[Feature Request] Optimize fragmented text spans for better browser translation compatibility

[021gink]

Is the feature relevant to the Firefox PDF Viewer?

No

Feature description

Background & Motivation

I'm developing a tool to help users translate PDF documents using browser built-in translation features (e.g., Chrome Translate, Edge Translator, Firefox Translations). However, I encountered a critical limitation:

The Problem:

PDF.js renders text content as highly fragmented <span> elements, which breaks browser translation engines.

Here's what happens when you try to translate an MIT course PDF using Edge Translator:
Original text (readable):

Translation result (completely unusable):

When I first encountered these translation issues, the obvious solution would have been to abandon PDF.js and use specialized PDF translation tools (e.g., Adobe Acrobat with built-in translation, dedicated PDF translators, or OCR-based solutions).
However, I made a deliberate choice to invest time in fixing this because:

PDF.js's Layout Fidelity is Unmatched
PDF.js provides pixel-perfect rendering that preserves:

• Original typography and font rendering
• Precise positioning of diagrams and figures
• Complex multi-column layouts
• Mathematical notation positioning
• Embedded vector graphics

My Solution
After studying the PDF.js architecture, I developed an algorithm that:

Merges fragmented text items into natural text blocks before DOM rendering
Detects superscripts/subscripts by analyzing Y-offset and font scale
Marks them with zero-width Unicode characters (\u200B for superscript, \u200C for subscript)
Converts to HTML _{/ ^{tags during rendering}}

After Merged spans：

After translate:

Here is the repository for my project: https://github.com/021gink/PDFtranslate

Although my project still has several shortcomings—such as text expansion when translating Chinese into English, and changes to styles like hyperlinks and italics after line-level merging—I believe this line-level span merging algorithm offers a new approach. If the span merging fits the requirements of the official library, I plan to fork it later, integrate this algorithm into the appropriate place, and submit a PR.

Other PDF viewers

No response