iiegn commited on
Commit
130e2d2
·
1 Parent(s): ab26265

Update dataset card and migration docs

Browse files

Use the official universal-dependencies/universal_dependencies Hub namespace in active examples, add former-location and v2.0 rewrite notes, and document the pending UPOS/HEAD schema change.

Refs: https://huggingface.co/datasets/universal-dependencies/universal_dependencies/discussions/2

Refs: https://huggingface.co/datasets/universal-dependencies/universal_dependencies/discussions/3

ADDING_NEW_UD_VERSION.md CHANGED
@@ -1,6 +1,6 @@
1
  # Adding a New Universal Dependencies Version
2
 
3
- This guide explains how to add a new Universal Dependencies release (e.g., UD 2.18, 2.19, etc.) to the `commul/universal_dependencies` HuggingFace dataset.
4
 
5
  **Quick reference:** See [tools/README.md](tools/README.md) for concise commands and script documentation.
6
 
@@ -12,7 +12,7 @@ This guide explains how to add a new Universal Dependencies release (e.g., UD 2.
12
  ```bash
13
  pip install ud-hf-parquet-tools pyyaml python-dotenv jinja2
14
  ```
15
- - Access to push to `commul/universal_dependencies` on HuggingFace Hub
16
  - `huggingface-cli` installed and authenticated:
17
  ```bash
18
  pip install huggingface_hub
@@ -361,7 +361,7 @@ If you've cloned the HuggingFace repository with git-lfs:
361
 
362
  ```bash
363
  # Add HF Hub as remote (if not already)
364
- git remote add hf https://huggingface.co/datasets/commul/universal_dependencies
365
 
366
  # Push to HuggingFace
367
  git push hf ${NEW_VER}
@@ -372,7 +372,7 @@ git push hf --tags
372
 
373
  ```bash
374
  # Upload entire directory
375
- huggingface-cli upload commul/universal_dependencies . --repo-type dataset --revision ${NEW_VER}
376
  ```
377
 
378
  **Expected upload time:** 2-6 hours depending on network speed and HuggingFace server load.
@@ -381,7 +381,7 @@ huggingface-cli upload commul/universal_dependencies . --repo-type dataset --rev
381
 
382
  ### 17. Verify on HuggingFace Hub
383
 
384
- Visit: https://huggingface.co/datasets/commul/universal_dependencies
385
 
386
  **Checklist:**
387
  1. ✅ Branch `2.18` exists in the "Branches" dropdown
@@ -397,7 +397,7 @@ Visit: https://huggingface.co/datasets/commul/universal_dependencies
397
  from datasets import load_dataset
398
 
399
  # Load from new version
400
- ds = load_dataset("commul/universal_dependencies", "en_ewt", revision="2.18")
401
  print(ds)
402
  ```
403
 
@@ -427,14 +427,14 @@ This makes the new version the default when users don't specify a revision.
427
 
428
  ### 19. Refresh Existing UD Branches for a Framework Release
429
 
430
- Use this when framework logic changes (for example loader `v2.1.1`) and you
431
  need to rebuild all existing UD branches (`2.7` ... `2.17`).
432
 
433
  ```bash
434
  # Repository root
435
- export FRAMEWORK_VER=v2.1.1
436
 
437
- 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
438
  git switch "${UD_VER}"
439
  # Keep the branch aligned with framework changes (pick one):
440
  git rebase main
 
1
  # Adding a New Universal Dependencies Version
2
 
3
+ This guide explains how to add a new Universal Dependencies release (e.g., UD 2.18, 2.19, etc.) to the `universal-dependencies/universal_dependencies` HuggingFace dataset.
4
 
5
  **Quick reference:** See [tools/README.md](tools/README.md) for concise commands and script documentation.
6
 
 
12
  ```bash
13
  pip install ud-hf-parquet-tools pyyaml python-dotenv jinja2
14
  ```
15
+ - Access to push to `universal-dependencies/universal_dependencies` on HuggingFace Hub
16
  - `huggingface-cli` installed and authenticated:
17
  ```bash
18
  pip install huggingface_hub
 
361
 
362
  ```bash
363
  # Add HF Hub as remote (if not already)
364
+ git remote add hf https://huggingface.co/datasets/universal-dependencies/universal_dependencies
365
 
366
  # Push to HuggingFace
367
  git push hf ${NEW_VER}
 
372
 
373
  ```bash
374
  # Upload entire directory
375
+ huggingface-cli upload universal-dependencies/universal_dependencies . --repo-type dataset --revision ${NEW_VER}
376
  ```
377
 
378
  **Expected upload time:** 2-6 hours depending on network speed and HuggingFace server load.
 
381
 
382
  ### 17. Verify on HuggingFace Hub
383
 
384
+ Visit: https://huggingface.co/datasets/universal-dependencies/universal_dependencies
385
 
386
  **Checklist:**
387
  1. ✅ Branch `2.18` exists in the "Branches" dropdown
 
397
  from datasets import load_dataset
398
 
399
  # Load from new version
400
+ ds = load_dataset("universal-dependencies/universal_dependencies", "en_ewt", revision="2.18")
401
  print(ds)
402
  ```
403
 
 
427
 
428
  ### 19. Refresh Existing UD Branches for a Framework Release
429
 
430
+ Use this when framework logic changes and you
431
  need to rebuild all existing UD branches (`2.7` ... `2.17`).
432
 
433
  ```bash
434
  # Repository root
435
+ export FRAMEWORK_VER=v2.2.1
436
 
437
+ 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 2.18; do
438
  git switch "${UD_VER}"
439
  # Keep the branch aligned with framework changes (pick one):
440
  git rebase main
CHANGELOG.md CHANGED
@@ -5,6 +5,24 @@ All notable changes to the Universal Dependencies HuggingFace dataset loader wil
5
  The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
6
  and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8
  ## [2.2.1] - 2026-05-29
9
 
10
  ### Added
 
5
  The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
6
  and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
7
 
8
+ ## [Unreleased]
9
+
10
+ ### Changed
11
+
12
+ - **Parquet Schema**
13
+ - Regular-token `upos` values are now stored as strings instead of HuggingFace `ClassLabel` indices.
14
+ - Regular-token `head` values are now stored as 16-bit integers instead of strings.
15
+ - Empty-node `head` values remain strings because CoNLL-U empty nodes may use `_`.
16
+
17
+ - **Dataset Card**
18
+ - Updated examples and links for the official Hugging Face namespace: `universal-dependencies/universal_dependencies`.
19
+ - Aligned generated card metadata with the Hub-side cleanup: `license: other`, `size_categories: 1M<n<10M`, and no empty leaderboard/contact fields.
20
+ - Added clearer citation guidance and a short note about the former `commul/universal_dependencies` location and the v2.0 Parquet rewrite.
21
+
22
+ - **Migration Documentation**
23
+ - Reduced old v1.x-to-v2.0 migration notes to an archived summary.
24
+ - Added current migration notes for the `upos` and `head` schema change.
25
+
26
  ## [2.2.1] - 2026-05-29
27
 
28
  ### Added
MIGRATION.md CHANGED
@@ -1,483 +1,112 @@
1
- # Migration Guide: v1.x → v2.0
2
 
3
- This guide helps you migrate from Universal Dependencies dataset loader v1.x to v2.0.
 
 
4
 
5
- ## What's New in v2.0
6
 
7
- **Architecture Changes:**
8
- - **Parquet format**: Native support with datasets >=4.0.0 (5-10x faster loading)
9
- - **No Python script**: Dataset no longer requires `trust_remote_code=True`
10
- - **External helper library**: CoNLL-U processing utilities moved to [`ud-hf-parquet-tools`](https://github.com/bot-zen/ud-hf-parquet-tools)
11
 
12
- **Data Changes:**
13
- - **MWT bug fix**: Token sequences now correctly exclude Multi-Word Token surface forms
14
- - **MWT field added**: New structured `mwt` field with Multi-Word Token information
15
- - **Enhanced metadata**: Includes `num_fused` (MWT counts) in statistics
 
16
 
17
- ## Quick Start
18
 
19
- ### For Users with datasets >=4.0.0
20
-
21
- No code changes needed! Parquet files load automatically:
22
 
23
  ```python
24
  from datasets import load_dataset
25
 
26
- # v2.0: Works seamlessly with datasets >=4.0.0
27
- dataset = load_dataset("commul/universal_dependencies", "fr_gsd")
28
- # Automatically uses Parquet format (fast, secure)
29
- ```
30
-
31
- ### For Users with datasets <4.0.0
32
-
33
- **Option 1: Upgrade datasets (Recommended)**
34
-
35
- ```bash
36
- pip install --upgrade "datasets>=4.0.0"
37
- ```
38
-
39
- **Option 2: Continue using v1.x**
40
-
41
- ```python
42
- # v1.x: Requires trust_remote_code
43
- dataset = load_dataset("commul/universal_dependencies", "fr_gsd", trust_remote_code=True, revision="v1.0")
44
- ```
45
-
46
- ## Breaking Changes
47
-
48
- ### 1. Token Sequences Now Exclude MWT Forms
49
-
50
- **Impact:** Token counts and sequences have changed for treebanks with Multi-Word Tokens (MWTs).
51
-
52
- **What Changed:**
53
- - v1.x incorrectly included MWT surface forms in token sequences
54
- - v2.0 correctly excludes them, matching UD guidelines
55
-
56
- **Example (French "des" → "de" + "les"):**
57
-
58
- ```python
59
- # v1.x (BUGGY):
60
- {
61
- "tokens": ["Elle", "des", "de", "les", "pommes", "."], # WRONG: "des" included
62
- "lemmas": ["elle", "_", "de", "le", "pomme", "."],
63
- "upos": ["PRON", "_", "ADP", "DET", "NOUN", "PUNCT"],
64
- }
65
-
66
- # v2.0 (CORRECT):
67
- {
68
- "tokens": ["Elle", "de", "les", "pommes", "."], # CORRECT: only syntactic words
69
- "lemmas": ["elle", "de", "le", "pomme", "."],
70
- "upos": ["PRON", "ADP", "DET", "NOUN", "PUNCT"],
71
- "mwt": [{"id": "2-3", "form": "des", "misc": ""}], # MWT info preserved
72
- }
73
  ```
74
 
75
- **Affected Treebanks (50+):**
76
-
77
- Languages with common MWTs include:
78
- - **French** (fr_*): du, au, des, aux (~2-5% of tokens)
79
- - **Italian** (it_*): del, della, nel, alla (~1-3%)
80
- - **Portuguese** (pt_*): do, da, no, pelo (~2-4%)
81
- - **Spanish** (es_*): del, al (~0.5-1%)
82
- - **Arabic** (ar_*): various clitics (~1-2%)
83
- - **German** (de_*): zum, vom, am (~0.1-0.5%)
84
- - **Catalan** (ca_*): del, al, pels (~1-2%)
85
- - **Indonesian** (id_*): reduplications (~0.1%)
86
-
87
- **Action Required:**
88
-
89
- If your code assumes specific token counts or positions:
90
 
91
  ```python
92
- # v1.x code that might break:
93
- def get_third_token(example):
94
- return example["tokens"][2] # May return different token in v2.0
95
-
96
- # Migration fix:
97
- def get_third_syntactic_word(example):
98
- # v2.0: This is now correct - gets the 3rd syntactic word
99
- return example["tokens"][2]
100
-
101
- def get_third_surface_token(example):
102
- # v2.0: If you need surface forms, reconstruct from MWTs
103
- tokens = example["tokens"][:]
104
- mwts = example["mwt"]
105
-
106
- # Insert MWT forms at appropriate positions
107
- for mwt in reversed(mwts): # Process in reverse to maintain indices
108
- start, end = map(int, mwt["id"].split("-"))
109
- tokens[start-1:end] = [mwt["form"]]
110
-
111
- return tokens[2]
112
  ```
113
 
114
- ### 2. New Schema Field: `mwt`
115
-
116
- **Impact:** Dataset schema now includes an `mwt` field.
117
-
118
- **What Changed:**
119
- - Added: `mwt` field containing structured MWT information
120
- - Schema: `[{"id": "1-2", "form": "surface_form", "misc": "metadata"}]`
121
- - Empty list for treebanks without MWTs
122
-
123
- **Example Usage:**
124
 
