# Getting PDFs for DefExtra

## Overview

You must supply PDFs yourself. The dataset does **not** ship any copyrighted excerpts.

See [`README.md`](../README.md) for the high‑level overview and examples.

## Recommended sources

- **Semantic Scholar**: most `paper_id` values are Semantic Scholar IDs. You can open
  `https://www.semanticscholar.org/paper/<paper_id>` and download the PDF when available.
- **DOI landing pages**: use `https://doi.org/<paper_doi>` to locate the publisher PDF.
- **arXiv**: use `https://arxiv.org/abs/<paper_arxiv>`.
- **ACL Anthology**: if a `paper_id` looks like `2024.lrec-main.952`, use
  `https://aclanthology.org/2024.lrec-main.952`.

## Helper script

Generate a CSV of required PDFs and links (pre-generated [in the repo](../defextra_required_pdfs.csv)):

```bash
uv run python scripts/list_defextra_pdfs.py \
  --legal-csv data/defextra_legal.csv \
  --output-csv defextra_required_pdfs.csv \
  --output-md defextra_required_pdfs.md
```

The output includes:

- `preferred_pdf_name` (the filename we recommend you use)
- `alias_pdf_names` (acceptable filename aliases)
- `url_semanticscholar`, `url_doi`, `url_arxiv`, `url_acl`

## File naming

Place PDFs in one folder and name them as `<paper_id>.pdf` when possible.
DOI/arXiv/PII aliases are supported (see `alias_pdf_names` in the helper output).