ReglyphReglyph vs Google Translate, Side by Side: Scans Get Refused. What About Digital PDFs?
Reglyph is our product, so we can't claim to be neutral. Both tools got the same files, and every result is shown as it came out.
- Scanned PDF: Google Translate refuses it outright with "Can't translate scanned PDFs". Reglyph translates it with the layout and background intact.
- Digital PDF: Google Translate does better than many people expect, keeping columns, figures and paragraph indents. But on both papers the equations and tables broke, so Reglyph comes out ahead there too.
| Page | Languages | Better result | What decided it |
|---|---|---|---|
| Guide page with colored panels (scan) | EN → ES | Reglyph | Google Translate refused it |
| Dense two-column paper (digital) | EN → FR | Reglyph | Google broke the chirp-mass equation into fragments; we lost the indents |
| Paper with a parameter table (digital) | EN → KO | Reglyph | Google flattened the parameter table into a paragraph |
Scanned PDFs: it stops at the upload
Google Translate doesn't run OCR, and a scan has no text layer, so there's nothing for it to translate. We uploaded a scanned page from the US Dietary Guidelines; a message popped up in the corner and that was the end of it. Other scans get the same response. This guide explains why.
Here is the same page through Reglyph. The Spanish sits where the English was, and the green title bars, photos and dark blue tab are all still there:
Digital PDFs: the page survives, the equations and tables don't
Google Translate does translate digital papers, and the columns and figures stay where they were. That makes these two pages a real side-by-side comparison. The differences are in the parts of a paper that aren't plain sentences.
A dense two-column paper (English → French)
Google's French text is cleanly set and keeps the paragraph indents, which ours doesn't. The problem is the math. The chirp-mass equation in the left column comes apart into loose pieces ("M ¼ (m1m2)3=5 …"), and in the running text equals signs turn into "¼" and the solar-mass symbol disappears, so "M ≃ 30 M☉" reads "M 30 M". On a physics paper that's the part you can least afford to lose. Ours keeps the equation whole.
A paper with a parameter table (English → Korean)
Look at Table 1 on the right. Google Translate turned the two-column parameter table into one paragraph of text ("크기 4096, 레이어 수 32, 헤드 크기 128, …") and translated the configuration names along the way, so n_layers is no longer something you can search for in the code. The figure captions stayed in English, and the body text switches between two font sizes from one paragraph to the next. Ours keeps the table as a table.
Which one should you use?
- Scans, phone photos, anything without a text layer: Reglyph. Google Translate won't accept them.
- Digital PDFs that are mostly text and that you only need to read: Google Translate is free, needs no account, and does the job.
- Pages with equations or tables, or when you need a translated PDF you can actually use: Reglyph. 5 pages a day are free: try it here.
More side-by-side tests: Reglyph vs DeepL, Reglyph vs Reflo and Reglyph vs Smallpdf.
Translate your scanned document now
Upload a scanned PDF or a photo — Reglyph OCRs it, translates it, and rebuilds the page so tables, stamps, and figures stay exactly where they were.
Translate 5 pages free, every dayFrequently asked
Does Google Translate keep the layout of a digital PDF?
Partly. In our test on two research papers it kept the columns, figures and paragraph indents, but it broke a displayed equation into fragments, replaced equals signs with "¼", and flattened a parameter table into a single paragraph.
Which is better for translating PDFs, Reglyph or Google Translate?
For scans, only Reglyph works; Google Translate refuses them. Both handle digital PDFs. Google Translate is free and fine for text-only pages; on pages with equations or tables, Reglyph did better in our test.
You make Reglyph. Is this comparison fair?
We can't be neutral, so we kept it checkable: the same files for both tools, Google Translate's screen and output shown as they came out, and our own lost paragraph indents left in. The source pages are public, so you can run the test yourself.