125
  ```python
126
- from datasets import load_dataset
127
-
128
- dataset = load_dataset("commul/universal_dependencies", "fr_gsd", split="train")
129
-
130
- # Access MWT information
131
- for example in dataset:
132
- if example["mwt"]: # Has MWTs
133
- for mwt in example["mwt"]:
134
- print(f"MWT {mwt['id']}: {mwt['form']}")
135
- # Extract range
136
- start, end = map(int, mwt["id"].split("-"))
137
- syntactic_words = example["tokens"][start-1:end]
138
- print(f" → {' + '.join(syntactic_words)}")
139
-
140
- # Output example:
141
- # MWT 2-3: des
142
- # → de + les
143
  ```
144
 
145
- **Research Use Cases:**
 
146
 
147
  ```python
148
- # Count MWTs per treebank
149
- def count_mwts(dataset):
150
- return sum(len(ex["mwt"]) for ex in dataset)
151
-
152
- # Analyze MWT patterns
153
- def analyze_mwt_patterns(dataset):
154
- patterns = {}
155
- for ex in dataset:
156
- for mwt in ex["mwt"]:
157
- form = mwt["form"]
158
- patterns[form] = patterns.get(form, 0) + 1
159
- return patterns
160
 
161
- fr_gsd = load_dataset("commul/universal_dependencies", "fr_gsd", split="train")
162
- print(analyze_mwt_patterns(fr_gsd))
163
- # Output: {'des': 1234, 'du': 987, 'au': 654, 'aux': 321, ...}
164
  ```
165
 
166
- ### 3. Requires datasets >=4.0.0
167
-
168
- **Impact:** The Python script loader is deprecated (datasets >=4.0.0 policy).
169
 
170
- **What Changed:**
171
- - v1.x: Uses Python script with `trust_remote_code=True`
172
- - v2.0: Uses Parquet format (no remote code execution)
173
 
174
- **Security Benefit:**
175
- - No arbitrary code execution from dataset loading
176
- - Parquet files are data-only, sandboxed
177
- - Aligns with HuggingFace security policies
178
 
179
- **Migration:**
 
 
180
 
181
- ```bash
182
- # Check your datasets version
183
- python -c "import datasets; print(datasets.__version__)"
184
 
185
- # Upgrade if needed
186
- pip install --upgrade "datasets>=4.0.0"
187
- ```
 
188
 
189
- If you cannot upgrade datasets:
190
 
191
  ```python
192
- # Use v1.x with revision pinning
193
  dataset = load_dataset(
194
- "commul/universal_dependencies",
195
  "fr_gsd",
196
  trust_remote_code=True,
197
- revision="v1.0" # Pin to v1.x
198
  )
199
  ```
200
 
201
- ## Helper Functions Moved to External Library
202
-
203
- **Important:** Helper functions for CoNLL-U processing are now in a separate package.
204
-
205
- ### What Moved
206
-
207
- The following functions are **no longer part of the dataset**:
208
- - `parse_feats()`, `parse_misc()`, `parse_deps()` - Parse CoNLL-U field strings
209
- - `write_conllu()`, `example_to_conllu()` - Export data to CoNLL-U format
210
- - Various internal conversion utilities
211
-
212
- ### How to Access Helper Functions
213
-
214
- If you need CoNLL-U processing utilities, install the external library:
215
-
216
- ```bash
217
- pip install ud-hf-parquet-tools
218
- ```
219
-
220
- Then import from the package:
221
-
222
- ```python
223
- from datasets import load_dataset
224
- from ud_hf_parquet_tools import parse_feats, parse_misc, write_conllu
225
-
226
- # Load dataset
227
- ds = load_dataset("commul/universal_dependencies", "en_ewt", split="train")
228
-
229
- # Parse optional fields
230
- sentence = ds[0]
231
- for i, token in enumerate(sentence['tokens']):
232
- feats = parse_feats(sentence['feats'][i]) # Returns dict or {}
233
- misc = parse_misc(sentence['misc'][i]) # Returns dict or {}
234
- print(f"{token}: UPOS={sentence['upos'][i]}, feats={feats}, misc={misc}")
235
-
236
- # Export back to CoNLL-U format
237
- write_conllu(ds, "output.conllu")
238
- ```
239
-
240
- **Library Documentation:** https://github.com/bot-zen/ud-hf-parquet-tools
241
-
242
- ### If You Don't Need Helper Functions
243
-
244
- Most users only need the dataset itself and can work directly with the fields:
245
-
246
- ```python
247
- from datasets import load_dataset
248
-
249
- ds = load_dataset("commul/universal_dependencies", "en_ewt", split="train")
250
-
251
- # Access data directly
252
- sentence = ds[0]
253
- print(f"Tokens: {sentence['tokens']}")
254
- print(f"POS tags: {sentence['upos']}")
255
- print(f"Dependencies: {sentence['deprel']}")
256
-
257
- # FEATS and MISC are strings in CoNLL-U format
258
- print(f"Features (raw): {sentence['feats'][0]}") # e.g., "Case=Nom|Number=Sing"
259
- print(f"Misc (raw): {sentence['misc'][0]}") # e.g., "SpaceAfter=No"
260
-
261
- # Parse manually if needed (simple cases)
262
- feats_str = sentence['feats'][0]
263
- if feats_str:
264
- feats_dict = dict(kv.split('=') for kv in feats_str.split('|'))
265
- print(f"Features (parsed): {feats_dict}")
266
- ```
267
-
268
- ## New Features in v2.0
269
-
270
- ### 1. Parquet Format (5-10x Faster Loading)
271
-
272
- ```python
273
- # v1.x: Downloads CoNLL-U, parses on-the-fly (~10-30 seconds)
274
- dataset = load_dataset("commul/universal_dependencies", "fr_gsd", trust_remote_code=True)
275
-
276
- # v2.0: Loads pre-processed Parquet (~1-3 seconds)
277
- dataset = load_dataset("commul/universal_dependencies", "fr_gsd")
278
- ```
279
-
280
- **Benefits:**
281
- - Much faster loading (especially for large treebanks)
282
- - Lower memory usage
283
- - Better compression
284
- - Native support in datasets >=4.0.0
285
-
286
- ### 2. Multi-Word Token (MWT) Information
287
-
288
- ```python
289
- dataset = load_dataset("commul/universal_dependencies", "fr_gsd", split="train")
290
-
291
- # Find sentences with MWTs
292
- sentences_with_mwts = [ex for ex in dataset if ex["mwt"]]
293
- print(f"Sentences with MWTs: {len(sentences_with_mwts)}/{len(dataset)}")
294
-
295
- # Analyze MWT complexity
296
- complex_mwts = [ex for ex in dataset if any(
297
- int(mwt["id"].split("-")[1]) - int(mwt["id"].split("-")[0]) > 2
298
- for mwt in ex["mwt"]
299
- )]
300
- print(f"Sentences with 3+ word MWTs: {len(complex_mwts)}")
301
- ```
302
-
303
- ### 3. Enhanced Metadata
304
-
305
- ```python
306
- # Load dataset info
307
- from datasets import load_dataset_builder
308
-
309
- builder = load_dataset_builder("commul/universal_dependencies", "fr_gsd")
310
- info = builder.info
311
-
312
- # Now includes MWT statistics
313
- print(info.description) # Contains num_fused counts
314
- ```
315
-
316
- ## Verification Steps
317
-
318
- ### 1. Verify Token Counts Match UD Stats
319
-
320
- ```python
321
- from datasets import load_dataset
322
- import json
323
-
324
- # Load dataset and metadata
325
- dataset = load_dataset("commul/universal_dependencies", "fr_gsd", split="train")
326
-
327
- # Count syntactic words
328
- word_count = sum(len(ex["tokens"]) for ex in dataset)
329
-
330
- # Load metadata (if available)
331
- with open("metadata.json") as f:
332
- metadata = json.load(f)
333
-
334
- expected_words = int(metadata["fr_gsd"]["splits"]["train"]["num_words"])
335
- print(f"Dataset words: {word_count}")
336
- print(f"Expected words: {expected_words}")
337
- print(f"Match: {word_count == expected_words}")
338
-
339
- # This should be True in v2.0 (was False in v1.x for MWT treebanks)
340
- ```
341
-
342
- ### 2. Verify MWT Extraction
343
-
344
- ```python
345
- # Count MWTs
346
- mwt_count = sum(len(ex["mwt"]) for ex in dataset)
347
-
348
- expected_mwts = int(metadata["fr_gsd"]["splits"]["train"]["num_fused"])
349
- print(f"Dataset MWTs: {mwt_count}")
350
- print(f"Expected MWTs: {expected_mwts}")
351
- print(f"Match: {mwt_count == expected_mwts}")
352
- ```
353
-
354
- ### 3. Compare v1.x vs v2.0 Output
355
-
356
- ```python
357
- # Load both versions (if v1.x still available)
358
- v1 = load_dataset("commul/universal_dependencies", "en_ewt", split="test[:10]", revision="v1.0", trust_remote_code=True)
359
- v2 = load_dataset("commul/universal_dependencies", "en_ewt", split="test[:10]")
360
-
361
- # English-EWT has no MWTs, so should be identical except for new field
362
- for i in range(10):
363
- assert v1[i]["tokens"] == v2[i]["tokens"], f"Example {i} differs"
364
- assert v2[i]["mwt"] == [], f"Example {i} has unexpected MWTs"
365
-
366
- print("✓ English-EWT unchanged (no MWTs)")
367
-
368
- # French-GSD has MWTs, so v2.0 will differ
369
- v1_fr = load_dataset("commul/universal_dependencies", "fr_gsd", split="test[:10]", revision="v1.0", trust_remote_code=True)
370
- v2_fr = load_dataset("commul/universal_dependencies", "fr_gsd", split="test[:10]")
371
-
372
- # v1.x token count includes MWTs (WRONG)
373
- v1_token_count = sum(len(ex["tokens"]) for ex in v1_fr)
374
-
375
- # v2.0 token count excludes MWTs (CORRECT)
376
- v2_token_count = sum(len(ex["tokens"]) for ex in v2_fr)
377
-
378
- print(f"v1.x French token count: {v1_token_count} (includes MWT forms)")
379
- print(f"v2.0 French token count: {v2_token_count} (syntactic words only)")
380
- print(f"Difference: {v1_token_count - v2_token_count} MWT forms removed")
381
- ```
382
-
383
- ## Common Issues
384
-
385
- ### Issue 1: "Dataset script not supported" Error
386
-
387
- **Error:**
388
- ```
389
- RuntimeError: Dataset scripts are no longer supported
390
- ```
391
-
392
- **Cause:** Using datasets >=4.0.0 with v1.x loader
393
-
394
- **Solution:**
395
- ```bash
396
- pip install --upgrade "datasets>=4.0.0"
397
- # Then use v2.0 (Parquet-based)
398
- ```
399
-
400
- ### Issue 2: Token Count Mismatch
401
-
402
- **Issue:** Your code expects specific token counts that changed in v2.0
403
-
404
- **Solution:** Update your code to use `num_words` from metadata instead of `num_tokens`
405
-
406
- ```python
407
- # v1.x: Used num_tokens (WRONG for MWT treebanks)
408
- expected_count = metadata["splits"]["train"]["num_tokens"]
409
-
410
- # v2.0: Use num_words (CORRECT)
411
- expected_count = metadata["splits"]["train"]["num_words"]
412
- ```
413
-
414
- ### Issue 3: MWT Field Not Found (v1.x Code)
415
-
416
- **Issue:** Old code doesn't handle the new `mwt` field
417
-
418
- **Solution:** Gracefully handle the field or upgrade
419
-
420
- ```python
421
- # Backwards compatible code
422
- tokens = example["tokens"]
423
- mwts = example.get("mwt", []) # Empty list if not present
424
- ```
425
-
426
- ### Issue 4: Helper Function Import Errors
427
 
428
- **Error:**
429
  ```python
430
- from universal_dependencies import parse_feats
431
- # ImportError: No module named 'universal_dependencies'
 
 
 
432
  ```
433
 
434
- **Cause:** Helper functions moved to separate library
435
-
436
- **Solution:**
437
- ```bash
438
- # Install the helper library
439
- pip install ud-hf-parquet-tools
440
 
441
- # Update imports
442
- from ud_hf_parquet_tools import parse_feats, parse_misc, write_conllu
443
- ```
444
 
445
- Or work with raw strings directly (see "If You Don't Need Helper Functions" section above).
 
