tools/README.md

tools/ Directory

Quick reference for the Universal Dependencies dataset generation pipeline.

Quick Start: Add New UD Version (Release Run)

# Set version
export NEW_VER=2.18
cd tools
# Optional convenience (all scripts also support explicit version flags)
# echo "UD_VER=${NEW_VER}" > .env

# Fetch metadata
uv run 00_fetch_ud_clarin-dspace_metadata.py -o --ud-ver "${NEW_VER}"
./00_fetch_ud_codes_and_flags.sh -o --ud-ver "${NEW_VER}"
./01_fetch_ud_snapshot.sh --ud-ver "${NEW_VER}"

# Generate files
uv run 02_generate_metadata.py -o --ud-ver "${NEW_VER}"
uv run 03_generate_README.py -o --ud-ver "${NEW_VER}"

# Keep root metadata/card in sync before parquet generation
cp README-${NEW_VER} ../README.md
cp metadata-${NEW_VER}.json ../metadata.json

# Generate and fully validate parquet
uv run 04_generate_parquet.py --ud-ver "${NEW_VER}" --prune-extra --check-extra
uv run 05_validate_parquet.py --ud-ver "${NEW_VER}" --local

# Stage generated outputs
git add ../README.md ../metadata.json ../parquet
git add README-${NEW_VER} metadata-${NEW_VER}.json
git add etc/citation-${NEW_VER} etc/description-${NEW_VER} etc/codes_and_flags-${NEW_VER}.yaml

For detailed instructions, see: ADDING_NEW_UD_VERSION.md

Quick Start: Refresh Existing UD Branches for a Framework Release

Use this workflow when the loader/framework changes (for example v2.1.1) and you need to regenerate artifacts in all UD version branches.

# Run from repository root
export FRAMEWORK_VER=v2.1.1

for UD_VER in 2.7 2.8 2.9 2.10 2.11 2.12 2.13 2.14 2.15 2.16 2.17; do
  git switch "${UD_VER}"
  # Keep branch aligned with framework changes (pick one)
  git rebase main
  # git merge main --strategy-option=theirs

  cd tools
  ./00_fetch_ud_clarin-dspace_metadata.py -o --ud-ver "${UD_VER}"
  ./00_fetch_ud_codes_and_flags.sh -o --ud-ver "${UD_VER}"
  ./01_fetch_ud_snapshot.sh --ud-ver "${UD_VER}"
  uv run 02_generate_metadata.py -o --ud-ver "${UD_VER}"
  uv run 03_generate_README.py -o --ud-ver "${UD_VER}"
  cp README-${UD_VER} ../README.md
  cp metadata-${UD_VER}.json ../metadata.json
  uv run 04_generate_parquet.py --ud-ver "${UD_VER}" --overwrite --prune-extra --check-extra
  uv run 05_validate_parquet.py --ud-ver "${UD_VER}" --local
  cd ..

  git add README.md metadata.json parquet
  git add tools/README-${UD_VER} tools/metadata-${UD_VER}.json
  git add tools/etc/citation-${UD_VER} tools/etc/description-${UD_VER} tools/etc/codes_and_flags-${UD_VER}.yaml
  git commit -m "Rebuild UD ${UD_VER} artifacts with loader ${FRAMEWORK_VER}"
  git tag -a "ud${UD_VER}-loader-${FRAMEWORK_VER}" -m "UD ${UD_VER} loader ${FRAMEWORK_VER}"
done

Pipeline Scripts

00 - Fetch Metadata

00_fetch_ud_clarin-dspace_metadata.py

• Downloads citation and description from LINDAT/CLARIN repository
• Creates: etc/citation-{VER}, etc/description-{VER}
• Uses: etc/ud_release_handles.tsv for version -> handle mapping
• Optional filter: --ud-ver {VER} to fetch one version

00_fetch_ud_codes_and_flags.sh

• Downloads language codes and flags from UD docs-automation
• Default output: etc/codes_and_flags-{UD_VER}.yaml, etc/codes_and_flags-latest.yaml
• --all output: all mapped etc/codes_and_flags-{VER}.yaml files
• Update VER_MAPPING for new versions

01 - Fetch Snapshot

01_fetch_ud_snapshot.sh

• Downloads official UD release snapshot from LINDAT/CLARIN (archive resolved via API)
• Creates: ud-treebanks-v{VER}/
• Creates/updates symlink: UD_repos -> ud-treebanks-v{VER}
• Options: --ud-ver, --online, --keep-archive, --keep-backup

02 - Generate Metadata

02_generate_metadata.py

• Extracts metadata from local UD directories
• Collects: summaries, licenses, splits, statistics, blocked status
• Creates: metadata-{VER}.json
• Reads: blocked_treebanks.yaml for license restrictions
• Options: --ud-ver, --override

