account to set up Google's Adaptive Translation model. You will need to create a training set, upload it to Google Console, and specify the path in the Smartling MT Profile. You will also need to bring your own API key in the form of JSON credentials.

Use a Smartling Key

Smartling offers the option to use a Smartling key. With this service, Smartling can optimize your Translation Memory and upload it to Google Console as a training set on your behalf. If you're interested in this service, please reach out to your Customer Success Manager to discuss this option.  

Supported Languages

Please refer to the Google documentation for the most up-to-date list of supported languages. 

How It Works

Users create a training set with example translations as segment pairs, each consisting of a sentence in a source language and its corresponding translation in the target language. Adaptive translation then uses these examples to tailor its translation output. 

To use Google Adaptive in Smartling, the dataset must be created and populated via API. You will first create a dataset and then populate it with example translations from a TSV file or a Translation Memory eXchange (TMX) file(s). See Setup in Google. 

For more information please refer to the Google adaptive documentation.

Important Considerations

Google Adaptive is generally recommended for smaller datasets where you don’t have enough Translation Memory data for custom engine training. If your TM contains more than 5,000 translation units, custom MT engine training may produce better results. Additionally, high latency is a known limitation with Google Adaptive. For example, a medium-sized job of 30,000 words can take 30 minutes or more to receive translations. Therefore, Google Adaptive cannot be used as the MT provider in the CAT tool, and we do not recommend using it with Smartling’s account-wide MT integrations.

Getting Started with Google Adaptive

First, set up Google Adaptive within Google. Then, follow the instructions to add Google Adaptive as a translation provider in Smartling. 

Setup in Google

You must enable Cloud Translation API, have a billing account associated with your Google account, and then enable Adaptive Translation. You will need several pieces of information from your Google Cloud Account: 

1. Project ID: A unique identifier for your project in Google
2. Location: The location of your data (us-central1 or eu)
3. Credentials: A Service User account with access to the necessary project 
4. Dataset ID: A dataset is required for Google Adaptive and can only be created via API. You will name the dataset and create an ID through the API. 

To create and populate your dataset via API, please follow steps 1 and 2 as outlined here. Once you have this information and have created your dataset, follow the steps below to configure Google Adaptive in Smartling. 

Setup in Smartling

To set up Google Adaptive as a translation provider in Smartling, you will need to create a new MT profile, which can then be assigned to a machine translation workflow step. 

Before getting started, reach out to your Customer Success Manager to enable Google Adaptive for your account.

To configure Google Adaptive in Smartling:

1. Create a new MT Profile by following these steps.
2. Complete all required fields.
   • Profile Name (profile name is visible in Smartling)
   • Project ID: example_project_id
   • Location: your location will either beus-central1 or eu
   • JSON credentials: copy and paste the entire JSON file into this field, including the brackets
3. Define language-specific rules, by clicking New Language-Specific Rule
   
   • Bilateral Mode: enabling this feature allows for specific language pairs to be machine translated as both source > target and target > source.
     • This allows you to reverse the language pairs and leverage source content as translations
   • Enter the Source Language and Target Languages: Since there is no generic model for all language pairs, you will need to add language-specific rules. Each rule will always consist of a source-target language pair (e.g., en <> de). 
   • Enter the Dataset ID: This is the unique name you gave your dataset when you created it via API. 
   • Click Test Integration to ensure the credentials are correct and that everything is configured properly.  
4. Once finished, click Save. 

You can now assign the MT profile to a machine translation workflow step, see Configure MT Profiles in Workflows.