 
 
446
 
447
  ## Support
448
 
449
- If you encounter issues during migration:
450
-
451
- 1. Check the [CHANGELOG.md](CHANGELOG.md) for detailed changes
452
- 2. Review the [README.md](README.md) for updated examples
453
- 3. Helper library documentation: https://github.com/bot-zen/ud-hf-parquet-tools
454
- 4. Report issues at: https://huggingface.co/datasets/commul/universal_dependencies/discussions
455
-
456
- ## Summary
457
-
458
- **Key Takeaways:**
459
-
460
- ✅ **v2.0 is more correct:** Fixes critical MWT bug
461
- ✅ **v2.0 is faster:** Parquet format is 5-10x quicker
462
- ✅ **v2.0 is more secure:** No remote code execution
463
- ✅ **v2.0 adds features:** MWT information now available
464
- ✅ **v2.0 is modular:** Helper functions available as separate library
465
-
466
- **Migration Checklist:**
467
-
468
- - [ ] Upgrade to datasets >=4.0.0
469
- - [ ] Test your code with v2.0 data
470
- - [ ] Update token count expectations (if using MWT treebanks)
471
- - [ ] Update any hard-coded token indices (if applicable)
472
- - [ ] If using helper functions: Install `ud-hf-parquet-tools` and update imports
473
- - [ ] If exporting to CoNLL-U: Use `write_conllu()` from `ud-hf-parquet-tools`
474
- - [ ] Utilize new MWT field for research (optional)
475
-
476
- **Estimated Migration Time:**
477
- - Basic usage: 15-30 minutes
478
- - With helper functions: +10 minutes (install library, update imports)
479
-
480
- **Resources:**
481
- - Dataset repository: https://huggingface.co/datasets/commul/universal_dependencies
482
  - Helper library: https://github.com/bot-zen/ud-hf-parquet-tools
483
- - CHANGELOG: [CHANGELOG.md](CHANGELOG.md)
 
1
+ # Migration Guide
2
 
3
+ This repository uses Parquet-only HuggingFace dataset branches for Universal
4
+ Dependencies releases. Current examples use the official dataset location:
5
+ `universal-dependencies/universal_dependencies`.
6
 
7
+ ## Current Development Version
8
 
9
+ The current development version updates the generated Parquet schema for regular
10
+ CoNLL-U token fields:
 
 
11
 
12
+ - `upos` is now stored as a list of strings instead of HuggingFace `ClassLabel`
13
+ indices.
14
+ - `head` is now stored as a list of 16-bit integers instead of strings.
15
+ - Empty-node `head` values remain strings because CoNLL-U empty nodes may use
16
+ `_` for HEAD.
17
 
18
+ ### What Usually Changes
19
 
20
+ Most users can load and iterate over datasets as before:
 
 
21
 
22
  ```python
23
  from datasets import load_dataset
24
 
25
+ ds = load_dataset(
26
+ "universal-dependencies/universal_dependencies",
27
+ "en_ewt",
28
+ revision="2.18",
29
+ split="train",
30
+ )
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
31
  ```
32
 
33
+ Code that expected `upos` as integer class IDs should now work directly with
34
+ strings:
 
 
 
 
 
 
 
 
 
 
 
 
 
35
 
36
  ```python
37
+ upos = ds[0]["upos"]
38
+ assert upos[0] == "PROPN"
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
39
  ```
40
 
41
+ Code that compared `head` to strings should compare to integers:
 
 
 
 
 
 
 
 
 
42
 
43
  ```python
44
+ heads = ds[0]["head"]
45
+ root_positions = [i for i, head in enumerate(heads) if head == 0]
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
46
  ```
47
 
48
+ If you need string output for CoNLL-U serialization, convert at the boundary or
49
+ use the helper package:
50
 
51
  ```python
52
+ from ud_hf_parquet_tools import write_conllu
 
 
 
 
 
 
 
 
 
 
 
53
 
54
+ write_conllu(ds, "output.conllu")
 
 
55
  ```
56
 
57
+ ### Compatibility Notes
 
 
58
 
59
+ The next `ud-hf-parquet-tools` release can still reconstruct older Parquet
60
+ outputs where `upos` was stored as a `ClassLabel`. New outputs should use the
61
+ updated schema.
62
 
63
+ ## Archived v1.x to v2.0 Notes
 
 
 
64
 
65
+ The v1 loader used a Python dataset script and required `trust_remote_code=True`.
66
+ The v2 line removed the script loader and distributes Parquet files directly,
67
+ which requires `datasets>=4.0.0`.
68
 
69
+ Important v2.0 data changes:
 
 
70
 
71
+ - Multi-word token surface forms are no longer included in regular `tokens`.
72
+ - Multi-word token information is preserved in the structured `mwt` field.
73
+ - Helper functions moved out of the dataset repository and into
74
+ `ud-hf-parquet-tools`.
75
 
76
+ Old v1 code like this:
77
 
78
  ```python
 
79
  dataset = load_dataset(
80
+ "universal-dependencies/universal_dependencies",
81
  "fr_gsd",
82
  trust_remote_code=True,
83
+ revision="v1.0",
84
  )
85
  ```
86
 
87
+ should be replaced with current Parquet loading:
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
88
 
 
89
  ```python
90
+ dataset = load_dataset(
91
+ "universal-dependencies/universal_dependencies",
92
+ "fr_gsd",
93
+ revision="2.18",
94
+ )
95
  ```
96
 
97
+ For token-count checks, use `num_words` from `metadata.json` rather than
98
+ `num_tokens` when comparing against regular syntactic-word sequences.
 
 
 
 
99
 
100
+ ## Repository Move
 
 
101
 
102
+ The dataset was formerly maintained at `commul/universal_dependencies` and now
103
+ lives under the official Universal Dependencies organization at
104
+ `universal-dependencies/universal_dependencies`:
105
+ https://huggingface.co/datasets/universal-dependencies/universal_dependencies
106
 
107
  ## Support
108
 
109
+ - Changelog: [CHANGELOG.md](CHANGELOG.md)
110
+ - Dataset repository: https://huggingface.co/datasets/universal-dependencies/universal_dependencies
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
111
  - Helper library: https://github.com/bot-zen/ud-hf-parquet-tools
112
+ - Discussions: https://huggingface.co/datasets/universal-dependencies/universal_dependencies/discussions
README.md CHANGED
@@ -218,6 +218,7 @@ tags:
218
  - constituency-parsing
219
  - dependency-parsing
220
  - part-of-speech-tagging
 
221
  configs:
222
  - config_name: abq_atb
223
  data_files:
@@ -2332,7 +2333,7 @@ configs:
2332
  - **Parquet Format:** faster loading with HuggingFace datasets >=4.0.0
2333
  - **MWT Support:** New `mwt` field provides structured multi-word token information
2334
  - **Enhanced Security:** No more `trust_remote_code=True` required
2335
- - **Separate Versioning:** Loader version (2.0.0) distinct from UD data version (2.17)
2336
 
2337
  **Breaking Changes:**
2338
  - Token sequences now exclude MWT surface forms (matches UD guidelines)
@@ -2378,8 +2379,10 @@ configs:
2378
 
2379
  Universal Dependencies is a project that seeks to develop cross-linguistically consistent treebank annotation for many languages, with the goal of facilitating multilingual parser development, cross-lingual learning, and parsing research from a language typology perspective. The annotation scheme is based on (universal) Stanford dependencies (de Marneffe et al., 2006, 2008, 2014), Google universal part-of-speech tags (Petrov et al., 2012), and the Interset interlingua for morphosyntactic tagsets (Zeman, 2008).
2380
 
2381
- This is a (temporary) fork of
2382
- [/universal-dependencies/universal_dependencies](https://huggingface.co/datasets/universal-dependencies/universal_dependencies).
 
 
2383
 
2384
  ### Data Quality & Fidelity
2385
 
@@ -2406,7 +2409,12 @@ For technical details, see [CONLLU_PARSING.md](https://github.com/bot-zen/ud-hf-
2406
  from datasets import load_dataset
2407
 
2408
  # Load a specific treebank
2409
- ds = load_dataset("commul/universal_dependencies", "en_ewt", revision="2.18", split="train")
 
 
 
 
 
2410
 
2411
  # Access sentence data
2412
  sentence = ds[0]
@@ -2414,11 +2422,8 @@ print(f"Sentence ID: {sentence['sent_id']}")
2414
  print(f"Text: {sentence['text']}")
2415
  print(f"Tokens: {sentence['tokens']}")
2416
 
2417
- ## TODO: Make helper functions available
2418
- ## post v2.0 universal_dependencies.py is not part of the Dataset any longer!
2419
- ##
2420
  # Parse optional fields using helper functions
2421
- from universal_dependencies import parse_feats, parse_misc
2422
 
2423
  for i, token in enumerate(sentence['tokens']):
2424
  feats = parse_feats(sentence['feats'][i]) # Returns dict or {}
@@ -2426,7 +2431,7 @@ for i, token in enumerate(sentence['tokens']):
2426
  print(f"{token}: UPOS={sentence['upos'][i]}, feats={feats}, misc={misc}")
2427
 
2428
  # Export back to CoNLL-U format
2429
- from universal_dependencies import write_conllu
2430
 
2431
  # Write to stdout
2432
  write_conllu(ds)
@@ -2443,7 +2448,9 @@ conllu_text = buffer.getvalue()
2443
 
2444
  ### Supported Tasks and Leaderboards
2445
 
2446
- [More Information Needed]
 
 
2447
 
2448
  ## Dataset Structure
2449
 
@@ -2463,7 +2470,7 @@ This dataset has 353 configurations (treebanks).
2463
  from datasets import get_dataset_config_names, load_dataset
2464
 
2465
  # Get all available treebank configurations for revision="2.18"
2466
- configs = get_dataset_config_names("commul/universal_dependencies", revision="2.18")
2467
  print(f"Available treebanks: {len(configs)}")
2468
 
2469
  # Example configurations:
@@ -2477,10 +2484,10 @@ print(f"Available treebanks: {len(configs)}")
2477
  # ]
2478
 
2479
  # Get the latest configurations
2480
- get_dataset_config_names("commul/universal_dependencies")
2481
 
2482
  # Load a specific treebank
2483
- dataset = load_dataset("commul/universal_dependencies", "en_ewt")
2484
  print(dataset)
2485
 
