Datasets:

ArXiv:
License:
holylovenia commited on
Commit
c2da684
1 Parent(s): adde2ab

Upload unimorph.py with huggingface_hub

Browse files
Files changed (1) hide show
  1. unimorph.py +447 -0
unimorph.py ADDED
@@ -0,0 +1,447 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from pathlib import Path
2
+ from typing import Any, Dict, List, Tuple
3
+
4
+ import datasets
5
+ from datasets.download.download_manager import DownloadManager
6
+
7
+ from seacrowd.utils import schemas
8
+ from seacrowd.utils.configs import SEACrowdConfig
9
+ from seacrowd.utils.constants import Licenses, Tasks
10
+
11
+ _CITATION = """
12
+ @misc{batsuren2022unimorph,
13
+ title={UniMorph 4.0: Universal Morphology},
14
+ author={
15
+ Khuyagbaatar Batsuren and Omer Goldman and Salam Khalifa and Nizar
16
+ Habash and Witold Kieraś and Gábor Bella and Brian Leonard and Garrett
17
+ Nicolai and Kyle Gorman and Yustinus Ghanggo Ate and Maria Ryskina and
18
+ Sabrina J. Mielke and Elena Budianskaya and Charbel El-Khaissi and Tiago
19
+ Pimentel and Michael Gasser and William Lane and Mohit Raj and Matt
20
+ Coler and Jaime Rafael Montoya Samame and Delio Siticonatzi Camaiteri
21
+ and Benoît Sagot and Esaú Zumaeta Rojas and Didier López Francis and
22
+ Arturo Oncevay and Juan López Bautista and Gema Celeste Silva Villegas
23
+ and Lucas Torroba Hennigen and Adam Ek and David Guriel and Peter Dirix
24
+ and Jean-Philippe Bernardy and Andrey Scherbakov and Aziyana Bayyr-ool
25
+ and Antonios Anastasopoulos and Roberto Zariquiey and Karina Sheifer and
26
+ Sofya Ganieva and Hilaria Cruz and Ritván Karahóǧa and Stella
27
+ Markantonatou and George Pavlidis and Matvey Plugaryov and Elena
28
+ Klyachko and Ali Salehi and Candy Angulo and Jatayu Baxi and Andrew
29
+ Krizhanovsky and Natalia Krizhanovskaya and Elizabeth Salesky and Clara
30
+ Vania and Sardana Ivanova and Jennifer White and Rowan Hall Maudslay and
31
+ Josef Valvoda and Ran Zmigrod and Paula Czarnowska and Irene Nikkarinen
32
+ and Aelita Salchak and Brijesh Bhatt and Christopher Straughn and Zoey
33
+ Liu and Jonathan North Washington and Yuval Pinter and Duygu Ataman and
34
+ Marcin Wolinski and Totok Suhardijanto and Anna Yablonskaya and Niklas
35
+ Stoehr and Hossep Dolatian and Zahroh Nuriah and Shyam Ratan and Francis
36
+ M. Tyers and Edoardo M. Ponti and Grant Aiton and Aryaman Arora and
37
+ Richard J. Hatcher and Ritesh Kumar and Jeremiah Young and Daria
38
+ Rodionova and Anastasia Yemelina and Taras Andrushko and Igor Marchenko
39
+ and Polina Mashkovtseva and Alexandra Serova and Emily Prud'hommeaux and
40
+ Maria Nepomniashchaya and Fausto Giunchiglia and Eleanor Chodroff and
41
+ Mans Hulden and Miikka Silfverberg and Arya D. McCarthy and David
42
+ Yarowsky and Ryan Cotterell and Reut Tsarfaty and Ekaterina Vylomova},
43
+ year={2022},
44
+ eprint={2205.03608},
45
+ archivePrefix={arXiv},
46
+ primaryClass={cs.CL}
47
+ }
48
+ """
49
+
50
+ _LOCAL = False
51
+ _LANGUAGES = ["ind", "kod", "ceb", "hil", "tgl"]
52
+ _DATASETNAME = "unimorph"
53
+ _DESCRIPTION = """\
54
