Student or educator? Sign up with your school email and get 6 months of Pro free.

The DOI matches its Crossref record; 36 of the 36 references it cites were found in a registry or an index; 20 of the 50 sources its claims cite were checked for support.

iTagPDF: Towards Finally Automating PDF Accessibility

Mowar, Peya, Steinfeld, Aaron, Bigham, Jeffrey P·2026·Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems

Abstract

Most academic research is ultimately disseminated through documents in the PDF format. This format has advantages in flexibility and portability, but presents challenges for accessibility that have stubbornly resisted solutions despite decades of attempts. Tagging PDFs is hard to automate because tags are currently generated visually, not semantically, which makes the output cluttered and manual correction tedious and error-prone. Ironically, this semantic structure already exists during authoring but is discarded during PDF rendering. This raises an obvious question, can we use this lost semantic information to better automate tagging in PDFs? In this paper, we develop iTagPDF, a system that refines generated metadata using the semantics in the source documents of research papers. We demonstrate that the metadata generated by iTagPDF already surpasses what authors currently submit to ACM conferences on many criteria. Our approach represents a concrete step toward finally automating accessibility remediation in research paper PDFs.

Overview

CoChat Summary

TL;DR

iTagPDF automatically generates PDF accessibility metadata by combining visual PDF processing with semantic context from LaTeX source documents. It outperforms Adobe Acrobat auto-tagging and author-submitted PDFs on many accessibility criteria, though it still relies on author verification and has limitations for some content types.

Key findings

  • On 40 ACM CHI/ASSETS paper PDFs, iTagPDF achieved 95.28% bounding box accuracy, 95.96% classification accuracy, and 97.26% reading order accuracy.
  • In a comparison on 25 papers, iTagPDF scored 81.36% average accessibility criteria versus 49.68% for Adobe Acrobat auto-tags and 55.59% for author-submitted PDFs; normalized scores were 95.19%, 87.03%, and 80.97%, respectively.
  • iTagPDF surpassed author-submitted metadata on reading order (96.42% vs. 88.62%), headings tagged (91.58% vs. 71.92%), lists tagged (83.09% vs. 46.46%), tables tagged (89.79% vs. 66.67%), and captions tagged (82.92% vs. 0.0%).
  • Adobe Acrobat performed better on table structure in the smaller subset (81.29% vs. iTagPDF’s 72.40%, N=5), and authors’ submitted PDFs had higher figure alt-text coverage (81.60% vs. iTagPDF’s 75.20%).

Methods

  • iTagPDF processes PDF page images with a YOLOv11 object detector fine-tuned on DocLayNet, extracts text with Tesseract OCR, parses LaTeX source into semantic chunks, and aligns visual regions to source chunks using word-overlap heuristics plus a GPT-4o LLM agent.
  • It generates document-level tags such as `<H1:H3>`, `<P>`, `<Caption>`, `<Table>`, `<L>`, `<Figure>`, `<Artifact>`, `<Note>`, `<Formula>`, `<BibEntry>`, and `<Author>`, plus content-specific tags `<TH>`, `<TD>`, and `<LI>`, then embeds metadata using iText Core.
  • Evaluation used 40 papers (20 CHI, 20 ASSETS; 554 pages; over 10,000 visual segments) from arXiv LaTeX sources and ACM Digital Library PDFs, with manually computed Box Error Rate, Classification Error Rate, and Reading Order Error Rate; accessibility criteria were compared against Acrobat auto-tagging and author-submitted PDFs on a 25-paper subset.

Limitations

  • The paper states iTagPDF currently focuses on LaTeX-authored research paper PDFs and may not extend directly to Microsoft Word or Adobe InDesign sources.
  • It relies on heuristics and LLM-in-the-loop mapping, which can fail when no reliable mapping is found, especially for figures or captions, leading to redundant `<Artifact>` tags; LLM outputs are not strictly deterministic, though the paper reports near-deterministic behavior for most tags.
  • The paper states accuracy is limited by off-the-shelf modules, iTagPDF does not produce some tags such as `<Link>`, does not identify inline formulas, does not handle references as list elements, and still relies on author verification for figure alt text.
  • Evident from the text: table structure evaluation used only N=5, formula tagging N=2, and alt-text quality was not evaluated.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1145/3772318.3790289

Cites

36 of 36 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

17 Valid · 9 Not valid · 16 Not checked · 83 Own results and interpretations

49 more claims · 2 with their cited sources checked

Create a free account to see all 125 claims and to verify the 6 references still open.

See all claims — free account
  • It may state two separate claims as one. 36
  • Probably not a substantial claim. 52

Create a free account to see these claims.

Citations

Cites

Show all 34 references

Research works printed but not linked: 1

  1. Adobe Inc.2023. Adobe Acrobat Pro DC Accessibility Features. Adobe Systems Incorporated. https://helpx.adobe.com/acrobat/using/create-verify-pdf-accessibility.html.

Metadata

Authors
Mowar, P., Steinfeld, A., & Bigham, J. P.
Affiliations
Carnegie Mellon University (US)
Venue
Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems
Publisher
ACM
Published
2026-04-13
Pages
1-17
OpenAlex
W7154067339
Cited by
2
References
34
Citations by year
2026: 2
FWCI
17.47
Type
conference paper
Language
en
Open access
closed
Locations
CHI Conference on Human Factors in Computing Systems (publishedVersion)
Topics
  • Social Sciences › Human Factors and Ergonomics › Digital Accessibility for Disabilities
  • Social Sciences › Business, Management and Accounting › Management Information Systems › Information Systems and Technology Applications
  • Physical Sciences › Computer Science › Computational Theory and Mathematics › Mathematics, Computing, and Information Processing
Keywords
PDF accessibility, document tagging, assistive technology

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-25, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-25 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: iTagPDF: Towards Finally Automating PDF Accessibility. https://app.cochat.ai/papers/10.1145/3772318.3790289
The paper
Mowar, P., Steinfeld, A., & Bigham, J. P. (2026). iTagPDF: Towards Finally Automating PDF Accessibility. Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems. https://doi.org/10.1145/3772318.3790289

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.