2486
  # Output:
@@ -2511,7 +2518,7 @@ Each example in the dataset contains the following fields:
2511
  - **upos** (list of strings): Universal POS tags
2512
  - **xpos** (list of strings): Language-specific POS tags
2513
  - **feats** (list of strings): Morphological features in UD format
2514
- - **head** (list of strings): Head indices for dependency relations
2515
  - **deprel** (list of strings): Dependency relation labels
2516
  - **deps** (list of strings): Enhanced dependency graph
2517
  - **misc** (list of strings): Miscellaneous annotations
@@ -2527,7 +2534,7 @@ Each example in the dataset contains the following fields:
2527
  ```python
2528
  from datasets import load_dataset
2529
 
2530
- dataset = load_dataset("commul/universal_dependencies", "en_ewt", split="train")
2531
  print(dataset[0])
2532
 
2533
  # Output:
@@ -2542,10 +2549,10 @@ print(dataset[0])
2542
  ],
2543
  'tokens': ['Al', '-', 'Zaman', ':', 'American', 'forces', 'killed', 'Shaikh', 'Abdullah', 'al', '-', 'Ani', ',', 'the', 'preacher', 'at', 'the', 'mosque', 'in', 'the', 'town', 'of', 'Qaim', ',', 'near', 'the', 'Syrian', 'border', '.'],
2544
  'lemmas': ['Al', '-', 'Zaman', ':', 'American', 'force', 'kill', 'Shaikh', 'Abdullah', 'al', '-', 'Ani', ',', 'the', 'preacher', 'at', 'the', 'mosque', 'in', 'the', 'town', 'of', 'Qaim', ',', 'near', 'the', 'Syrian', 'border', '.'],
2545
- 'upos': [10, 1, 10, 1, 6, 0, 16, 10, 10, 10, 1, 10, 1, 8, 0, 2, 8, 0, 2, 8, 0, 2, 10, 1, 2, 8, 6, 0, 1],
2546
  'xpos': ['NNP', 'HYPH', 'NNP', ':', 'JJ', 'NNS', 'VBD', 'NNP', 'NNP', 'NNP', 'HYPH', 'NNP', ',', 'DT', 'NN', 'IN', 'DT', 'NN', 'IN', 'DT', 'NN', 'IN', 'NNP', ',', 'IN', 'DT', 'JJ', 'NN', '.'],
2547
  'feats': ['Number=Sing', None, 'Number=Sing', None, 'Degree=Pos', 'Number=Plur', 'Mood=Ind|Number=Plur|Person=3|Tense=Past|VerbForm=Fin', 'Number=Sing', 'Number=Sing', 'Number=Sing', None, 'Number=Sing', None, 'Definite=Def|PronType=Art', 'Number=Sing', None, 'Definite=Def|PronType=Art', 'Number=Sing', None, 'Definite=Def|PronType=Art', 'Number=Sing', None, 'Number=Sing', None, None, 'Definite=Def|PronType=Art', 'Degree=Pos', 'Number=Sing', None],
2548
- 'head': ['0', '3', '1', '7', '6', '7', '1', '7', '8', '8', '12', '8', '15', '15', '8', '18', '18', '15', '21', '21', '18', '23', '21', '28', '28', '28', '28', '21', '1'],
2549
  'deprel': ['root', 'punct', 'flat', 'punct', 'amod', 'nsubj', 'parataxis', 'obj', 'flat', 'flat', 'punct', 'flat', 'punct', 'det', 'appos', 'case', 'det', 'nmod', 'case', 'det', 'nmod', 'case', 'nmod', 'punct', 'case', 'det', 'amod', 'nmod', 'punct'],
2550
  'deps': ['0:root', '3:punct', '1:flat', '7:punct', '6:amod', '7:nsubj', '1:parataxis', '7:obj', '8:flat', '8:flat', '12:punct', '8:flat', '15:punct', '15:det', '8:appos', '18:case', '18:det', '15:nmod:at', '21:case', '21:det', '18:nmod:in', '23:case', '21:nmod:of', '28:punct', '28:case', '28:det', '28:amod', '21:nmod:near', '1:punct'],
2551
  'misc': ['SpaceAfter=No', 'SpaceAfter=No', None, None, None, None, None, None, None, 'SpaceAfter=No', 'SpaceAfter=No', 'SpaceAfter=No', None, None, None, None, None, None, None, None, None, None, 'SpaceAfter=No', None, None, None, None, 'SpaceAfter=No', None],
@@ -2557,7 +2564,7 @@ print(dataset[0])
2557
  *MWT Example (French):**
2558
 
2559
  ```python
2560
- dataset = load_dataset("commul/universal_dependencies", "fr_gsd", split="train")
2561
  # Find sentence with MWT
2562
  example = [ex for ex in dataset if ex['mwt']][0]
2563
  print(example['mwt'])
@@ -2575,7 +2582,7 @@ The file `metadata.json` stores additional information about the data, for examp
2575
  from huggingface_hub import hf_hub_download
2576
  import json
2577
 
2578
- md = hf_hub_download(repo_id="commul/universal_dependencies", filename="metadata.json", repo_type="dataset")
2579
 
2580
  with open(md, "r", encoding="utf-8") as f:
2581
  metadata = json.load(f)
@@ -2631,7 +2638,9 @@ with open(md, "r", encoding="utf-8") as f:
2631
 
2632
  ### Dataset Curators
2633
 
2634
- [More Information Needed]
 
 
2635
 
2636
  ### Licensing Information
2637
  Universal Dependencies is a collection of linguistic data and tools. Each of
@@ -2646,6 +2655,11 @@ The `./tools/` are licensed under the [Apache-2.0](https://www.apache.org/licens
2646
 
2647
  ### Citation Information
2648
 
 
 
 
 
 
2649
  ```bibtex
2650
  @misc{11234/1-6149,
2651
  title = {Universal Dependencies 2.18},
@@ -2659,4 +2673,8 @@ The `./tools/` are licensed under the [Apache-2.0](https://www.apache.org/licens
2659
 
2660
  ### Contributions
2661
 
2662
- Thanks to [universal-dependencies](https://huggingface.co/universal-dependencies) for [the original of this dataset](https://huggingface.co/datasets/universal-dependencies/universal_dependencies).
 
 
 
 
 
218
  - constituency-parsing
219
  - dependency-parsing
220
  - part-of-speech-tagging
221
+
222
  configs:
223
  - config_name: abq_atb
224
  data_files:
 
2333
  - **Parquet Format:** faster loading with HuggingFace datasets >=4.0.0
2334
  - **MWT Support:** New `mwt` field provides structured multi-word token information
2335
  - **Enhanced Security:** No more `trust_remote_code=True` required
2336
+ - **Separate Versioning:** Loader version (2.0.0) distinct from UD data version (2.18)
2337
 
2338
  **Breaking Changes:**
2339
  - Token sequences now exclude MWT surface forms (matches UD guidelines)
 
2379
 
2380
  Universal Dependencies is a project that seeks to develop cross-linguistically consistent treebank annotation for many languages, with the goal of facilitating multilingual parser development, cross-lingual learning, and parsing research from a language typology perspective. The annotation scheme is based on (universal) Stanford dependencies (de Marneffe et al., 2006, 2008, 2014), Google universal part-of-speech tags (Petrov et al., 2012), and the Interset interlingua for morphosyntactic tagsets (Zeman, 2008).
2381
 
2382
+ This is the official Hugging Face Parquet distribution for Universal
2383
+ Dependencies at
2384
+ [`universal-dependencies/universal_dependencies`](https://huggingface.co/datasets/universal-dependencies/universal_dependencies).
2385
+ The dataset was formerly maintained at `commul/universal_dependencies`.
2386
 
2387
  ### Data Quality & Fidelity
2388
 
 
2409
  from datasets import load_dataset
2410
 
2411
  # Load a specific treebank
2412
+ ds = load_dataset(
2413
+ "universal-dependencies/universal_dependencies",
2414
+ "en_ewt",
2415
+ revision="2.18",
2416
+ split="train",
2417
+ )
2418
 
2419
  # Access sentence data
2420
  sentence = ds[0]
 
2422
  print(f"Text: {sentence['text']}")
2423
  print(f"Tokens: {sentence['tokens']}")
2424
 
 
 
 
2425
  # Parse optional fields using helper functions
2426
+ from ud_hf_parquet_tools import parse_feats, parse_misc
2427
 
2428
  for i, token in enumerate(sentence['tokens']):
2429
  feats = parse_feats(sentence['feats'][i]) # Returns dict or {}
 
2431
  print(f"{token}: UPOS={sentence['upos'][i]}, feats={feats}, misc={misc}")
2432
 
2433
  # Export back to CoNLL-U format
2434
+ from ud_hf_parquet_tools import write_conllu
2435
 
2436
  # Write to stdout
2437
  write_conllu(ds)
 
2448
 
2449
  ### Supported Tasks and Leaderboards
2450
 
2451
+ This dataset supports dependency parsing, universal and language-specific
2452
+ part-of-speech tagging, lemmatization, and morphology-oriented token
2453
+ classification. No single leaderboard is maintained in this dataset card.
2454
 
2455
  ## Dataset Structure
2456
 
 
2470
  from datasets import get_dataset_config_names, load_dataset
2471
 
2472
  # Get all available treebank configurations for revision="2.18"
2473
+ configs = get_dataset_config_names("universal-dependencies/universal_dependencies", revision="2.18")
2474
  print(f"Available treebanks: {len(configs)}")
2475
 
2476
  # Example configurations:
 
2484
  # ]
2485
 
2486
  # Get the latest configurations
2487
+ get_dataset_config_names("universal-dependencies/universal_dependencies")
2488
 
2489
  # Load a specific treebank
2490
+ dataset = load_dataset("universal-dependencies/universal_dependencies", "en_ewt")
2491
  print(dataset)
2492
 
2493
  # Output:
 
2518
  - **upos** (list of strings): Universal POS tags
2519
  - **xpos** (list of strings): Language-specific POS tags
2520
  - **feats** (list of strings): Morphological features in UD format
2521
+ - **head** (list of integers): Head indices for dependency relations
2522
  - **deprel** (list of strings): Dependency relation labels
2523
  - **deps** (list of strings): Enhanced dependency graph
2524
  - **misc** (list of strings): Miscellaneous annotations
 
2534
  ```python
2535
  from datasets import load_dataset
2536
 
2537
+ dataset = load_dataset("universal-dependencies/universal_dependencies", "en_ewt", split="train")
2538
  print(dataset[0])
2539
 
2540
  # Output:
 
2549
  ],
2550
  'tokens': ['Al', '-', 'Zaman', ':', 'American', 'forces', 'killed', 'Shaikh', 'Abdullah', 'al', '-', 'Ani', ',', 'the', 'preacher', 'at', 'the', 'mosque', 'in', 'the', 'town', 'of', 'Qaim', ',', 'near', 'the', 'Syrian', 'border', '.'],
2551
  'lemmas': ['Al', '-', 'Zaman', ':', 'American', 'force', 'kill', 'Shaikh', 'Abdullah', 'al', '-', 'Ani', ',', 'the', 'preacher', 'at', 'the', 'mosque', 'in', 'the', 'town', 'of', 'Qaim', ',', 'near', 'the', 'Syrian', 'border', '.'],
2552
+ 'upos': ['PROPN', 'PUNCT', 'PROPN', 'PUNCT', 'ADJ', 'NOUN', 'VERB', ...],
2553
  'xpos': ['NNP', 'HYPH', 'NNP', ':', 'JJ', 'NNS', 'VBD', 'NNP', 'NNP', 'NNP', 'HYPH', 'NNP', ',', 'DT', 'NN', 'IN', 'DT', 'NN', 'IN', 'DT', 'NN', 'IN', 'NNP', ',', 'IN', 'DT', 'JJ', 'NN', '.'],
2554
  'feats': ['Number=Sing', None, 'Number=Sing', None, 'Degree=Pos', 'Number=Plur', 'Mood=Ind|Number=Plur|Person=3|Tense=Past|VerbForm=Fin', 'Number=Sing', 'Number=Sing', 'Number=Sing', None, 'Number=Sing', None, 'Definite=Def|PronType=Art', 'Number=Sing', None, 'Definite=Def|PronType=Art', 'Number=Sing', None, 'Definite=Def|PronType=Art', 'Number=Sing', None, 'Number=Sing', None, None, 'Definite=Def|PronType=Art', 'Degree=Pos', 'Number=Sing', None],
2555
+ 'head': [0, 3, 1, 7, 6, 7, 1, ...],
2556
  'deprel': ['root', 'punct', 'flat', 'punct', 'amod', 'nsubj', 'parataxis', 'obj', 'flat', 'flat', 'punct', 'flat', 'punct', 'det', 'appos', 'case', 'det', 'nmod', 'case', 'det', 'nmod', 'case', 'nmod', 'punct', 'case', 'det', 'amod', 'nmod', 'punct'],
2557
  'deps': ['0:root', '3:punct', '1:flat', '7:punct', '6:amod', '7:nsubj', '1:parataxis', '7:obj', '8:flat', '8:flat', '12:punct', '8:flat', '15:punct', '15:det', '8:appos', '18:case', '18:det', '15:nmod:at', '21:case', '21:det', '18:nmod:in', '23:case', '21:nmod:of', '28:punct', '28:case', '28:det', '28:amod', '21:nmod:near', '1:punct'],
2558
  'misc': ['SpaceAfter=No', 'SpaceAfter=No', None, None, None, None, None, None, None, 'SpaceAfter=No', 'SpaceAfter=No', 'SpaceAfter=No', None, None, None, None, None, None, None, None, None, None, 'SpaceAfter=No', None, None, None, None, 'SpaceAfter=No', None],
 
2564
  *MWT Example (French):**
2565
 
2566
  ```python
2567
+ dataset = load_dataset("universal-dependencies/universal_dependencies", "fr_gsd", split="train")
2568
  # Find sentence with MWT
2569
  example = [ex for ex in dataset if ex['mwt']][0]
2570
  print(example['mwt'])
 
2582
  from huggingface_hub import hf_hub_download
2583
  import json
2584
 
2585
+ md = hf_hub_download(repo_id="universal-dependencies/universal_dependencies", filename="metadata.json", repo_type="dataset")
2586
 
2587
  with open(md, "r", encoding="utf-8") as f:
2588
  metadata = json.load(f)
 
2638
 
2639
  ### Dataset Curators
2640
 
2641
+ Universal Dependencies is maintained by the Universal Dependencies community.
2642
+ This Hugging Face Parquet distribution is maintained under the official
2643
+ `universal-dependencies` organization.
2644
 
2645
  ### Licensing Information
2646
  Universal Dependencies is a collection of linguistic data and tools. Each of
 
2655
 
2656
  ### Citation Information
2657
 
2658
+ Please cite the Universal Dependencies release used in your work. If you use a
2659
+ specific treebank, also cite the treebank-specific references listed in that
2660
+ treebank's upstream README when available. For general UD background, cite the
2661
+ UD v2 paper listed in the dataset description.
2662
+
2663
  ```bibtex
2664
  @misc{11234/1-6149,
2665
  title = {Universal Dependencies 2.18},
 
