Skip to content
English
Submit a request
  • There are no suggestions because the search field is empty.

How Learning Mode, Incognito Mode, and customer assets affect model training

Lara lets customers control whether their translation content can be retained and used for model improvement.

This article explains:

  • how Learning Mode and Incognito Mode differ;
  • how Translation Memories and glossaries are used;
  • how to prevent translation content from being used for training;
  • when an Enterprise configuration may be appropriate.

In brief
  • In standard Learning Mode, translation inputs and outputs may be retained and used to improve Lara’s translation models.
  • In Incognito Mode, the content of the translation request is not retained or used for model training.
  • Existing Translation Memories and glossaries can still guide translations in Incognito Mode.
  • Customers requiring a contractual no-training commitment while continuing to use Learning Mode can request an Enterprise configuration. Contact our Sales team to discuss your requirements


Learning Mode


In Learning Mode, source text, translated content and context may be securely stored and used to improve Lara’s translation models, including through model training.

This processing is described in Sections 4.2 and 4.7 of our Terms and Conditions.

Learning Mode is intended for customers who want Lara to learn from their translation activity over time and are comfortable with the relevant content being retained for that purpose.

How Translation Memories and glossaries are used


Translation Memories and glossaries are customer-specific resources used to guide translations at request time. This process is called adaptation.

When configured for an account, these resources can be applied through both the Lara web interface and the API.

Translated does not take customer Translation Memories or glossaries and use them directly as standalone training datasets for shared models.

There is, however, an important qualification: when a translation is processed in Learning Mode, the resulting content may be retained and used for model improvement. Because that output may reflect a customer’s Translation Memories or glossaries, the adapted output may also be used for that purpose.


Incognito Mode


Incognito Mode prevents the content of a translation request from being retained or used for model training.

When Incognito Mode is enabled:

  • the source content is not retained;
  • the generated translation is not retained;
  • the request is not used to improve models;
  • existing configured Translation Memories and glossaries may still be referenced at translation time.

Incognito Mode does not delete previously stored Translation Memories or glossaries. It prevents Lara from retaining or learning from the content of that specific translation request.

Operational metadata required for security, billing and service integrity may still be processed in accordance with our Data Processing Addendum.

Enabling Incognito Mode through the API


For API requests, Incognito Mode must be enabled for every relevant request by setting:

"no_trace": true

Where supported, Translation Memories and glossaries can be used together with no_trace to guide the translation without retaining the source content or generated output.

For implementation details and code examples, see the Translate Text documentation in the Lara Developer Hub

The configured assets guide the translation at inference time, while the source content and generated translation are not retained or used for model training.

Enabling Incognito Mode in the web interface


Before submitting content through the Lara web interface:

  1. Open the translation settings.
  2. Enable Incognito Mode.
  3. Confirm that Incognito Mode is active.
  4. Submit the content for translation.

Configured Translation Memories and glossaries can continue to guide the translation, but the input and output of that translation are not retained or used for model training.

Team enforcement

A Team owner can enforce Incognito Mode for the entire Team account. When enforcement is enabled:

  • Incognito Mode applies to all translations submitted by Team members;
  • Team members cannot disable or override it;
  • configured Translation Memories and glossaries remain available for translation-time adaptation.

The Team owner retains the ability to change this setting.

Enterprise enforcement

Enterprise customers can request server-side enforcement of Incognito Mode. Under this configuration:

  • Incognito Mode is enforced for the entire account;
  • neither Team members nor the Team owner can disable or override it;
  • the restriction can only be changed by Translated following an authorized customer request;
  • the configuration can be documented in the applicable Enterprise agreement.

Contact our Sales team to discuss Enterprise enforcement.


Enterprise no-training configuration

Customers who need to continue using Learning Mode in the Lara web interface while contractually excluding their content from model training can request an Enterprise configuration.

Under this configuration, Lara configures the customer’s account infrastructure so that customer assets, translation inputs and outputs submitted through the web interface are excluded from model training, while Learning Mode and stored Translation Memories remain available.

This configuration does not apply to API requests. For API usage, customers must enable Incognito Mode by setting "no_trace": true on each relevant request.

Contact our Sales team to discuss an Enterprise no-training configuration.


Frequently asked questions


Are uploaded Translation Memories used directly to train shared models?

No. Customer Translation Memories and glossaries are not taken and used directly as standalone training datasets.
However, translations generated in Learning Mode may reflect those resources, and the resulting adapted output may be retained and used for model improvement.


Can I use Translation Memories with Incognito Mode?

Yes. Existing configured Translation Memories and glossaries can be referenced to guide the translation at request time. The request itself is not retained, does not update the memory and is not used for model training.


Is no_trace an account-wide setting?

No. Through the API, no_trace=true must be included in every request for which Incognito Mode is required.
In the web interface, the user must enable Incognito Mode before submitting the content.

 

Is Incognito Mode equivalent to an Enterprise no-training configuration?

Both provide a no-training outcome for the relevant translation content, but they operate differently.

Incognito Mode prevents individual requests from being retained. An Enterprise configuration allows the customer to continue using Learning Mode and stored resources under a contractual no-training commitment.


Where can I find the applicable contractual terms?


See:

 

 

This article is about:

  • Learning Mode vs. Incognito Mode: Understanding how data retention and model training differ between the two settings.
  • Translation Memories and glossaries: How these assets guide your translations without being used directly as standalone training datasets.
  • Data protection: How to effectively prevent your translation inputs and outputs from being stored or used for model training.
  • Enterprise configuration: When to request a dedicated infrastructure to maintain a contractual no-training commitment.