Translation Memories

The Translation Proxy can be configured to maintain and leverage internal translation memories. These memories can contain more than one target locale allowing leveraging them for any pair of locales contained within.

As opposed to project dictionaries, translation memories are keyed to the user creating them, and can be assigned to any project the user has Backup Owner privileges or higher.Any project can contain an arbitrary number of memories, but one must always be designated the default: only this memory will be utilized when segments are being written; while pre-translation and suggestions are fed from all memories assigned to the project with applicable locale configurations.

Using TMs

Translation memories are initialized empty, and must be first configured with locales. After the target locales are defined, the memory can be populated. There are three ways a segment can be injected into the memory:

  • TMX-import: The Proxy can digest a standard TMX (Translation Memory eXchange) file and populate a designated memory based on its contents. The memory must be configured with at least one of the target locales of the TMX file. Duplicate segments are either merged (if for different locales) or discarded during import.
  • Project population: The Proxy can populate the memory from the project it is currently assigned to. The memory must be configured with at least one of the project’s target locales for this to work. If there are several locales assigned to the memory, the UI will treat them as a set, and offer the intersection of the memory and the project’s locales as the default. This set can be further restricted by removing locales from the population task before committing it. This action is logged in the project’s Audit Log.
  • Individual Injection: If a memory is assigned to the project with at least one locale present on both, it will be available on the Workbench for use. Confirming one or more segments will trigger the saveToMemory action, injecting the segment in its current form into the memory.

Memories are used for two tasks on the UI:

  • Pre-translation tasks can leverage any memories assigned to the project, provided the memory is configured with the correct locale. This applies to user-triggered Pre-translation, as well as Automatic Pre-translation triggered by new content. Only content with confidence levels above the user-configured threshold will be used, matches with lower percentages are discarded.
  • The Workbench automatically leverages any memories with the appropriate locales on segment selection. Matches are displayed in the Suggestions tab of the sidebar, along with their match percentages. Additionally, all memories on the project with the applicable target segments can be queried at will by entering a search term.

Confidence levels

The Proxy differentiates five levels of similarity between individual segments/entries (see here). Memory application yields the best results between 101% and 99% - 98% matches disregard tagging, and may need manual adjustment. However, searching below 98% is also possible, using the Google Search API, but these matches should be used with caution, as there is no guarantee regarding their accuracy due to the Search API’s word stemming.