2673
 
2674
  ### Contributions
2675
 
2676
+ The v2.0 Parquet rewrite was originally authored by Egon W. Stemle. This
2677
+ dataset was formerly hosted at `commul/universal_dependencies` and is now
2678
+ maintained at the official
2679
+ [`universal-dependencies/universal_dependencies`](https://huggingface.co/datasets/universal-dependencies/universal_dependencies)
2680
+ location.
RELEASE.md CHANGED
@@ -284,7 +284,7 @@ Do not use lightweight tags (`git tag v${NEW_VER}`) for framework releases.
284
 
285
  This release updates the Universal Dependencies dataset **framework** to v2.1.0.
286
 
287
- **This is a framework release, not a UD data release.** It applies to all UD data versions (2.7-2.17).
288
 
289
  ### What's New
290
 
@@ -300,7 +300,7 @@ Do not use lightweight tags (`git tag v${NEW_VER}`) for framework releases.
300
 
301
  For users: See [MIGRATION.md](MIGRATION.md) if breaking changes exist.
302
 
303
- For maintainers: Framework has been applied to all UD data branches (2.7-2.17).
304
 
305
  ### Full Changelog
306
 
@@ -375,7 +375,7 @@ pytest tests/test_conversion.py
375
  1. **Schema validation:**
376
  ```python
377
  from datasets import load_dataset
378
- ds = load_dataset("commul/universal_dependencies", "en_ewt", revision="2.17")
379
  assert "mwt" in ds["train"].features
380
  assert "empty_nodes" in ds["train"].features
381
  ```
@@ -389,7 +389,7 @@ pytest tests/test_conversion.py
389
  3. **Backwards compatibility (if minor/patch):**
390
  ```python
391
  # Old code should still work
392
- ds = load_dataset("commul/universal_dependencies", "fr_gsd", revision="2.17")
393
  tokens = ds["train"][0]["tokens"]
394
  # No errors
395
  ```
 
284
 
285
  This release updates the Universal Dependencies dataset **framework** to v2.1.0.
286
 
287
+ **This is a framework release, not a UD data release.** It applies to all UD data versions (2.7-2.18).
288
 
289
  ### What's New
290
 
 
300
 
301
  For users: See [MIGRATION.md](MIGRATION.md) if breaking changes exist.
302
 
303
+ For maintainers: Framework has been applied to all UD data branches (2.7-2.18).
304
 
305
  ### Full Changelog
306
 
 
375
  1. **Schema validation:**
376
  ```python
377
  from datasets import load_dataset
378
+ ds = load_dataset("universal-dependencies/universal_dependencies", "en_ewt", revision="2.17")
379
  assert "mwt" in ds["train"].features
380
  assert "empty_nodes" in ds["train"].features
381
  ```
 
389
  3. **Backwards compatibility (if minor/patch):**
390
  ```python
391
  # Old code should still work
392
+ ds = load_dataset("universal-dependencies/universal_dependencies", "fr_gsd", revision="2.17")
393
  tokens = ds["train"][0]["tokens"]
394
  # No errors
395
  ```
tools/04_generate_parquet.py CHANGED
@@ -5,7 +5,7 @@ Generate Parquet files from Universal Dependencies CoNLL-U data.
5
  This script is a wrapper around the ud-hf-parquet-tools library.
6
  For the actual implementation, see: https://github.com/bot-zen/ud-hf-parquet-tools
7
 
8
- Repository: commul/universal_dependencies
9
 
10
  Usage:
11
  python 04_generate_parquet.py [OPTIONS]
 
5
  This script is a wrapper around the ud-hf-parquet-tools library.
6
  For the actual implementation, see: https://github.com/bot-zen/ud-hf-parquet-tools
7
 
8
+ Repository: universal-dependencies/universal_dependencies
9
 
10
  Usage:
11
  python 04_generate_parquet.py [OPTIONS]
tools/05_validate_parquet.py CHANGED
@@ -9,7 +9,7 @@ This script can validate both:
9
  - Remote: Downloaded from HuggingFace Hub
10
  - Local: Local parquet files in ../parquet/
11
 
12
- Repository: commul/universal_dependencies
13
 
14
  Usage:
15
  python 05_validate_parquet.py [OPTIONS]
 
9
  - Remote: Downloaded from HuggingFace Hub
10
  - Local: Local parquet files in ../parquet/
11
 
12
+ Repository: universal-dependencies/universal_dependencies
13
 
14
  Usage:
15
  python 05_validate_parquet.py [OPTIONS]
tools/BLOCKED_TREEBANKS.md CHANGED
@@ -46,7 +46,7 @@ The dataset loader (`universal_dependencies.py`) handles blocked treebanks grace
46
  from datasets import load_dataset
47
 
48
  # Fast loading from Parquet
49
- ds = load_dataset("commul/universal_dependencies", "en_ewt", split="train")
50
  ```
51
 
52
  ### Loading a Blocked Treebank
@@ -55,7 +55,7 @@ from datasets import load_dataset
55
 
56
  # Falls back to CoNLL-U loading (requires trust_remote_code)
57
  ds = load_dataset(
58
- "commul/universal_dependencies",
59
  "pt_cintil", # Example blocked treebank
60
  split="train",
61
  trust_remote_code=True
 
46
  from datasets import load_dataset
47
 
48
  # Fast loading from Parquet
49
+ ds = load_dataset("universal-dependencies/universal_dependencies", "en_ewt", split="train")
50
  ```
51
 
52
  ### Loading a Blocked Treebank
 
55
 
56
  # Falls back to CoNLL-U loading (requires trust_remote_code)
57
  ds = load_dataset(
58
+ "universal-dependencies/universal_dependencies",
59
  "pt_cintil", # Example blocked treebank
60
  split="train",
61
  trust_remote_code=True
tools/README-2.18 CHANGED
@@ -1,13 +1,9 @@
1
  ---
2
- ### THIS IS A GENERATED FILE.
3
  viewer: true
4
-
5
  annotations_creators:
6
  - expert-generated
7
-
8
  language_creators:
9
  - crowdsourced
10
-
11
  language:
12
  - abq
13
  - ab
@@ -202,30 +198,21 @@ language:
202
  - ess
203
  - say
204
  - zza
205
-
206
- license:
207
- - apache-2.0
208
-
209
  multilinguality:
210
  - multilingual
211
-
212
  size_categories:
213
- - '1K<n<10K'
214
-
215
  source_datasets:
216
  - original
217
-
218
  task_categories:
219
  - token-classification
220
-
221
  task_ids:
222
  - parsing
223
  - part-of-speech
224
  - lemmatization
225
-
226
  paperswithcode_id: universal-dependencies
227
  pretty_name: Universal Dependencies Treebank
228
-
229
  tags:
230
  - text
231
  - constituency-parsing
@@ -2346,7 +2333,7 @@ configs:
2346
  - **Parquet Format:** faster loading with HuggingFace datasets >=4.0.0
2347
  - **MWT Support:** New `mwt` field provides structured multi-word token information
2348
  - **Enhanced Security:** No more `trust_remote_code=True` required
2349
- - **Separate Versioning:** Loader version (2.0.0) distinct from UD data version (2.17)
2350
 
2351
  **Breaking Changes:**
2352
  - Token sequences now exclude MWT surface forms (matches UD guidelines)
@@ -2387,16 +2374,15 @@ configs:
2387
  - **Repository:** [Universal Dependencies](https://github.com/UniversalDependencies)
2388
  - **Paper:** [Universal Dependencies v2: An Evergrowing Multilingual Treebank Collection](https://arxiv.org/abs/2004.10643)
2389
  - **Paper:** [From Treebank Metadata to Sentence-Level Genre in Universal Dependencies: A Reproducible, Versioned Resource](https://universaldependencies.org/udw26/papers/41_Paper.pdf)
2390
- - **Leaderboard:**
2391
- - **Point of Contact:** [appliedlinguisticsdevs@eurac.edu](mailto:appliedlinguisticsdevs@eurac.edu)
2392
- - **Point of Contact:** [IAL Homepage](https://www.eurac.edu/linguistics)
2393
 
2394
  ### Dataset Summary
2395
 
2396
  Universal Dependencies is a project that seeks to develop cross-linguistically consistent treebank annotation for many languages, with the goal of facilitating multilingual parser development, cross-lingual learning, and parsing research from a language typology perspective. The annotation scheme is based on (universal) Stanford dependencies (de Marneffe et al., 2006, 2008, 2014), Google universal part-of-speech tags (Petrov et al., 2012), and the Interset interlingua for morphosyntactic tagsets (Zeman, 2008).
2397
 
2398
- This is a (temporary) fork of
2399
- [/universal-dependencies/universal_dependencies](https://huggingface.co/datasets/universal-dependencies/universal_dependencies).
 
 
2400
 
2401
  ### Data Quality & Fidelity
2402
 
@@ -2423,7 +2409,12 @@ For technical details, see [CONLLU_PARSING.md](https://github.com/bot-zen/ud-hf-
2423
  from datasets import load_dataset
2424
 
2425
  # Load a specific treebank
2426
- ds = load_dataset("commul/universal_dependencies", "en_ewt", revision="2.18", split="train")
 
 
 
 
 
2427
 
2428
  # Access sentence data
2429
  sentence = ds[0]
@@ -2431,11 +2422,8 @@ print(f"Sentence ID: {sentence['sent_id']}")
2431
  print(f"Text: {sentence['text']}")
2432
  print(f"Tokens: {sentence['tokens']}")
2433
 
2434
- ## TODO: Make helper functions available
2435
- ## post v2.0 universal_dependencies.py is not part of the Dataset any longer!
2436
- ##
2437
  # Parse optional fields using helper functions
2438
- from universal_dependencies import parse_feats, parse_misc
2439
 
2440
  for i, token in enumerate(sentence['tokens']):
2441
  feats = parse_feats(sentence['feats'][i]) # Returns dict or {}
@@ -2443,7 +2431,7 @@ for i, token in enumerate(sentence['tokens']):
2443
  print(f"{token}: UPOS={sentence['upos'][i]}, feats={feats}, misc={misc}")
2444
 
2445
  # Export back to CoNLL-U format
2446
- from universal_dependencies import write_conllu
2447
 
2448
  # Write to stdout
2449
  write_conllu(ds)
@@ -2460,7 +2448,9 @@ conllu_text = buffer.getvalue()
2460
 
2461
  ### Supported Tasks and Leaderboards
2462
 
2463
- [More Information Needed]
 
 
2464
 
2465
  ## Dataset Structure
2466
 
@@ -2480,7 +2470,7 @@ This dataset has 353 configurations (treebanks).
2480
  from datasets import get_dataset_config_names, load_dataset
2481
 
2482
  # Get all available treebank configurations for revision="2.18"
2483
- configs = get_dataset_config_names("commul/universal_dependencies", revision="2.18")
2484
  print(f"Available treebanks: {len(configs)}")
2485
 
2486
  # Example configurations:
@@ -2494,10 +2484,10 @@ print(f"Available treebanks: {len(configs)}")
2494
  # ]
2495
 
2496
  # Get the latest configurations
2497
- get_dataset_config_names("commul/universal_dependencies")
2498
 
2499
  # Load a specific treebank
2500
- dataset = load_dataset("commul/universal_dependencies", "en_ewt")
2501
  print(dataset)
2502
 
