Translation Memory

[object Object]

Translation Memory

A translation memory (TM) stores source–target segment pairs. Inside projects it provides automatic matching, zero-cost pre-translation, and a way to keep terminology stable before delivery.

All memory data is parsed, stored and indexed locally. Nothing is sent to an external server, and search works offline.

Managing Memories

Open Corpus Management → Memory.

List and actions

Columns: index, name, languages, entry count, creation time. Filter by name; select rows to merge.

Row actions:

Details: browse and search entries
Import: add entries to this memory
Export: four export options (below)
More: rename / clear / re-index / delete

The number of memories is plan-limited ("Memories" in plan benefits), and each memory has an entry cap ("Entries per memory"). Creating beyond the limit is blocked; once a memory is full you can no longer import into it (a single import may overshoot the cap once).

Import

Format	Notes
`.tmx`	Standard TMX, compatible with mainstream CAT tools
`.sdltm`	Trados Studio memory file
`.xlsx`	Spreadsheet format; download the sample from the import hint and follow its columns

After import the entry count and language list refresh. Wrong format, wrong encoding or mismatched language pairs raise an error.

Export

Export TMX: the whole memory as standard TMX
Export Excel: the whole memory as a spreadsheet
Split export by file count: enter how many files, and the memory is cut into that many TMX files
Split export by entry count: enter a per-file entry cap and split accordingly

"Split export" cuts one memory into several files; it does not split content into separate memories. To get independent memories, create new ones and import into each.

Merge

Select two or more memories, click Merge, and give the new memory a name. A merged copy is created; the originals stay.

Re-index

If search results look incomplete after a large import, run Re-index on that memory to rebuild the full-text index.

Browsing and Searching a Memory

The table shows one column per language contained in the memory
The search box matches keywords; the selector to its left limits the language (or searches all)
Matches are highlighted, with paged results
Double-click a row (or click Edit) to change its language texts; Delete removes a single entry

One-click dedupe

Combine any of these conditions (matching entries collapse to one):

Ignore case while deduplicating (on by default)
Delete entries containing empty text
Delete an entry when any two of its language texts are identical
When the selected language is duplicated, keep only one entry (multiple languages selectable)

A confirmation is required, and the number of removed entries is reported.

Using Memories in Projects

When you switch segments, the right-hand Corpus panel shows memory matches, applicable in one click
Ctrl + 1 applies the first memory match
In Pre-translate, choose TM first or TM only and set the minimum match rate (default 70%)
The match-rate buckets in File analysis and Project statistics are computed from memory matching

Best Practices

One memory per project or domain: cleaner matches than a single giant memory
Review before importing: a bad entry gets reused forever
Dedupe regularly: duplicates slow search and distort match statistics
Export important memories: TMX is the safest long-term archive

Notes

Capacity: the entry cap per memory depends on the plan
Languages: the language list is derived from imported content; one memory can hold several languages
Clearing and deleting are irreversible
Location: memories live in the local database and migrate with Data backup
Search: based on the local tokenized full-text index

FAQ

Q: What is TMX?

A: TMX (Translation Memory eXchange) is the standard exchange format for translation memories, supported by Trados, memoQ and other CAT tools. It is the usual way to move memories between tools.

Q: I imported a file but the entry count did not change.

A: Check the file format and language pairs, that the file has content, and whether the capacity cap was hit. .xlsx imports must follow the sample column layout.

Q: How do I handle duplicate entries?

A: Use One-click dedupe with the case, empty-text, identical-languages and per-language options. The number of removed entries is reported.

Q: Can one memory serve several projects?

A: Yes. Memories are global resources; matching and pre-translation use the local memories, and projects can share them.

Q: How is the match rate computed?

A: From source similarity (after segmentation), on a 0–100 scale. 100% means an identical source; matches below the minimum rate are not used for pre-translation but still show as reference in the Corpus panel.

Q: Can I back up a memory?

A: Two ways: export it as TMX/Excel, or export the whole database via Settings → Network & advanced → Data backup.

Q: I found a wrong translation. How do I fix it?

A: Open the memory, find the entry, double-click and edit it. New translations use the corrected text; existing caches and finished translations are not rewritten automatically.

Settings

Termbase