Translation QA for SDL Trados, Word and PDF files
Find the errors in a translation before you deliver it — numbers, tags, terminology, spelling and consistency — and fix them right in the report.
Word, PDF or a scan / photo — the paragraphs are paired automatically.
By running a check, you agree to our Terms of Service and Privacy Policy.
- Trados, Word & PDF: return packages, SDLXLIFF, bilingual Word files, or an original with its translation — even scans
- Every error type: numbers, tags, terminology, spelling, punctuation and consistency
- Fix it in the report: correct the translation and download the corrected file
- Rule-based, no AI: free, no account needed, files deleted after 24 hours
Translated documentrequired
Two separate files — Word (.docx), PDF or a scan / photo (JPG, PNG, TIFF). Scanned pages are read with OCR, and the paragraphs are aligned automatically, even when one was split, merged or left out.
Bilingual filesrequired
Return package or project package (.sdlrpx, .sdlppx) or one or more .sdlxliff files.
Translation memoriesoptional
.sdltm or .tmx — used to find translations that differ from your TM. Packages already include theirs.
Termbasesoptional
.sdltb, .tbx, .xlsx or .csv (source; target; status) — forbidden and approved terms.
Used for the spelling, number and punctuation rules — and, for scanned PDFs and images, to recognise the text. “Detect automatically” reads the language from the file; for scans in other than Western European languages, choose it here.
Content
Punctuation
Tags
Terminology
Consistency
Spaces & characters
Numbers & placeholders
Spelling
What the QA checker looks for
Comparable to the QA verification in desktop CAT tools, but in your browser. Choose exactly which checks run — every result points to the segment, highlights the problem and tells you why it was flagged.
Content
Empty / untranslated target. Source has content but the target is empty.
Target identical to source. Target is a copy of the source (possibly left untranslated).
Suspicious length. Target is far longer or shorter than the source.
Repeated words. The same word twice in a row in the target ("the the").
First-letter capitalisation. Source starts with a capital letter but the target does not (or vice versa).
Punctuation
Ending punctuation. Source and target must end with the same punctuation (. ? ! : ; … and script equivalents).
Leading punctuation. Leading bullets/dashes and Spanish ¿ ¡ must match the source.
Punctuation spacing & doubling. Doubled punctuation, stray space before , . ; : ! ? and missing space after them.
Brackets & quotation marks. Bracket/quote counts differ from the source, or brackets are unbalanced.
Tags
Tag errors. Missing tags, extra tags and broken tag nesting compared with the source.
Tag order. Same tags as the source but in a different order (often legitimate — off by default).
Terminology
Forbidden terms. The target uses a termbase term that is marked forbidden / deprecated.
Termbase terms not used. A termbase source term occurs in the source but no approved target term is in the translation.
Consistency
Same source, different translations. Identical source segments translated in different ways.
Same translation, different sources. Different source segments given the identical translation.
Differs from translation memory. The source matches a TM entry exactly but the target differs from every TM translation.
Spaces & characters
Unbreakable (non-breaking) spaces. Non-breaking spaces of the source missing from the target, extra ones, and French typography spaces.
Spaces, tabs & line breaks. Double spaces, leading/trailing space, tabs and line breaks that differ from the source.
Invalid / invisible characters. Replacement characters, zero-width characters, soft hyphens and control characters not in the source.
Numbers & placeholders
Numbers. Numbers in the source that are missing from the target, or extra ones.
Placeholders, URLs & e-mails. Variables such as %s, {0}, $name$, URLs and e-mail addresses must match the source.
Spelling
Spelling (target language). Words in the target that are not in the Hunspell dictionary for the target language, with suggestions. Names, codes and words that also appear in the source or termbase are ignored.
Guides:
Which files can I check?
SDLRPX return packages, SDLPPX project packages and SDLXLIFF files, Word translations (.docx) and PDF translations, including scanned PDFs and images. A package already contains its translation memory and termbase; you can add more TM (.sdltm, .tmx) and termbase files (.sdltb, .tbx, .xlsx, .csv).
Can I check a Word translation?
Yes, in two ways. Under One file, upload a bilingual Word file — a table with source and target columns, such as the Trados Studio “Export for bilingual review” or a memoQ / Phrase bilingual DOCX. Under Two files, upload the original and the translation as two separate Word files. The paragraphs of the two files are paired automatically, also when the translator split, merged or left out a paragraph; the report says how the files were aligned, and the context view shows the neighbouring pairs. All checks, the search, termbase and TM work the same way, and your corrections are written into the translated Word file, which you can download.
Can I check a PDF or a scanned translation?
Yes. Choose Two files and upload PDFs, or scans and photos (JPG, PNG, TIFF) — you can also mix them, e.g. a Word original with a PDF translation. Pages without a text layer are read with OCR (up to 50 scanned pages per file); for scans in languages outside Western Europe, set the language under Advanced. A PDF translation can be checked, searched and exported, but corrections cannot be saved into a PDF — fix them in the document the PDF was made from.
Is anything sent to AI?
No. The QA checker uses fixed rules only. Your text, terminology and translation memory stay on our server and are deleted automatically.
How does the spell check work?
The target language is read from your file and checked with Hunspell — the open-source spell checker used by LibreOffice, Firefox and Chrome — running on our server, with dictionaries for more than 50 languages. Names, codes and words that also appear in the source or your termbase are ignored. Languages written without spaces between words (Chinese, Japanese, Thai) are not spell-checked.
How are forbidden and approved terms handled?
Termbase entries marked forbidden or deprecated are flagged as errors when they appear in the translation. If a source term occurs but none of its approved translations does, it is reported as a warning — inflected forms are recognised, but please verify.
Can I fix the issues here?
Yes. Click Edit on any target (or double-click it), correct the text and press Ctrl+Enter. Tags stay protected, the segment is saved as approved in the SDLXLIFF and the report refreshes, so fixed issues disappear. When you are done, download the corrected SDLXLIFF or — for Trados packages — the corrected SDLRPX return package. The list can also be exported to Excel or CSV.
How does the search work?
The Search segments tab searches every segment, not only those with issues. Fill in the source box, the target box or both, and tick does not contain where needed — for example “source contains Admin, target does not contain Yönetici”. Options include match case, whole words, regular expressions, tag names, status, tag and QA-issue filters.
What can I do with inconsistent translations?
The consistency checks show every other segment in the group with the differences highlighted and how often each translation is used. You can copy the most used translation into one segment or into all differing segments at once, or edit any of them by hand.
Can I see the segments around an issue?
Yes. Click any segment number to open it in context with the five segments before and after it, and load more if you need them. You can edit segments right there.
Are Trados tracked changes supported?
Yes. Deleted text is ignored by all checks, so a struck-through word never causes a false error. When you save a correction, the tracked changes in that segment are accepted.