Skip to content
LionrapidLionRapid Docs

Translation memories

A translation memory stores translation units — source–target pairs from your past work — so the platform can reuse them automatically. When the same or similar text comes back in a later project, the stored translation is offered (or applied) instead of being translated from scratch.

A TM has one language pair and works in that direction only: an English → French TM does not apply to French → English work.

Creating and filling a TM

Section titled “Creating and filling a TM”

There are three ways a TM gains units:

  • Import a TMX file on the Resources page. The language pair and unit metadata are read from the file.
  • Create an empty TM — on the Resources page or during translation setup — and attach it to a project. Confirmed work in that project adds units to it.
  • Project additions. Any TM attached to a project accumulates units from the work done there. Units added this way are withheld from exports until the project is fully paid — see withheld additions.

The TM viewer

Section titled “The TM viewer”

Click a TM card on the Resources page to open the viewer: a table of all units with the source text on the left and the target text on the right. The list loads incrementally, so even large memories open quickly.

  • Search the units by text. You can scope the search to the source side, the target side, or both, and jump between matches. A filter toggle restricts the list to matching units only.
  • Edit a unit by double-clicking either cell. Press Enter (or click elsewhere) to save, Escape to cancel. The corrected unit is what every future reuse will get — fixing it here fixes it everywhere it is offered from now on.
  • Remove a unit from the right-click menu (or the row’s menu button) when it should never be reused again.
  • Download TMX exports the whole memory.

How TM leverage works

Section titled “How TM leverage works”

When a TM is attached to a project’s language pair, its units are matched against the project’s source segments.

  • An exact match (100%) means the stored source text is identical to the segment (ignoring surrounding whitespace).
  • A fuzzy match is a similar-but-not-identical unit, scored from 0 to 100 by text similarity. Word order and punctuation affect the score. Matches are grouped into the conventional bands: 95–99, 85–94, and 75–84. Candidates below 75 are not offered — at that distance, editing a wrong suggestion costs more than translating fresh.
  • A unit that differs only in casing scores in the 95–99 band, not as an exact match.

Where matches reach you

Section titled “Where matches reach you”
  • In the editor, the best match (and a runner-up when there is one) is shown for the segment you are working on, with its score, so you can take it over or adapt it. See TMs, termbases & glossaries in the editor.
  • Before machine translation, exact matches and high-scoring fuzzy matches can be filled from the TM so those segments never go to MT at all. The controls for this — exact reuse and the “Skip MT at TM match ≥” threshold — are part of translation setup; how they interact with the MT pipeline is described in MT engines and routing.
  • In the project analysis, the classic word-count breakdown used for quoting and reporting buckets every translatable word by its best leverage: repetitions, exact matches, the three fuzzy bands, segments that will be copied rather than translated, and new words. The same matching rules drive the analysis, the editor, and the pipeline, so the numbers always agree.

Authoritative TMs

Section titled “Authoritative TMs”

A project manager can mark a TM attachment as authoritative. Exact matches from an authoritative TM are applied automatically instead of being offered as suggestions, and when several TMs offer the same match, the authoritative one wins. Use this for a memory you trust completely — for example a client-approved TM of previously delivered work.

Section titled “Related pages”