2503
  # Output:
@@ -2528,7 +2518,7 @@ Each example in the dataset contains the following fields:
2528
  - **upos** (list of strings): Universal POS tags
2529
  - **xpos** (list of strings): Language-specific POS tags
2530
  - **feats** (list of strings): Morphological features in UD format
2531
- - **head** (list of strings): Head indices for dependency relations
2532
  - **deprel** (list of strings): Dependency relation labels
2533
  - **deps** (list of strings): Enhanced dependency graph
2534
  - **misc** (list of strings): Miscellaneous annotations
@@ -2544,7 +2534,7 @@ Each example in the dataset contains the following fields:
2544
  ```python
2545
  from datasets import load_dataset
2546
 
2547
- dataset = load_dataset("commul/universal_dependencies", "en_ewt", split="train")
2548
  print(dataset[0])
2549
 
2550
  # Output:
@@ -2559,10 +2549,10 @@ print(dataset[0])
2559
  ],
2560
  'tokens': ['Al', '-', 'Zaman', ':', 'American', 'forces', 'killed', 'Shaikh', 'Abdullah', 'al', '-', 'Ani', ',', 'the', 'preacher', 'at', 'the', 'mosque', 'in', 'the', 'town', 'of', 'Qaim', ',', 'near', 'the', 'Syrian', 'border', '.'],
2561
  'lemmas': ['Al', '-', 'Zaman', ':', 'American', 'force', 'kill', 'Shaikh', 'Abdullah', 'al', '-', 'Ani', ',', 'the', 'preacher', 'at', 'the', 'mosque', 'in', 'the', 'town', 'of', 'Qaim', ',', 'near', 'the', 'Syrian', 'border', '.'],
2562
- 'upos': [10, 1, 10, 1, 6, 0, 16, 10, 10, 10, 1, 10, 1, 8, 0, 2, 8, 0, 2, 8, 0, 2, 10, 1, 2, 8, 6, 0, 1],
2563
  'xpos': ['NNP', 'HYPH', 'NNP', ':', 'JJ', 'NNS', 'VBD', 'NNP', 'NNP', 'NNP', 'HYPH', 'NNP', ',', 'DT', 'NN', 'IN', 'DT', 'NN', 'IN', 'DT', 'NN', 'IN', 'NNP', ',', 'IN', 'DT', 'JJ', 'NN', '.'],
2564
  'feats': ['Number=Sing', None, 'Number=Sing', None, 'Degree=Pos', 'Number=Plur', 'Mood=Ind|Number=Plur|Person=3|Tense=Past|VerbForm=Fin', 'Number=Sing', 'Number=Sing', 'Number=Sing', None, 'Number=Sing', None, 'Definite=Def|PronType=Art', 'Number=Sing', None, 'Definite=Def|PronType=Art', 'Number=Sing', None, 'Definite=Def|PronType=Art', 'Number=Sing', None, 'Number=Sing', None, None, 'Definite=Def|PronType=Art', 'Degree=Pos', 'Number=Sing', None],
2565
- 'head': ['0', '3', '1', '7', '6', '7', '1', '7', '8', '8', '12', '8', '15', '15', '8', '18', '18', '15', '21', '21', '18', '23', '21', '28', '28', '28', '28', '21', '1'],
2566
  'deprel': ['root', 'punct', 'flat', 'punct', 'amod', 'nsubj', 'parataxis', 'obj', 'flat', 'flat', 'punct', 'flat', 'punct', 'det', 'appos', 'case', 'det', 'nmod', 'case', 'det', 'nmod', 'case', 'nmod', 'punct', 'case', 'det', 'amod', 'nmod', 'punct'],
2567
  'deps': ['0:root', '3:punct', '1:flat', '7:punct', '6:amod', '7:nsubj', '1:parataxis', '7:obj', '8:flat', '8:flat', '12:punct', '8:flat', '15:punct', '15:det', '8:appos', '18:case', '18:det', '15:nmod:at', '21:case', '21:det', '18:nmod:in', '23:case', '21:nmod:of', '28:punct', '28:case', '28:det', '28:amod', '21:nmod:near', '1:punct'],
2568
  'misc': ['SpaceAfter=No', 'SpaceAfter=No', None, None, None, None, None, None, None, 'SpaceAfter=No', 'SpaceAfter=No', 'SpaceAfter=No', None, None, None, None, None, None, None, None, None, None, 'SpaceAfter=No', None, None, None, None, 'SpaceAfter=No', None],
@@ -2574,7 +2564,7 @@ print(dataset[0])
2574
  *MWT Example (French):**
2575
 
2576
  ```python
2577
- dataset = load_dataset("commul/universal_dependencies", "fr_gsd", split="train")
2578
  # Find sentence with MWT
2579
  example = [ex for ex in dataset if ex['mwt']][0]
2580
  print(example['mwt'])
@@ -2592,7 +2582,7 @@ The file `metadata.json` stores additional information about the data, for examp
2592
  from huggingface_hub import hf_hub_download
2593
  import json
2594
 
2595
- md = hf_hub_download(repo_id="commul/universal_dependencies", filename="metadata.json", repo_type="dataset")
2596
 
2597
  with open(md, "r", encoding="utf-8") as f:
2598
  metadata = json.load(f)
@@ -2648,7 +2638,9 @@ with open(md, "r", encoding="utf-8") as f:
2648
 
2649
  ### Dataset Curators
2650
 
2651
- [More Information Needed]
 
 
2652
 
2653
  ### Licensing Information
2654
  Universal Dependencies is a collection of linguistic data and tools. Each of
@@ -2663,6 +2655,11 @@ The `./tools/` are licensed under the [Apache-2.0](https://www.apache.org/licens
2663
 
2664
  ### Citation Information
2665
 
 
 
 
 
 
2666
  ```bibtex
2667
  @misc{11234/1-6149,
2668
  title = {Universal Dependencies 2.18},
@@ -2676,4 +2673,8 @@ The `./tools/` are licensed under the [Apache-2.0](https://www.apache.org/licens
2676
 
2677
  ### Contributions
2678
 
2679
- Thanks to [universal-dependencies](https://huggingface.co/universal-dependencies) for [the original of this dataset](https://huggingface.co/datasets/universal-dependencies/universal_dependencies).
 
 
 
 
 
1
  ---
 
2
  viewer: true
 
3
  annotations_creators:
4
  - expert-generated
 
5
  language_creators:
6
  - crowdsourced
 
7
  language:
8
  - abq
9
  - ab
 
198
  - ess
199
  - say
200
  - zza
201
+ license: other
 
 
 
202
  multilinguality:
203
  - multilingual
 
204
  size_categories:
205
+ - 1M<n<10M
 
206
  source_datasets:
207
  - original
 
208
  task_categories:
209
  - token-classification
 
210
  task_ids:
211
  - parsing
212
  - part-of-speech
213
  - lemmatization
 
214
  paperswithcode_id: universal-dependencies
215
  pretty_name: Universal Dependencies Treebank
 
216
  tags:
217
  - text
218
  - constituency-parsing
 
2333
  - **Parquet Format:** faster loading with HuggingFace datasets >=4.0.0
2334
  - **MWT Support:** New `mwt` field provides structured multi-word token information
2335
  - **Enhanced Security:** No more `trust_remote_code=True` required
2336
+ - **Separate Versioning:** Loader version (2.0.0) distinct from UD data version (2.18)
2337
 
2338
  **Breaking Changes:**
2339
  - Token sequences now exclude MWT surface forms (matches UD guidelines)
 
2374
  - **Repository:** [Universal Dependencies](https://github.com/UniversalDependencies)
2375
  - **Paper:** [Universal Dependencies v2: An Evergrowing Multilingual Treebank Collection](https://arxiv.org/abs/2004.10643)
2376
  - **Paper:** [From Treebank Metadata to Sentence-Level Genre in Universal Dependencies: A Reproducible, Versioned Resource](https://universaldependencies.org/udw26/papers/41_Paper.pdf)
 
 
 
2377
 
2378
  ### Dataset Summary
2379
 
2380
  Universal Dependencies is a project that seeks to develop cross-linguistically consistent treebank annotation for many languages, with the goal of facilitating multilingual parser development, cross-lingual learning, and parsing research from a language typology perspective. The annotation scheme is based on (universal) Stanford dependencies (de Marneffe et al., 2006, 2008, 2014), Google universal part-of-speech tags (Petrov et al., 2012), and the Interset interlingua for morphosyntactic tagsets (Zeman, 2008).
2381
 
2382
+ This is the official Hugging Face Parquet distribution for Universal
2383
+ Dependencies at
2384
+ [`universal-dependencies/universal_dependencies`](https://huggingface.co/datasets/universal-dependencies/universal_dependencies).
2385
+ The dataset was formerly maintained at `commul/universal_dependencies`.
2386
 
2387
  ### Data Quality & Fidelity
2388
 
 
2409
  from datasets import load_dataset
2410
 
2411
  # Load a specific treebank
2412
+ ds = load_dataset(
2413
+ "universal-dependencies/universal_dependencies",
2414
+ "en_ewt",
2415
+ revision="2.18",
2416
+ split="train",
2417
+ )
2418
 
2419
  # Access sentence data
2420
  sentence = ds[0]
 
2422
  print(f"Text: {sentence['text']}")
2423
  print(f"Tokens: {sentence['tokens']}")
2424
 
 
 
 
2425
  # Parse optional fields using helper functions
2426
+ from ud_hf_parquet_tools import parse_feats, parse_misc
2427
 
2428
  for i, token in enumerate(sentence['tokens']):
2429
  feats = parse_feats(sentence['feats'][i]) # Returns dict or {}
 
2431
  print(f"{token}: UPOS={sentence['upos'][i]}, feats={feats}, misc={misc}")
2432
 
2433
  # Export back to CoNLL-U format
2434
+ from ud_hf_parquet_tools import write_conllu
2435
 
2436
  # Write to stdout
2437
  write_conllu(ds)
 
2448
 
2449
  ### Supported Tasks and Leaderboards
2450
 
2451
+ This dataset supports dependency parsing, universal and language-specific
2452
+ part-of-speech tagging, lemmatization, and morphology-oriented token
2453
+ classification. No single leaderboard is maintained in this dataset card.
2454
 
2455
  ## Dataset Structure
2456
 
 
2470
  from datasets import get_dataset_config_names, load_dataset
2471
 
2472
  # Get all available treebank configurations for revision="2.18"
2473
+ configs = get_dataset_config_names("universal-dependencies/universal_dependencies", revision="2.18")
2474
  print(f"Available treebanks: {len(configs)}")
2475
 
2476
  # Example configurations:
 
2484
  # ]
2485
 
2486
  # Get the latest configurations
2487
+ get_dataset_config_names("universal-dependencies/universal_dependencies")
2488
 
2489
  # Load a specific treebank
2490
+ dataset = load_dataset("universal-dependencies/universal_dependencies", "en_ewt")
2491
  print(dataset)
2492
 
2493
  # Output:
 
2518
  - **upos** (list of strings): Universal POS tags
2519
  - **xpos** (list of strings): Language-specific POS tags
2520
  - **feats** (list of strings): Morphological features in UD format
2521
+ - **head** (list of integers): Head indices for dependency relations
2522
  - **deprel** (list of strings): Dependency relation labels
2523
  - **deps** (list of strings): Enhanced dependency graph
2524
  - **misc** (list of strings): Miscellaneous annotations
 
2534
  ```python
2535
  from datasets import load_dataset
2536
 
2537
+ dataset = load_dataset("universal-dependencies/universal_dependencies", "en_ewt", split="train")
2538
  print(dataset[0])
2539
 
2540
  # Output:
 
2549
  ],
2550
  'tokens': ['Al', '-', 'Zaman', ':', 'American', 'forces', 'killed', 'Shaikh', 'Abdullah', 'al', '-', 'Ani', ',', 'the', 'preacher', 'at', 'the', 'mosque', 'in', 'the', 'town', 'of', 'Qaim', ',', 'near', 'the', 'Syrian', 'border', '.'],
2551
  'lemmas': ['Al', '-', 'Zaman', ':', 'American', 'force', 'kill', 'Shaikh', 'Abdullah', 'al', '-', 'Ani', ',', 'the', 'preacher', 'at', 'the', 'mosque', 'in', 'the', 'town', 'of', 'Qaim', ',', 'near', 'the', 'Syrian', 'border', '.'],
2552
+ 'upos': ['PROPN', 'PUNCT', 'PROPN', 'PUNCT', 'ADJ', 'NOUN', 'VERB', ...],
2553
  'xpos': ['NNP', 'HYPH', 'NNP', ':', 'JJ', 'NNS', 'VBD', 'NNP', 'NNP', 'NNP', 'HYPH', 'NNP', ',', 'DT', 'NN', 'IN', 'DT', 'NN', 'IN', 'DT', 'NN', 'IN', 'NNP', ',', 'IN', 'DT', 'JJ', 'NN', '.'],
2554
  'feats': ['Number=Sing', None, 'Number=Sing', None, 'Degree=Pos', 'Number=Plur', 'Mood=Ind|Number=Plur|Person=3|Tense=Past|VerbForm=Fin', 'Number=Sing', 'Number=Sing', 'Number=Sing', None, 'Number=Sing', None, 'Definite=Def|PronType=Art', 'Number=Sing', None, 'Definite=Def|PronType=Art', 'Number=Sing', None, 'Definite=Def|PronType=Art', 'Number=Sing', None, 'Number=Sing', None, None, 'Definite=Def|PronType=Art', 'Degree=Pos', 'Number=Sing', None],
2555
+ 'head': [0, 3, 1, 7, 6, 7, 1, ...],
2556
  'deprel': ['root', 'punct', 'flat', 'punct', 'amod', 'nsubj', 'parataxis', 'obj', 'flat', 'flat', 'punct', 'flat', 'punct', 'det', 'appos', 'case', 'det', 'nmod', 'case', 'det', 'nmod', 'case', 'nmod', 'punct', 'case', 'det', 'amod', 'nmod', 'punct'],
2557
  'deps': ['0:root', '3:punct', '1:flat', '7:punct', '6:amod', '7:nsubj', '1:parataxis', '7:obj', '8:flat', '8:flat', '12:punct', '8:flat', '15:punct', '15:det', '8:appos', '18:case', '18:det', '15:nmod:at', '21:case', '21:det', '18:nmod:in', '23:case', '21:nmod:of', '28:punct', '28:case', '28:det', '28:amod', '21:nmod:near', '1:punct'],
2558
  'misc': ['SpaceAfter=No', 'SpaceAfter=No', None, None, None, None, None, None, None, 'SpaceAfter=No', 'SpaceAfter=No', 'SpaceAfter=No', None, None, None, None, None, None, None, None, None, None, 'SpaceAfter=No', None, None, None, None, 'SpaceAfter=No', None],
 
