Clean Smartcat TMX exports
Smartcat's strength — a marketplace where many linguists work against shared TMs, with MT woven through the flow — is also what fills those TMs with conflicting variants and machine output. Export the TM to TMX and reconcile it here into one clean, authoritative memory.
━━ the problem
When a rotating pool of freelancers writes to one shared Smartcat TM, the same source ends up with many slightly different targets of varying quality. Consensus (most-popular) or a trusted-reviewer ranking picks one authoritative variant per segment instead of a lottery.
Smartcat's integrated MT and pretranslation write machine-origin units into the memory. For a human-quality delivery TM — or clean training data — those need filtering out by their creationid/changeid origin signature.
Not everything in a busy Smartcat workspace is finished work. Empty targets, untranslated (target = source) units, and half-done drafts slip into the TMX export and quietly degrade your match quality until they're removed.
━━ how TM Cleaner handles it
Upload the exported TMX, preview the projected reduction and the duplicate / junk / MT samples, and download a reconciled TM ready to re-import into Smartcat or hand to a client.
- ✓consensus or trusted-reviewer dedup tuned for TMs written by a rotating freelancer pool
- ✓filter MT-pretranslated units by origin so the delivery TM stays human-quality
- ✓drop untranslated, empty-target, and draft leakage that inflates the memory
- ✓author and date filters isolate a chosen reviewer's work or a specific project window
━━ frequently asked
how do I export a TM from Smartcat to TMX?+
in Smartcat, open the Translation Memories section, select the TM, and export it as TMX. upload the resulting .tmx here; the cleaned output re-imports the same way or hands off to any TMX-capable tool.
can I keep only my trusted linguists' translations?+
yes. rank trusted reviewers by their changeid so their variant wins any conflicting duplicate group, or use the author filter to keep only chosen contributors and exclude the rest. combine with a date range to isolate a period of reviewed work.
is my TM content kept private?+
the upload goes to storage via a one-time signed URL, is parsed XXE-hardened, and is hard-deleted the moment cleaning finishes. download links expire after 24h and no TM content is retained.
━━ also clean
free tier · pay-as-you-go after · no subscription · see pricing