How to Translate a PDF from English to Chinese

The short version: upload the PDF, set the target language to 中文, pick an engine, download the translated PDF or the bilingual side-by-side copy. PDFPal does this while keeping the columns, tables and figures where they were — you get a PDF that mirrors the source page for page, not a wall of text.

The longer version is worth reading: English-to-Chinese is one of the harder pairs to get right in a PDF, and the ways it goes wrong are specific and predictable.

What makes English to Chinese hard inside a PDF

A PDF is not a document the way a Word file is. It is a set of instructions for painting glyphs at fixed coordinates on a fixed page. Translating into Chinese fights four properties of that model at once.

1. Chinese is more compact, so every text box changes shape. A paragraph of English routinely becomes 40–60 percent shorter in characters once translated. A text box sized for six lines of English may only need three lines of Chinese. If a tool simply drops the translation into the original box, you get half-empty boxes, orphaned headings, and captions that no longer sit under their figure. The fix is to re-anchor the translated text onto the original layout and rescale each paragraph to fit — which is what PDFPal's pipeline does rather than overwriting glyphs in place.

2. The source PDF almost certainly has no Chinese font in it. PDFs embed subsets of the fonts they use. An English fact sheet embeds Latin glyphs and nothing else. Write Chinese characters using that font resource and the viewer has no glyph to draw — you get tofu (empty rectangles), missing characters, or blank text that still copies correctly but does not render. A correct English-to-Chinese output PDF has to embed a CJK font in the file itself. PDFPal's renderer does this, which is why the output opens the same way on a colleague's laptop, a phone, and a print shop's RIP.

3. Line breaking works on completely different rules. English breaks on spaces and hyphenation points. Chinese has no spaces — a line can break almost anywhere between characters, except where it must not: a line may not begin with a closing bracket or with 。， 、 or ？, and it may not end with an opening bracket. A line breaker written for space-delimited text either refuses to break at all (text runs off the box) or breaks in a way that looks wrong to any Chinese reader.

4. Punctuation and mixed scripts change the metrics. Chinese punctuation is full-width: 。 and ， each occupy a full em, not the narrow advance of a Latin period. Meanwhile technical documents keep Latin runs inside Chinese sentences — product names, units, gene symbols, citation numbers, API names. Every one of those is a script boundary where spacing and line-breaking decisions get made.

The failure that happens before translation even starts

There is an older, cruder failure mode that no amount of language skill can fix: the text was extracted in the wrong order.

Here is real output from running a standard text extractor (pypdf) over page 1 of a two-column USGS fact sheet:

S
ince the early 1990's, the
magnitude and location of an
Not long ago, the fi rst thing that most
people did after feeling an earthquake was to

Three things went wrong mechanically: the drop cap S was severed from ince, two different columns got interleaved mid-sentence, and the fi ligature came apart with a space in it.

That is what happens when a pipeline flattens a PDF to a linear text stream before translating: the reading order is already wrong, so what gets translated is scrambled text. In English you can often spot the damage at a glance. Once the same scramble comes back as fluent Chinese, it reads as plausible nonsense and is much harder to catch — which is exactly why layout-aware extraction matters more for this pair, not less.

Choosing an engine for English to Chinese

PDFPal lets you pick the engine per document, so you can match it to the text rather than settling on one for everything.

• Google — deterministic machine translation. Fast, literal, consistent across runs. Good for forms, spec sheets, tables of figures, regulatory boilerplate, and anything where you want the same input to produce the same output every time. 1 credit per page.
• Free pooled Qwen model — no key, no cost, available on the free plan. Reasonable for getting the gist of a document before deciding it is worth a careful pass.
• Premium LLM engines (OpenAI, Claude) — these see more context per unit of text, so they handle terminology, register, and English idiom far better on prose-heavy documents: reports, papers, contracts, marketing copy. They cost 2 credits per page and require a paid plan.

If you do not like the result, run the same file again on a different engine and compare. And if a job fails outright, PDFPal refunds the credits automatically and offers a retry — optionally on a different engine, which is usually the right move when an LLM engine is what timed out.

Step by step

1. Sign in. You get 20 free pages, no credit card.
2. Upload the PDF. Up to 100 MB. Input has to be a PDF with a real text layer (see the next section if yours is a scan).
3. Set source and target. Source can be English or Auto-detect; set the target to 中文. Chinese is one of 40 target languages.
4. Pick your engine using the guidance above.
5. Optionally add a glossary — a CSV of source,target pairs. This is the single highest-leverage step for technical English-to-Chinese work, because it pins down the terms you actually care about.
6. Translate, then download. You get a translated-only PDF and a bilingual side-by-side version. On the free plan the output carries a small "Translated with PDFPal.ai" line; paid output has none. Batch translation of several files at once is a paid feature — the free plan handles one file at a time.

If your PDF is a scan, translate it in two steps

PDFPal has no OCR. If the pages are images of text — a scanned book, a photographed contract, a fax — there is no text layer to translate, and the job will stop and refund your credits with exactly that message.

There is a "scanned-PDF mode" toggle, but be clear on what it does: it only improves rendering of scanned-looking pages (black text on white). It does not perform text recognition.

The working order is: run the file through a dedicated OCR tool first, save the OCR'd PDF (one that has a searchable text layer), then upload that to PDFPal. Do it in that order and everything above applies normally.

Checking the Chinese output

Run this checklist on the downloaded PDF before you send it anywhere.

• No tofu. Scroll the whole document looking for empty rectangles or missing characters, especially in headings and figure captions, which often use a different font than the body.
• The text is real text. Use Ctrl-F to search for a Chinese word you can see on the page. If it finds nothing, the glyphs were drawn but not encoded — and downstream tools will not be able to read them.
• Copy-paste survives. Select a translated paragraph, copy it, paste it into a plain text editor. It should come back as the same characters.
• Tables still line up. Chinese cell text is shorter, so check that column widths did not collapse and that numeric columns are untouched.
• Terminology is consistent. Pick your five most important terms and search each one across the document. Inconsistency here is what a glossary exists to prevent — add the terms and re-run.
• Punctuation reads natively. Full-width 。 and ， inside Chinese sentences; no stray Latin periods left behind at the end of Chinese lines.
• Latin runs are intact. Product names, units, equations and citation numbers should be unchanged, not transliterated.

For the final review, hand the bilingual side-by-side version to a colleague who reads both languages. They can check meaning against the source without flipping between two files — the fastest review loop we know of for this pair.

Real before-and-after samples of layout-preserving output are at examples, and the credit-based plans are on the pricing page. Start with the 20 free pages and put your hardest document through it first.