Translate stories

Introduction

When localizing a story, you have the possibility to get an automatic translation for each selected language.

For more information around how the translation works and how you can control it, refer to the dedicated paragraph below: How translation works.

The translation journey

1. Edit a story that contains text for at least one language. E.g., start with a story already written in English.
   

2. In the left language menu, select the target language for translation. E.g., select French.
   

3. In the popup window, select Suggested translation to get the automatic translation for the selected language. If you select Manual translation, the new localization gets the same text as the source one: no automatic translation, but you have to write it yourself.
   

4. The translation process starts. You can continue your work and come back after a while to check the translation.
   

5. Once available, the translation displays, and you can modify if needed.

   Publishing remains your responsibility

   The automatic translation workflow writes draft stories. Once you review and approve the translation, and only at this stage, it’s your decision to publish the story.

   On top, you have:
   - A banner reminding users that automatic translations require editor validation
   - The ability to Delete translation and remove the localization — which corresponds to a negative feedback for the translation
   - The possibility to close the reminder by selecting Looks good! — which corresponds to a positive feedback for the translation
   

How translation works

Tone of voice and accuracy

The automatic translation process involves multiple AI agents who handle translation, review, and collaborative improvements to ensure high-quality results.

The Deltatre team can adjust these agents to match each customer's specific business context, specific vocabulary, and tone of voice. Also, it is possible to add culture-specific instructions that improve drastically the output quality by matching the target audience culture.

The fine-tuning takes place during the FORGE deployment stage and is customized for each customer individually (customer segregation).

Feedback

Besides the direct feedback you provide by deleting a translation or choosing Looks good!, the automatic translation process also learns and improves over time by analyzing the differences between its automatic translations and the final published versions.

Data processing and management

Customers can select which AI platform powers the automatic translation workflow. The supported models are:

The selected AI platform determines the data processing and management.

* Deltatre certified means that the Deltatre team tested and validated the model for the vertical translation task.

Limitations

The selected AI platform determines the supported languages.

Be aware that the translation accuracy decreases with less widely spoken languages. Mitigation strategies depend on specific languages and require a dedicated discussion with the Deltatre team.