dsh-pdf-translate

Dropping an English academic or technical PDF into a generic translator usually yields a Chinese version that looks nothing like the source: formulas reduced to a1, a2, a3 fragments, font hierarchy lost, headers and footers scattered. dsh-pdf-translate closes that gap – drag an English PDF into the DSH sidebar and, within seconds to a few minutes, receive a Chinese PDF and Markdown that faithfully mirror the original typesetting.

Rather than applying a fixed template, it first measures every typographic parameter from the source – margins, font sizes at each level, line and paragraph spacing, accent colours, header and footer positions and text – then slices the content into semantic blocks (titles, body text, theorem statements, standalone equations, headers, footers) and translates them block by block. Formula fragments extracted from the PDF are reconstructed into LaTeX and rendered offline by the bundled KaTeX build, with equation numbers right-aligned via \tag. The re-typesetting stage generates HTML from the measured parameters, compressing each section independently (spacing first, then font size) under the hard rule that one extra page is acceptable but no content is ever clipped. The final PDF is produced through headless Edge via the CDP Page.printToPDF API, matching the source paper size.

It is aimed at readers who need an English academic or technical PDF translated into Chinese with strict layout fidelity: font hierarchy, equation numbering, headers, footers, and overall page count all expected to match the source. Note that it only processes electronic PDFs with a text layer – scanned documents are explicitly rejected (no OCR) – and translations are currently English-to-Chinese only, using the default model displayed on the panel.