Some time ago, we launched our AI Essentials package which provides access to LLM-powered features for translation and linguistic review. Included with AI Essentials are Generative Translation and Smart Review. I will provide a very quick recap of these two features here.

Generative Translation

Generative Translation uses existing linguistic assets and custom instructions to generate translations by prompting an LLM. It can translate from scratch or augment a previous machine translation. The following diagram illustrates the flow.

By using translation memory, Generative Translation ensures that your previous style and tone are respected. Terminology reference also ensure that the vocabulary is correct.

Smart Review

Smart Review similarly uses existing linguistic assets to perform Translation Quality Assessment (TQA) on your translations. The following diagram illustrates the flow.

By using terminology, Smart Review ensures that the vocabulary is correct. Any spelling mistakes and mistranslations are also caught.

LLM support

At launch, AI Essentials supported two possible LLMs, namely:

1. An AWS Bedrock model, provided by RWS, for those customers who did not want to bring their own model subscription.
2. A customer’s existing subscription to an Azure OpenAI model.

Extended LLM support

We have just released a new feature which brings the potential for Trados to support thousands of AI models. When configuring your Connected AI, you now have the choice of adding an Azure AI Foundry model to be used by Generative Translation and Smart Review. This opens the possibility to use models such as Grok, DeepSeek, Llama, and Cohere, to name but a few. If you already have a subscription to one of these models and would like to use it in your Trados tenant, you can now do so. When you add a new LLM provider, simply fill out the credentials which you can get from your Azure AI Foundry account, and you are good to go.

Implications for quality and cost

It is not feasible for us to evaluate every single LLM available in Azure AI Foundry, but we have started to look at some of the more popular models with a view to providing some guidance on quality and cost. Watch this space for more details about this in the future.

Disclaimer

Azure AI Foundry hosts thousands of compatible models and more are added frequently. Please note that not all models are supported for all operations, but we will continue to iterate to ensure a wider range of supported LLMs. If you encounter problems using a particular LLM, please contact support who can guide you to alternative models or raise an enhancement to have your model supported.