2564
  *MWT Example (French):**
2565
 
2566
  ```python
2567
+ dataset = load_dataset("universal-dependencies/universal_dependencies", "fr_gsd", split="train")
2568
  # Find sentence with MWT
2569
  example = [ex for ex in dataset if ex['mwt']][0]
2570
  print(example['mwt'])
 
2582
  from huggingface_hub import hf_hub_download
2583
  import json
2584
 
2585
+ md = hf_hub_download(repo_id="universal-dependencies/universal_dependencies", filename="metadata.json", repo_type="dataset")
2586
 
2587
  with open(md, "r", encoding="utf-8") as f:
2588
  metadata = json.load(f)
 
2638
 
2639
  ### Dataset Curators
2640
 
2641
+ Universal Dependencies is maintained by the Universal Dependencies community.
2642
+ This Hugging Face Parquet distribution is maintained under the official
2643
+ `universal-dependencies` organization.
2644
 
2645
  ### Licensing Information
2646
  Universal Dependencies is a collection of linguistic data and tools. Each of
 
2655
 
2656
  ### Citation Information
2657
 
2658
+ Please cite the Universal Dependencies release used in your work. If you use a
2659
+ specific treebank, also cite the treebank-specific references listed in that
2660
+ treebank's upstream README when available. For general UD background, cite the
2661
+ UD v2 paper listed in the dataset description.
2662
+
2663
  ```bibtex
2664
  @misc{11234/1-6149,
2665
  title = {Universal Dependencies 2.18},
 
2673
 
2674
  ### Contributions
2675
 
2676
+ The v2.0 Parquet rewrite was originally authored by Egon W. Stemle. This
2677
+ dataset was formerly hosted at `commul/universal_dependencies` and is now
2678
+ maintained at the official
2679
+ [`universal-dependencies/universal_dependencies`](https://huggingface.co/datasets/universal-dependencies/universal_dependencies)
2680
+ location.
tools/README.md CHANGED
@@ -43,9 +43,9 @@ you need to regenerate artifacts in all UD version branches.
43
 
44
  ```bash
45
  # Run from repository root
46
- export FRAMEWORK_VER=v2.1.1
47
 
48
- 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
49
  git switch "${UD_VER}"
50
  # Keep branch aligned with framework changes (pick one)
51
  git rebase main
 
43
 
44
  ```bash
45
  # Run from repository root
46
+ export FRAMEWORK_VER=v2.2.1
47
 
48
+ 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 2.18; do
49
  git switch "${UD_VER}"
50
  # Keep branch aligned with framework changes (pick one)
51
  git rebase main
tools/templates/README.tmpl CHANGED
@@ -1,13 +1,9 @@
1
  ---
2
- ### THIS IS A GENERATED FILE.
3
  viewer: true
4
-
5
  annotations_creators:
6
  - expert-generated
7
-
8
  language_creators:
9
  - crowdsourced
10
-
11
  language:
12
  {%- set languages = [] -%}
13
  {%- for name,metadata in data.items()|sort(attribute='1.dirname') -%}
@@ -16,30 +12,21 @@ language:
16
  {%- for language in languages|unique %}
17
  - {{ language if language != 'no' else "'no'" }}
18
  {%- endfor %}
19
-
20
- license:
21
- - apache-2.0
22
-
23
  multilinguality:
24
  - multilingual
25
-
26
  size_categories:
27
- - '1K<n<10K'
28
-
29
  source_datasets:
30
  - original
31
-
32
  task_categories:
33
  - token-classification
34
-
35
  task_ids:
36
  - parsing
37
  - part-of-speech
38
  - lemmatization
39
-
40
  paperswithcode_id: universal-dependencies
41
  pretty_name: Universal Dependencies Treebank
42
-
43
  tags:
44
  - text
45
  - constituency-parsing
@@ -69,7 +56,7 @@ configs:
69
  - **Parquet Format:** faster loading with HuggingFace datasets >=4.0.0
70
  - **MWT Support:** New `mwt` field provides structured multi-word token information
71
  - **Enhanced Security:** No more `trust_remote_code=True` required
72
- - **Separate Versioning:** Loader version (2.0.0) distinct from UD data version (2.17)
73
 
74
  **Breaking Changes:**
75
  - Token sequences now exclude MWT surface forms (matches UD guidelines)
@@ -110,16 +97,15 @@ configs:
110
  - **Repository:** [Universal Dependencies](https://github.com/UniversalDependencies)
111
  - **Paper:** [Universal Dependencies v2: An Evergrowing Multilingual Treebank Collection](https://arxiv.org/abs/2004.10643)
112
  - **Paper:** [From Treebank Metadata to Sentence-Level Genre in Universal Dependencies: A Reproducible, Versioned Resource](https://universaldependencies.org/udw26/papers/41_Paper.pdf)
113
- - **Leaderboard:**
114
- - **Point of Contact:** [appliedlinguisticsdevs@eurac.edu](mailto:appliedlinguisticsdevs@eurac.edu)
115
- - **Point of Contact:** [IAL Homepage](https://www.eurac.edu/linguistics)
116
 
117
  ### Dataset Summary
118
 
119
  {{ description }}
120
 
121
- This is a (temporary) fork of
122
- [/universal-dependencies/universal_dependencies](https://huggingface.co/datasets/universal-dependencies/universal_dependencies).
 
 
123
 
124
  ### Data Quality & Fidelity
125
 
@@ -146,7 +132,12 @@ For technical details, see [CONLLU_PARSING.md](https://github.com/bot-zen/ud-hf-
146
  from datasets import load_dataset
147
 
148
  # Load a specific treebank
149
- ds = load_dataset("commul/universal_dependencies", "en_ewt", revision="{{ ud_ver }}", split="train")
 
 
 
 
 
150
 
151
  # Access sentence data
152
  sentence = ds[0]
@@ -154,11 +145,8 @@ print(f"Sentence ID: {sentence['sent_id']}")
154
  print(f"Text: {sentence['text']}")
155
  print(f"Tokens: {sentence['tokens']}")
156
 
157
- ## TODO: Make helper functions available
158
- ## post v2.0 universal_dependencies.py is not part of the Dataset any longer!
159
- ##
160
  # Parse optional fields using helper functions
161
- from universal_dependencies import parse_feats, parse_misc
162
 
163
  for i, token in enumerate(sentence['tokens']):
164
  feats = parse_feats(sentence['feats'][i]) # Returns dict or {}
@@ -166,7 +154,7 @@ for i, token in enumerate(sentence['tokens']):
166
  print(f"{token}: UPOS={sentence['upos'][i]}, feats={feats}, misc={misc}")
167
 
168
  # Export back to CoNLL-U format
169
- from universal_dependencies import write_conllu
170
 
171
  # Write to stdout
172
  write_conllu(ds)
@@ -183,7 +171,9 @@ conllu_text = buffer.getvalue()
183
 
184
  ### Supported Tasks and Leaderboards
185
 
186
- [More Information Needed]
 
 
187
 
188
  ## Dataset Structure
189
 
@@ -203,7 +193,7 @@ This dataset has {{ data.items()|length }} configurations (treebanks).
203
  from datasets import get_dataset_config_names, load_dataset
204
 
205
  # Get all available treebank configurations for revision="{{ ud_ver }}"
206
- configs = get_dataset_config_names("commul/universal_dependencies", revision="{{ ud_ver }}")
207
  print(f"Available treebanks: {len(configs)}")
208
 
209
  # Example configurations:
@@ -217,10 +207,10 @@ print(f"Available treebanks: {len(configs)}")
217
  # ]
218
 
219
  # Get the latest configurations
220
- get_dataset_config_names("commul/universal_dependencies")
221
 
222
  # Load a specific treebank
223
- dataset = load_dataset("commul/universal_dependencies", "en_ewt")
224
  print(dataset)
225
 
226
  # Output:
@@ -251,7 +241,7 @@ Each example in the dataset contains the following fields:
251
  - **upos** (list of strings): Universal POS tags
252
  - **xpos** (list of strings): Language-specific POS tags
253
  - **feats** (list of strings): Morphological features in UD format
254
- - **head** (list of strings): Head indices for dependency relations
255
  - **deprel** (list of strings): Dependency relation labels
256
  - **deps** (list of strings): Enhanced dependency graph
257
  - **misc** (list of strings): Miscellaneous annotations
@@ -267,7 +257,7 @@ Each example in the dataset contains the following fields:
267
  ```python
268
  from datasets import load_dataset
269
 
270
- dataset = load_dataset("commul/universal_dependencies", "en_ewt", split="train")
271
  print(dataset[0])
272
 
273
  # Output:
@@ -282,10 +272,10 @@ print(dataset[0])
282
  ],
283
  'tokens': ['Al', '-', 'Zaman', ':', 'American', 'forces', 'killed', 'Shaikh', 'Abdullah', 'al', '-', 'Ani', ',', 'the', 'preacher', 'at', 'the', 'mosque', 'in', 'the', 'town', 'of', 'Qaim', ',', 'near', 'the', 'Syrian', 'border', '.'],
284
  'lemmas': ['Al', '-', 'Zaman', ':', 'American', 'force', 'kill', 'Shaikh', 'Abdullah', 'al', '-', 'Ani', ',', 'the', 'preacher', 'at', 'the', 'mosque', 'in', 'the', 'town', 'of', 'Qaim', ',', 'near', 'the', 'Syrian', 'border', '.'],
285
- 'upos': [10, 1, 10, 1, 6, 0, 16, 10, 10, 10, 1, 10, 1, 8, 0, 2, 8, 0, 2, 8, 0, 2, 10, 1, 2, 8, 6, 0, 1],
286
  'xpos': ['NNP', 'HYPH', 'NNP', ':', 'JJ', 'NNS', 'VBD', 'NNP', 'NNP', 'NNP', 'HYPH', 'NNP', ',', 'DT', 'NN', 'IN', 'DT', 'NN', 'IN', 'DT', 'NN', 'IN', 'NNP', ',', 'IN', 'DT', 'JJ', 'NN', '.'],
287
  'feats': ['Number=Sing', None, 'Number=Sing', None, 'Degree=Pos', 'Number=Plur', 'Mood=Ind|Number=Plur|Person=3|Tense=Past|VerbForm=Fin', 'Number=Sing', 'Number=Sing', 'Number=Sing', None, 'Number=Sing', None, 'Definite=Def|PronType=Art', 'Number=Sing', None, 'Definite=Def|PronType=Art', 'Number=Sing', None, 'Definite=Def|PronType=Art', 'Number=Sing', None, 'Number=Sing', None, None, 'Definite=Def|PronType=Art', 'Degree=Pos', 'Number=Sing', None],
288
- 'head': ['0', '3', '1', '7', '6', '7', '1', '7', '8', '8', '12', '8', '15', '15', '8', '18', '18', '15', '21', '21', '18', '23', '21', '28', '28', '28', '28', '21', '1'],
289
  'deprel': ['root', 'punct', 'flat', 'punct', 'amod', 'nsubj', 'parataxis', 'obj', 'flat', 'flat', 'punct', 'flat', 'punct', 'det', 'appos', 'case', 'det', 'nmod', 'case', 'det', 'nmod', 'case', 'nmod', 'punct', 'case', 'det', 'amod', 'nmod', 'punct'],
290
  'deps': ['0:root', '3:punct', '1:flat', '7:punct', '6:amod', '7:nsubj', '1:parataxis', '7:obj', '8:flat', '8:flat', '12:punct', '8:flat', '15:punct', '15:det', '8:appos', '18:case', '18:det', '15:nmod:at', '21:case', '21:det', '18:nmod:in', '23:case', '21:nmod:of', '28:punct', '28:case', '28:det', '28:amod', '21:nmod:near', '1:punct'],
291
  'misc': ['SpaceAfter=No', 'SpaceAfter=No', None, None, None, None, None, None, None, 'SpaceAfter=No', 'SpaceAfter=No', 'SpaceAfter=No', None, None, None, None, None, None, None, None, None, None, 'SpaceAfter=No', None, None, None, None, 'SpaceAfter=No', None],
@@ -297,7 +287,7 @@ print(dataset[0])
297
  *MWT Example (French):**
298
 
299
  ```python
300
- dataset = load_dataset("commul/universal_dependencies", "fr_gsd", split="train")
301
  # Find sentence with MWT
302
  example = [ex for ex in dataset if ex['mwt']][0]
303
  print(example['mwt'])
@@ -315,7 +305,7 @@ The file `metadata.json` stores additional information about the data, for examp
315
  from huggingface_hub import hf_hub_download
316
  import json
317
 
