Site Accessibility:

Research Leadership

Prof. Dr. Marcelo Akira Inuzuka

Principal Investigator Institute of Informatics — Federal University of Goiás
Role in the project
Principal Investigator and scientific coordinator of Acessília. Sets the research agenda, supervises the team and is responsible for the project's scientific results.
Institution
Federal University of Goiás (Universidade Federal de Goiás, UFG), Brazil
Academic unit
Institute of Informatics (INF/UFG) — faculty member
Research areas
Artificial Intelligence, Natural Language Processing, Document Understanding and Digital Accessibility
Key publication
First author of Acessilia at Dr.DocBench 2026: Agentic Document Parsing through Canonical Tree Adjudication (ACL Anthology, 2026)

Research Team

Scholarship holders and collaborators with documented participation in the project's scientific output. Affiliations as stated in the paper published in the ACL Anthology.

Students and volunteers from the A11yDevs university outreach program (INF/UFG) also contribute to software development. Members are listed individually on this page only with their consent and with documented participation.

Research Activities

  • Multimodal Document Understanding

    Joint interpretation of text, layout, tables, formulas and images in complex academic documents.

  • Agentic Artificial Intelligence

    Agents that combine tools and vision-language models to review and resolve disagreements between document parsers.

  • Semantic Document Parsing

    Representing documents as canonical trees, aligning alternative interpretations and adjudicating structural disagreements.

  • Accessible Educational Materials

    Converting course notes, textbooks and papers into formats accessible to blind and low-vision students.

  • Document Accessibility and Assistive Technologies

    Producing screen-reader-friendly outputs such as semantic HTML, PDF/UA and narrated audio.

Research Outputs

Published paper

Marcelo Akira Inuzuka, Marcos Paulo C. M. Queiroz, Jhonata Fernandes Cordeiro, and Eduardo Fraga Pereira. 2026. Acessilia at Dr.DocBench 2026: Agentic Document Parsing through Canonical Tree Adjudication. In The Proceedings of Workshop on Document Intelligence and Understanding, pages 448–459, Budapest, Hungary. Association for Computational Linguistics.

Official results on the Dr.DocBench hidden test set (509 pages, EvalAI evaluation)
System Text Reading order Tables (TEDS) Formulas (CDM) Overall
Baseline (MinerU only) 67.65 53.44 47.36 0.00 58.09
Acessília (final submission) 74.67 72.30 58.31 0.22 70.37

Source: Table 1 in Inuzuka et al. (2026). A 12.28-point overall gain over the baseline. Formula recognition (CDM) remains a limitation acknowledged in the paper.

Open-source research software

Benchmarks and datasets

Institutional Affiliations and Verification

Host institution

The project is carried out at the Institute of Informatics (opens in a new window) of the Federal University of Goiás (opens in a new window), Goiânia, Brazil. Acessília is a research project; it is not a formal laboratory or administrative unit of UFG.

Infrastructure and institutional support

As acknowledged in the published paper:

  • Laboratório Multiusuário de Computação de Alto Desempenho (LaMCAD/UFG), the multi-user high-performance computing laboratory — computing infrastructure.
  • Centro de Excelência em Inteligência Artificial (CEIA/UFG), the Center of Excellence in Artificial Intelligence — computing infrastructure.
  • Advanced Knowledge Center for Immersive Technologies (AKCIT) — institutional support.
  • Fundação de Amparo à Pesquisa do Estado de Goiás (FAPEG), the Goiás State Research Foundation — research and innovation support.

How to verify this information