ArXiv now offers papers in HTML format

Original Article ↗ Hacker News Discussion ↗

ArXiv’s new HTML versions of research papers are being welcomed as a major usability and accessibility upgrade over traditional PDFs, especially for mobile reading and for blind or low-vision users who rely on screen readers. Commenters contrast the responsiveness, copy‑and‑paste friendliness and potential for customization of HTML with the fixed, print‑oriented strengths of PDF, debating trade‑offs around layout, fonts, annotations and long‑term citability. Behind the scenes, they note that robustly converting LaTeX into faithful HTML remains technically challenging, but view arXiv’s move—building on projects like ar5iv and LaTeXML—as a significant step toward more accessible scholarship.

Reading experience: HTML vs PDF

  • Many praise HTML as vastly better on phones and small screens. No zooming or horizontal scrolling, dark mode support, and responsive layout reduce friction.
  • Two-column PDFs are widely disliked on mobile due to “zig‑zag” scrolling; others argue two columns are optimal on wider displays for faster scanning.
  • Several still prefer PDFs for large monitors, iPads, and printing. They value fixed layout, precise pagination (“top of page 7”), and easy annotation in PDF tools.
  • Fonts and color schemes are contentious. Some find the new HTML font and dark theme unpleasant; others override fonts in the browser. A few would like classic LaTeX fonts in HTML.
  • HTML is praised for text reflow, custom styling, and copy/paste; users want controls for font, line spacing, color schemes, and possibly multi-column toggles.

Accessibility and disabilities

  • A major benefit cited is accessibility, especially for blind users. PDFs—particularly LaTeX‑generated ones—often lose semantic structure and are hard for screen readers.
  • HTML and MathML are seen as more promising but not yet ideal for advanced math. Some blind users still prefer working directly with LaTeX source despite its own problems.
  • Justified text and certain visual choices may hurt accessibility; some argue for ragged-right text, others note trade-offs.

Conversion quality and technical challenges

  • The LaTeX→HTML pipeline (via LaTeXML / arxiv-readability) is acknowledged as fragile:
    • A small percentage of articles fail entirely; a sizable minority have known conversion errors.
    • Newer LaTeX packages worsen failure rates; figures, subfigures, and algorithms are common pain points.
  • Bugs are reported (e.g., reference links, author lists, spacing). Contributors expect gradual improvement but warn coverage will be “hit-or-miss” for now.
  • TeX is described as a complex, page-oriented typesetting language, making high-fidelity HTML output hard. Page-bound constructs, floats, and lack of semantics complicate conversion.
  • Suggestions to use computer vision or LLMs draw skepticism due to fidelity and reproducibility concerns.

Rollout, scope, and ecosystem

  • Currently, HTML is available only for new submissions; arXiv plans to backfill the corpus over time.
  • Previous third‑party efforts (ar5iv, arxiv‑vanity, etc.) are acknowledged as precursors; some will eventually redirect to the official HTML.
  • HTML is expected to enable better scraping, meta-analysis, TTS, Kindle/epub workflows, and richer interfaces (e.g., discovery tools), though interactive “project page” style content is seen as out of scope for arXiv itself.