A PDF as a bilingual book, with every word clickable

Why the Viewer does not re-draw the page as HTML

Re-rendering a PDF as HTML is what destroys columns, tables and figures — it is the reason most in-browser translation of documents produces something unreadable.

Instead the Viewer draws each page to a canvas, pixel for pixel as the document specifies, and lays a transparent text layer precisely on top. Every word in that layer has its own box, so it can be clicked, hovered, coloured and marked, while what you see is the untouched original. Pages render and release as you scroll, and zoom and fit-width behave normally.

Why the permission is requested at the button, not at install

An extension page runs on its own origin, so fetching an ordinary https://…/file.pdf is a cross-origin request — and almost no document server sends the header that would allow it.

Chrome therefore has to ask. The request is deliberately deferred until you press the button, so installing NoaLingua never presents a broad access prompt, and the permission is only ever used to read the document you just asked for.

Why translation arrives in batches

The Reader translates in series rather than all at once, so a long document becomes readable from the beginning before it has finished. Paragraphs are aligned so the original and its translation start at the same height, and hovering a sentence highlights its counterpart in the other column.