GROBID

Run GROBID as a local Docker service and convert PDFs into TEI XML.

ansible-playbook -i inventory -K playbooks/lab-stack.yml --tags grobid

🔧 Manual setup and configuration

  • Make sure Docker is installed and your user can run Docker commands.
  • Log out and log back in (or restart Nautilus/Nemo) so right-click actions become visible.
  • In Nautilus, right-click a PDF and use Convert PDF to TEI (grobid).
  • The action calls /usr/local/bin/pdf-to-tei and writes *.tei.xml next to each PDF.
  • The helper script sends PDFs to http://127.0.0.1:8070/api/processFulltextDocument.

✅ Check

docker ps --filter "name=grobid" --format "table {{.Names}}\t{{.Status}}\t{{.Ports}}"

Expected: container grobid is up and exposes port 8070.

curl -s -o /dev/null -w "%{http_code}\n" http://127.0.0.1:8070

Expected: an HTTP response (commonly 200 or 404, but not connection refused).

ls -l /usr/local/bin/pdf-to-tei
ls -l ~/.local/share/file-manager/actions/grobid-pdf-to-tei.desktop
ls -l ~/.local/share/nemo/actions/grobid-pdf-to-tei.nemo_action

Expected: helper script exists and at least one action file exists.

Nautilus right-click action behavior

  • Menu entry: Convert PDF to TEI (grobid)
  • File type: appears for PDF files (application/pdf)
  • Execution: runs /usr/local/bin/pdf-to-tei "%F"
  • Output naming: input article.pdf becomes article.tei.xml
  • Batch behavior: multiple selected PDFs are processed one-by-one
  • Timeout behavior: request timeout is based on role variable grobid_nfs_timeout_ms

Rationale

GROBID is the best performing open-source tools (used in many tools like ResearchGate, Mendeley). In particular, it supports the parsing of academic references and output like BibTeX. Reference parsing is still a shortcoming of docling and marker.

Tkaczyk, D., Collins, A., Sheridan, P., & Beel, J. (2018, May). Machine learning vs. rules and out-of-the-box vs. retrained: An evaluation of open-source bibliographic reference and citation parsers. In Proceedings of the 18th ACM/IEEE on joint conference on digital libraries (pp. 99-108).