TM Matches Not Appearing

Your TM is full of relevant entries, it opens fine in ✏️ Edit/Maintain TM, and still the match panel stays empty. Work through the checks below in order – the first two explain most cases. If you’re still stuck, the diagnostic script tells you which one it is.

1. Is the TM switched on for this project?

A TM gives matches only when its Read box is ticked for the open project, and a new project starts with every TM switched off, so that a new job doesn’t quietly pick up the last job’s resources.

1. Open the 💾 TMs tab.
2. Tick Read for the TM(s) you want to use (and Write if new translations should be saved into one).

When no TM is switched on, the log says so once per project: “No TM is switched on for this project, so no TM matches can be shown.” You’ll find the log under Settings → 📋 Log, or in its own window via Tools → 📋 Log Window….

2. Does the project have the right language pair?

TM matches are looked up for the project’s source and target language. A project created with the wrong pair finds nothing, even though the TM itself looks perfectly fine.

• See the project’s pair under Project → 📋 Project Info… (Languages).
• Compare it with the Languages column in the 💾 TMs tab.

Some imports in older versions could create a project with the wrong pair without telling you:

Import	Problem	Fixed in
CafeTran bilingual DOCX	Every project became English → Dutch	v1.10.372
Déjà Vu X3 bilingual RTF	Dutch → Spanish when no languages were found in the file	v1.10.372
memoQ bilingual RTF	Dutch → English when the header row couldn’t be read	v1.10.372
memoQ bilingual DOCX	English → Dutch when the column headers weren’t recognised	v1.10.370
memoQ XLIFF	Language “unknown” when the file declared none	v1.10.370

Current versions read the pair from the file where they can, and ask you to confirm or pick it where they can’t – see CAT Tool Integration Overview. To fix a project that was created with the wrong pair, import the file again and choose the right languages. If you’ve already translated in it, export the bilingual file first and import that one: the bilingual imports bring existing translations along.

Note

TMs with upper-case language codes. Before v1.10.370, TM entries tagged with a bare upper-case language code such as IT (as some older TMX files spell it) could never be found by any project. This mostly affected TMs imported before v1.10.253. Language codes are now compared regardless of case, so these entries match as soon as you update – nothing needs re-importing.

3. Exact matches appear, but never fuzzy ones

Fuzzy matches are found through a full-text index inside the TM database; exact (100%) matches don’t use it. If that index stops working, exact matches carry on as normal, ✏️ Edit/Maintain TM still lists every entry, and yet no fuzzy match ever appears – with no error anywhere.

Since v1.10.371, Supervertaler tests the index each time it starts and rebuilds it if it can’t find its own entries. The log then shows:

[TM] The full-text index cannot find its own rows, so fuzzy matches would come back empty. Rebuilding it from 200,000 translation units - this may take a moment.
[TM] Full-text index rebuilt; fuzzy matching is working again.

The rebuild takes a few seconds on a large TM (about 2.6 seconds for 200,000 entries), so that one start-up may take a little longer.

Fix: update to v1.10.371 or later and restart Supervertaler. If you can’t update yet, the diagnostic script can do the same repair.

4. Look in the log for “TM search failed”

Before v1.10.371, a match search that failed – because of a locked database, a damaged index or a bad language value, for example – simply showed an empty match panel, exactly as if the TM had nothing to offer. Now the log says so and names the cause:

⚠️ TM search failed, so the match pane is empty: <the reason>
   This is a fault, not an empty TM. Run scripts/sv_tm_diagnose.py for details.

If you see this, run the diagnostic script below and include both in your report.

Run the diagnostic script

sv_tm_diagnose.py is a small Python script that looks at your TM database and your most recent projects and tells you why matches are missing. By default it only reads: it opens the database read-only and changes nothing.

It isn’t included in the Windows download. You need Python 3 installed (the script uses only Python’s standard library, so there’s nothing else to install).

1. Download sv_tm_diagnose.py from GitHub (use the Download raw file button). If you run Supervertaler from source, it’s already in the scripts folder.

2. Open a command prompt (Terminal on macOS and Linux) in the folder where you saved it.

3. Run:

   python sv_tm_diagnose.py

   (On Windows, use py sv_tm_diagnose.py if python isn’t recognised.)

The script finds supervertaler.db in the Supervertaler folder in your home folder, and checks the most recently saved of your projects there and in Documents. If your data folder is somewhere else, or you want to check a different project, give both paths, the database first:

python sv_tm_diagnose.py "D:\Supervertaler\resources\supervertaler.db" "D:\Jobs\Manual\Manual.svproj"

What it reports

• your Workbench version, operating system, Python and SQLite version;
• the language pair of the project, and which TMs have Read ticked for it;
• the language pairs actually stored on the entries of each TM, and whether the project can see them;
• whether the full-text index can find entries that are known to exist;
• a Verdict, which is one of:
  • No TM is switched on for this project – see check 1;
  • The full-text index is dead: fuzzy matches cannot work, exact ones can – see check 3;
  • No active TM entry matches the project’s languages – see check 2;
  • TMs are on, languages line up and the index works – then look in the log for “TM search failed” (check 4).

When you ask for help, please send the entire output.

Repairing the index with the script

python sv_tm_diagnose.py --repair

With --repair, the script rebuilds the full-text index in place – but only if it found it dead – and then tests it again. If the index is working, nothing is changed. You can combine --repair with the two paths above.

Caution

--repair writes to your TM database. Close Supervertaler (and Supervertaler for Trados, which shares the same database) before you run it. Supervertaler v1.10.371 and later make the same repair automatically at start-up, so you only need this on an older version.

Still no matches?

Open an issue on GitHub with the script’s full output and any “TM search failed” lines from the log.