318
- md = hf_hub_download(repo_id="commul/universal_dependencies", filename="metadata.json", repo_type="dataset")
319
 
320
  with open(md, "r", encoding="utf-8") as f:
321
  metadata = json.load(f)
@@ -371,7 +361,9 @@ with open(md, "r", encoding="utf-8") as f:
371
 
372
  ### Dataset Curators
373
 
374
- [More Information Needed]
 
 
375
 
376
  ### Licensing Information
377
  Universal Dependencies is a collection of linguistic data and tools. Each of
@@ -386,10 +378,19 @@ The `./tools/` are licensed under the [Apache-2.0](https://www.apache.org/licens
386
 
387
  ### Citation Information
388
 
 
 
 
 
 
389
  ```bibtex
390
  {{ citation }}
391
  ```
392
 
393
  ### Contributions
394
 
395
- Thanks to [universal-dependencies](https://huggingface.co/universal-dependencies) for [the original of this dataset](https://huggingface.co/datasets/universal-dependencies/universal_dependencies).
 
 
 
 
 
1
  ---
 
2
  viewer: true
 
3
  annotations_creators:
4
  - expert-generated
 
5
  language_creators:
6
  - crowdsourced
 
7
  language:
8
  {%- set languages = [] -%}
9
  {%- for name,metadata in data.items()|sort(attribute='1.dirname') -%}
 
12
  {%- for language in languages|unique %}
13
  - {{ language if language != 'no' else "'no'" }}
14
  {%- endfor %}
15
+ license: other
 
 
 
16
  multilinguality:
17
  - multilingual
 
18
  size_categories:
19
+ - 1M<n<10M
 
20
  source_datasets:
21
  - original
 
22
  task_categories:
23
  - token-classification
 
24
  task_ids:
25
  - parsing
26
  - part-of-speech
27
  - lemmatization
 
28
  paperswithcode_id: universal-dependencies
29
  pretty_name: Universal Dependencies Treebank
 
30
  tags:
31
  - text
32
  - constituency-parsing
 
56
  - **Parquet Format:** faster loading with HuggingFace datasets >=4.0.0
57
  - **MWT Support:** New `mwt` field provides structured multi-word token information
58
  - **Enhanced Security:** No more `trust_remote_code=True` required
59
+ - **Separate Versioning:** Loader version (2.0.0) distinct from UD data version ({{ ud_ver }})
60
 
61
  **Breaking Changes:**
62
  - Token sequences now exclude MWT surface forms (matches UD guidelines)
 
97
  - **Repository:** [Universal Dependencies](https://github.com/UniversalDependencies)
98
  - **Paper:** [Universal Dependencies v2: An Evergrowing Multilingual Treebank Collection](https://arxiv.org/abs/2004.10643)
99
  - **Paper:** [From Treebank Metadata to Sentence-Level Genre in Universal Dependencies: A Reproducible, Versioned Resource](https://universaldependencies.org/udw26/papers/41_Paper.pdf)
 
 
 
100
 
101
  ### Dataset Summary
102
 
103
  {{ description }}
104
 
105
+ This is the official Hugging Face Parquet distribution for Universal
106
+ Dependencies at
107
+ [`universal-dependencies/universal_dependencies`](https://huggingface.co/datasets/universal-dependencies/universal_dependencies).
108
+ The dataset was formerly maintained at `commul/universal_dependencies`.
109
 
110
  ### Data Quality & Fidelity
111
 
 
132
  from datasets import load_dataset
133
 
134
  # Load a specific treebank
135
+ ds = load_dataset(
136
+ "universal-dependencies/universal_dependencies",
137
+ "en_ewt",
138
+ revision="{{ ud_ver }}",
139
+ split="train",
140
+ )
141
 
142
  # Access sentence data
143
  sentence = ds[0]
 
145
  print(f"Text: {sentence['text']}")
146
  print(f"Tokens: {sentence['tokens']}")
147
 
 
 
 
148
  # Parse optional fields using helper functions
149
+ from ud_hf_parquet_tools import parse_feats, parse_misc
150
 
151
  for i, token in enumerate(sentence['tokens']):
152
  feats = parse_feats(sentence['feats'][i]) # Returns dict or {}
 
154
  print(f"{token}: UPOS={sentence['upos'][i]}, feats={feats}, misc={misc}")
155
 
156
  # Export back to CoNLL-U format
157
+ from ud_hf_parquet_tools import write_conllu
158
 
159
  # Write to stdout
160
  write_conllu(ds)
 
171
 
172
  ### Supported Tasks and Leaderboards
173
 
174
+ This dataset supports dependency parsing, universal and language-specific
175
+ part-of-speech tagging, lemmatization, and morphology-oriented token
176
+ classification. No single leaderboard is maintained in this dataset card.
177
 
178
  ## Dataset Structure
179
 
 
193
  from datasets import get_dataset_config_names, load_dataset
194
 
195
  # Get all available treebank configurations for revision="{{ ud_ver }}"
196
+ configs = get_dataset_config_names("universal-dependencies/universal_dependencies", revision="{{ ud_ver }}")
197
  print(f"Available treebanks: {len(configs)}")
198
 
199
  # Example configurations:
 
207
  # ]
208
 
209
  # Get the latest configurations
210
+ get_dataset_config_names("universal-dependencies/universal_dependencies")
211
 
212
  # Load a specific treebank
213
+ dataset = load_dataset("universal-dependencies/universal_dependencies", "en_ewt")
214
  print(dataset)
215
 
216
  # Output:
 
241
  - **upos** (list of strings): Universal POS tags
242
  - **xpos** (list of strings): Language-specific POS tags
243
  - **feats** (list of strings): Morphological features in UD format
244
+ - **head** (list of integers): Head indices for dependency relations
245
  - **deprel** (list of strings): Dependency relation labels
246
  - **deps** (list of strings): Enhanced dependency graph
247
  - **misc** (list of strings): Miscellaneous annotations
 
257
  ```python
258
  from datasets import load_dataset
259
 
260
+ dataset = load_dataset("universal-dependencies/universal_dependencies", "en_ewt", split="train")
261
  print(dataset[0])
262
 
263
  # Output:
 
272
  ],
273
  'tokens': ['Al', '-', 'Zaman', ':', 'American', 'forces', 'killed', 'Shaikh', 'Abdullah', 'al', '-', 'Ani', ',', 'the', 'preacher', 'at', 'the', 'mosque', 'in', 'the', 'town', 'of', 'Qaim', ',', 'near', 'the', 'Syrian', 'border', '.'],
274
  'lemmas': ['Al', '-', 'Zaman', ':', 'American', 'force', 'kill', 'Shaikh', 'Abdullah', 'al', '-', 'Ani', ',', 'the', 'preacher', 'at', 'the', 'mosque', 'in', 'the', 'town', 'of', 'Qaim', ',', 'near', 'the', 'Syrian', 'border', '.'],
275
+ 'upos': ['PROPN', 'PUNCT', 'PROPN', 'PUNCT', 'ADJ', 'NOUN', 'VERB', ...],
276
  'xpos': ['NNP', 'HYPH', 'NNP', ':', 'JJ', 'NNS', 'VBD', 'NNP', 'NNP', 'NNP', 'HYPH', 'NNP', ',', 'DT', 'NN', 'IN', 'DT', 'NN', 'IN', 'DT', 'NN', 'IN', 'NNP', ',', 'IN', 'DT', 'JJ', 'NN', '.'],
277
  'feats': ['Number=Sing', None, 'Number=Sing', None, 'Degree=Pos', 'Number=Plur', 'Mood=Ind|Number=Plur|Person=3|Tense=Past|VerbForm=Fin', 'Number=Sing', 'Number=Sing', 'Number=Sing', None, 'Number=Sing', None, 'Definite=Def|PronType=Art', 'Number=Sing', None, 'Definite=Def|PronType=Art', 'Number=Sing', None, 'Definite=Def|PronType=Art', 'Number=Sing', None, 'Number=Sing', None, None, 'Definite=Def|PronType=Art', 'Degree=Pos', 'Number=Sing', None],
278
+ 'head': [0, 3, 1, 7, 6, 7, 1, ...],
279
  'deprel': ['root', 'punct', 'flat', 'punct', 'amod', 'nsubj', 'parataxis', 'obj', 'flat', 'flat', 'punct', 'flat', 'punct', 'det', 'appos', 'case', 'det', 'nmod', 'case', 'det', 'nmod', 'case', 'nmod', 'punct', 'case', 'det', 'amod', 'nmod', 'punct'],
280
  'deps': ['0:root', '3:punct', '1:flat', '7:punct', '6:amod', '7:nsubj', '1:parataxis', '7:obj', '8:flat', '8:flat', '12:punct', '8:flat', '15:punct', '15:det', '8:appos', '18:case', '18:det', '15:nmod:at', '21:case', '21:det', '18:nmod:in', '23:case', '21:nmod:of', '28:punct', '28:case', '28:det', '28:amod', '21:nmod:near', '1:punct'],
281
  'misc': ['SpaceAfter=No', 'SpaceAfter=No', None, None, None, None, None, None, None, 'SpaceAfter=No', 'SpaceAfter=No', 'SpaceAfter=No', None, None, None, None, None, None, None, None, None, None, 'SpaceAfter=No', None, None, None, None, 'SpaceAfter=No', None],
 
287
  *MWT Example (French):**
288
 
289
  ```python
290
+ dataset = load_dataset("universal-dependencies/universal_dependencies", "fr_gsd", split="train")
291
  # Find sentence with MWT
292
  example = [ex for ex in dataset if ex['mwt']][0]
293
  print(example['mwt'])
 
305
  from huggingface_hub import hf_hub_download
306
  import json
307
 
308
+ md = hf_hub_download(repo_id="universal-dependencies/universal_dependencies", filename="metadata.json", repo_type="dataset")
309
 
310
  with open(md, "r", encoding="utf-8") as f:
311
  metadata = json.load(f)
 
361
 
362
  ### Dataset Curators
363
 
364
+ Universal Dependencies is maintained by the Universal Dependencies community.
365
+ This Hugging Face Parquet distribution is maintained under the official
366
+ `universal-dependencies` organization.
367
 
368
  ### Licensing Information
369
  Universal Dependencies is a collection of linguistic data and tools. Each of
 
378
 
379
  ### Citation Information
380
 
381
+ Please cite the Universal Dependencies release used in your work. If you use a
382
+ specific treebank, also cite the treebank-specific references listed in that
383
+ treebank's upstream README when available. For general UD background, cite the
384
+ UD v2 paper listed in the dataset description.
385
+
386
  ```bibtex
387
  {{ citation }}
388
  ```
389
 
390
  ### Contributions
391
 
392
+ The v2.0 Parquet rewrite was originally authored by Egon W. Stemle. This
393
+ dataset was formerly hosted at `commul/universal_dependencies` and is now
394
+ maintained at the official
395
+ [`universal-dependencies/universal_dependencies`](https://huggingface.co/datasets/universal-dependencies/universal_dependencies)
396
+ location.