Translation Memory

Weblate comes with a built-in translation memory consisting of:

  • Manually imported translation memory (see User interface).

  • Automatically stored translations performed in Weblate (depending on Translation-memory scopes).

  • Automatically imported past translations.

Content in the translation memory can be applied to strings in several ways:

For installation tips, see Weblate Translation Memory, which is turned on by default.

Translation-memory scopes

The translation-memory scopes ensure privacy for different projects and users. Sharing of translations is also available.

Note

When upgrading from older Weblate releases, existing translation-memory entries are converted to scopes by a periodic Celery background task. Until this backfill finishes, existing entries can be temporarily unavailable in suggestions and memory management views. Keep Celery running after the upgrade so the periodic task can start or resume the migration. Site administrators can monitor the backfill and duplicate-entry consolidation in Administration > Performance report.

Imported translation memory

Importing arbitrary translation memory data using the import_memory command makes memory content available to all users and projects.

Per-user translation memory

Stores all user translations automatically in the personal translation memory of each respective user.

The default value for new users is automatically adjusted based on the Autoclean translation memory configuration. If automatic cleanup is enabled, this is disabled by default to prevent reintroducing inconsistent translations.

Per-project translation memory

All translations within a project are automatically stored in a project translation memory only available for this project.

Entries attributed to an existing restricted component are only available to users who can access that component. This also applies when project memory is downloaded or accessed using the REST API.

Legacy entries which can not be attributed to a current component use the project access rules. This can happen when the component path has changed or the component has been removed.

Workspace translation memory

Projects in the same workspace can share translation memory without enabling the global shared translation memory. Workspace translation memory has to be enabled both in the workspace settings and in the individual project workflow settings.

Entries attributed to an existing restricted component are only available to users who can access that component, even when they can access another project in the workspace. Unattributed legacy entries use the workspace access rules.

See also

Shared translation memory

All translations within projects with shared translation memory turned on are stored in a shared translation memory available to all projects. Turning off contribution to shared translation memory stops previously contributed automatic entries from being used as shared suggestions.

Restricted components do not contribute new entries to shared translation memory. On Hosted Weblate, a project with restricted components can not enable shared translation memory, and a component can not be restricted while its project uses shared translation memory.

Unattributed legacy entries which are already in shared translation memory use the shared translation memory access rules.

Please consider carefully whether to turn this feature on for shared Weblate installations, as it can have severe implications:

  • The translations can be used by anybody else.

  • This might lead to disclosing secret information.

  • Make sure the translations you share have good quality.

See also

Translation memory status

Added in version 5.13.

Translation memory entries can have two different statuses: active and pending. Pending entries are included in suggestions, but with a quality penalty applied. Depending on the configured score threshold, the penalty can exclude even exact matches from suggestions and automatic translation. Entries imported from translation memory files are stored as active. If Autoclean translation memory is enabled, matching pending entries are removed when a translation becomes active.

Autoclean translation memory

Added in version 5.13.

When enabled for a project, Weblate replaces older automatically created translation memory entries when a translation becomes active:

  • With Enable reviews enabled, cleanup happens when the translation is approved.

  • Without review, cleanup happens as soon as the translation reaches the translated state.

For the same source string, component, context, and source/target language pair, Weblate removes matching non-file entries from translation memory, including entries with a different target text. This applies across personal, project, and shared translation memory scopes. The current translation is then stored again as an active entry in the scopes that are enabled for that change.

Entries imported from external translation memory files are not cleaned up automatically. Entries with a different context are kept.

In the Docker container this can be configured using WEBLATE_DEFAULT_AUTOCLEAN_TM.

Note

Enabling automatic cleanup also changes the default for new user profiles to not contribute to personal translation memory. This prevents reintroducing inconsistent translations that were supposed to be cleaned up.

See also

Autoclean translation memory

Managing the Translation Memory

User interface

Translation memory can be managed from several places in the Weblate UI:

  • Open the user menu and choose Translation memory to manage your personal translation memory.

  • Open a project and choose Translation memory from the project menu to manage translation memory for that project.

  • Open a workspace and choose Operations ↓ Translation memory to manage automatically created translation memory for that workspace. Editing the workspace is required.

  • Open Administration and choose Translation memory to manage uploaded translation memory for the whole Weblate instance.

The translation memory page shows entry counts for the selected scope. Depending on the scope, it also lists entries by origin, component, or language pair. The listed entries can be downloaded as JSON or TMX. Users with the required permissions can delete entries, and the project view can rebuild translation memory for the whole project or for individual components from the current translations.

Removing a component preserves its automatically created translation memory by default. Because the component access rule no longer exists after removal, the retained entries use the access rules of their remaining translation-memory scope. To remove those entries together with a component, select Delete translation memory created from this component in the removal form. Category removal offers the equivalent option for every component in the category and its nested categories. This removes attributed entries and legacy entries matching the current component path. It does not infer paths used before a component or category rename or move. Such unmatched entries can be deleted by origin in the project or workspace translation memory management, or using the REST API. The REST API accepts the delete_memory boolean in the request body or query string for component and category removal. Personal and uploaded entries are preserved.

The project view also shows whether shared translation memory and autoclean translation memory are enabled for the project, with a link to the project workflow settings when the user can edit the project.

The workspace view can rebuild all currently contributing projects or an individual contributing component. It manages workspace-scoped automatic entries only. Translation-memory files cannot be uploaded into a workspace scope.

Translation memory files can be imported on the same page. Uploaded files are processed during the request, stored as active, and added to the selected scope:

  • Personal uploads are available in your personal translation memory.

  • Project uploads are available in the selected project’s translation memory.

  • Administration uploads are available as uploaded shared entries for the whole Weblate instance.

Hint

Translation memories in various formats can be imported into Weblate. The TMX format is provided for interoperability with other tools. All supported formats are TMX, JSON, XLIFF, PO, and CSV.

Source and target languages must be selected when uploading XLIFF, PO, or CSV files if the language information is not provided by the file itself.

See also

Weblate Translation Memory Schema

../_images/memory.webp

You can search translation memory while editing strings from the Automatic suggestions tab.

Admin interface

Administrative users have additional translation memory management controls. In Administration > Translation memory, the page lists uploaded shared entries, shared entries, and total entries for the Weblate instance. Depending on permissions, it can import uploaded shared memory, delete uploaded entries, and download uploaded, shared, or all entries as JSON or TMX.

Added in version 4.12.

The project translation memory view also allows rebuilding parts of or the entire project translation memory. Automatically generated entries for the selected component or project are deleted, and the memory is populated again from the current translations in the background. Entries imported from files are preserved.

Management interface

Several management commands can manipulate translation memory content. These operate on the translation memory as a whole, unfiltered by scopes (unless requested by parameters):

dump_memory

Exports the memory into JSON

import_memory

Imports TMX or JSON files into the translation memory

cleanup_memory

Removes all entries with pending status from the translation memory

Added in version 4.14.

The Weblate API covers the translation memory. This allows automated manipulation for different purposes, or based on events in the translation cycle.