03 - Generate README

03_generate_README.py

• Renders Jinja2 template with metadata
• Creates: README-{VER} (HuggingFace dataset card)
• Uses: templates/README.tmpl, metadata-{VER}.json, etc/citation-{VER}, etc/description-{VER}
• Options: --ud-ver, --override

04 - Generate Parquet

04_generate_parquet.py

• Wrapper script calling ud-hf-parquet-tools library
• Converts CoNLL-U files to Parquet format
• Creates: ../parquet/{treebank}/{split}.parquet
• Uses tools/metadata-{UD_VER}.json when present; otherwise falls back to ../metadata.json
• Options: --ud-ver, --test, --overwrite, --blocked-treebanks, --check-extra, --prune-extra
• Recommended for release runs: --prune-extra --check-extra to prevent stale parquet files when treebanks are renamed/removed across UD versions

05 - Validate Parquet

05_validate_parquet.py

• Wrapper script calling ud-hf-parquet-tools library
• Validates Parquet files against CoNLL-U source
• Uses tools/metadata-{UD_VER}.json when present; otherwise falls back to ../metadata.json
• Options: --ud-ver, --local, --parquet-dir, --revision, --test, -vv

Directory Structure

tools/
├── 00_fetch_ud_clarin-dspace_metadata.py   # Fetch citation/description
├── 00_fetch_ud_codes_and_flags.sh          # Fetch language metadata
├── 01_fetch_ud_snapshot.sh                 # Download UD snapshot from CLARIN
├── 02_generate_metadata.py                 # Extract treebank metadata
├── 03_generate_README.py                   # Generate dataset card
├── 04_generate_parquet.py                  # Generate Parquet files
├── 05_validate_parquet.py                  # Validate Parquet files
├── blocked_treebanks.yaml                  # License-restricted treebanks
├── etc/
│   ├── citation-{VER}                      # Generated citations
│   ├── description-{VER}                   # Generated descriptions
│   ├── codes_and_flags-{VER}.yaml          # Language metadata
│   └── ud_release_handles.tsv              # UD version -> CLARIN handle mapping
├── templates/
│   └── README.tmpl                         # Jinja2 template for dataset card
├── metadata-{VER}.json                     # Generated metadata (output)
├── README-{VER}                            # Generated dataset card (output)
├── ud-treebanks-v{VER}/                    # Extracted CLARIN snapshot
└── UD_repos -> ud-treebanks-v{VER}         # Symlink consumed by 02/04/05 scripts

Configuration

Environment Variables (.env, optional)

UD_VER=2.17  # Current UD version

You can skip .env and pass --ud-ver flags directly.

Blocked Treebanks (blocked_treebanks.yaml)

pt_cintil:
  reason: "Restrictive license prohibits redistribution in derived formats"
  license: "CC BY-NC-SA 4.0"
  url: "https://github.com/UniversalDependencies/UD_Portuguese-CINTIL"

Dependencies

• Python 3.12+
• uv (for running scripts with dependencies)
• ud-hf-parquet-tools (for parquet generation/validation)

Install dependencies:

pip install ud-hf-parquet-tools pyyaml python-dotenv jinja2

Common Operations

Test on subset:

uv run ./04_generate_parquet.py --ud-ver 2.17 --test --prune-extra --check-extra
uv run ./05_validate_parquet.py --ud-ver 2.17 --test --local

Full local validation (release run):

uv run ./05_validate_parquet.py --ud-ver 2.17 --local

Force regeneration:

./02_generate_metadata.py --ud-ver 2.17 -o
./03_generate_README.py --ud-ver 2.17 -o
uv run ./04_generate_parquet.py --ud-ver 2.17 --overwrite --prune-extra --check-extra

Skip blocked treebanks:

uv run ./04_generate_parquet.py --blocked-treebanks blocked_treebanks.yaml

Troubleshooting

Script not found:

• Scripts use #!/usr/bin/env -S uv run shebang
• Make executable: chmod +x *.py *.sh
• Or run with: uv run ./script.py

Missing dependencies:

pip install ud-hf-parquet-tools

Snapshot download/extraction fails:

./01_fetch_ud_snapshot.sh --online --ud-ver 2.17

Disk usage after repeated snapshot refreshes:

# Default behavior deletes archive and replaces old snapshot directory
./01_fetch_ud_snapshot.sh --online --ud-ver 2.17

# Keep archive and timestamped backup only when needed
./01_fetch_ud_snapshot.sh --online --ud-ver 2.17 --keep-archive --keep-backup

Documentation

• Comprehensive guide: ADDING_NEW_UD_VERSION.md
• Blocked treebanks: BLOCKED_TREEBANKS.md
• Migration guide: MIGRATION.md
• Parquet tools: https://github.com/bot-zen/ud-hf-parquet-tools