+ The Universal Morphology (UniMorph) project is a collaborative effort providing
55
+ broad-coverage instantiated normalized morphological inflection tables for
56
+ undreds of diverse world languages. The project comprises two major thrusts: a
57
+ language-independent feature schema for rich morphological annotation, and a
58
+ type-level resource of annotated data in diverse languages realizing that
59
+ schema. 5 Austronesian languages spoken in Southeast Asia, consisting 2
60
+ Malayo-Polynesian languages and 3 Greater Central Philippine languages, become
61
+ the part of UniMorph 4.0 release.
62
+ """
63
+
64
+ _HOMEPAGE = "https://unimorph.github.io"
65
+ _LICENSE = Licenses.CC_BY_SA_3_0.value
66
+ _URL = "https://raw.githubusercontent.com/unimorph/"
67
+
68
+ _SUPPORTED_TASKS = [Tasks.MORPHOLOGICAL_INFLECTION]
69
+ _SOURCE_VERSION = "4.0.0"
70
+ _SEACROWD_VERSION = "2024.06.20"
71
+
72
+
73
+ class UnimorphDataset(datasets.GeneratorBasedBuilder):
74
+ """Unimorh 4.0 dataset by Batsuren et al., (2022)"""
75
+
76
+ SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
77
+ SEACROWD_VERSION = datasets.Version(_SEACROWD_VERSION)
78
+
79
+ SEACROWD_SCHEMA_NAME = "pairs_multi"
80
+
81
+ dataset_names = sorted([f"{_DATASETNAME}_{lang}" for lang in _LANGUAGES])
82
+ BUILDER_CONFIGS = []
83
+ for name in dataset_names:
84
+ source_config = SEACrowdConfig(
85
+ name=f"{name}_source",
86
+ version=SOURCE_VERSION,
87
+ description=f"{_DATASETNAME} source schema",
88
+ schema="source",
89
+ subset_id=name,
90
+ )
91
+ BUILDER_CONFIGS.append(source_config)
92
+ seacrowd_config = SEACrowdConfig(
93
+ name=f"{name}_seacrowd_{SEACROWD_SCHEMA_NAME}",
94
+ version=SEACROWD_VERSION,
95
+ description=f"{_DATASETNAME} SEACrowd schema",
96
+ schema=f"seacrowd_{SEACROWD_SCHEMA_NAME}",
97
+ subset_id=name,
98
+ )
99
+ BUILDER_CONFIGS.append(seacrowd_config)
100
+
101
+ # Add configuration that allows loading all datasets at once.
102
+ BUILDER_CONFIGS.extend(
103
+ [
104
+ # unimorph_source
105
+ SEACrowdConfig(
106
+ name=f"{_DATASETNAME}_source",
107
+ version=SOURCE_VERSION,
108
+ description=f"{_DATASETNAME} source schema (all)",
109
+ schema="source",
110
+ subset_id=_DATASETNAME,
111
+ ),
112
+ # unimorph_seacrowd_pairs
113
+ SEACrowdConfig(
114
+ name=f"{_DATASETNAME}_seacrowd_{SEACROWD_SCHEMA_NAME}",
115
+ version=SEACROWD_VERSION,
116
+ description=f"{_DATASETNAME} SEACrowd schema (all)",
117
+ schema=f"seacrowd_{SEACROWD_SCHEMA_NAME}",
118
+ subset_id=_DATASETNAME,
119
+ ),
120
+ ]
121
+ )
122
+
123
+ DEFAULT_CONFIG_NAME = f"{_DATASETNAME}_source"
124
+ # https://huggingface.co/datasets/universal_morphologies/blob/main/universal_morphologies.py
125
+ CLASS_CATEGORIES = {
126
+ "Aktionsart": ["STAT", "DYN", "TEL", "ATEL", "PCT", "DUR", "ACH", "ACCMP", "SEMEL", "ACTY"],
127
+ "Animacy": ["ANIM", "INAN", "HUM", "NHUM"],
128
+ "Argument_Marking": [
129
+ "ARGNO1S",
130
+ "ARGNO2S",
131
+ "ARGNO3S",
132
+ "ARGNO1P",
133
+ "ARGNO2P",
134
+ "ARGNO3P",
135
+ "ARGAC1S",
136
+ "ARGAC2S",
137
+ "ARGAC3S",
138
+ "ARGAC1P",
139
+ "ARGAC2P",
140
+ "ARGAC3P",
141
+ "ARGAB1S",
142
+ "ARGAB2S",
143
+ "ARGAB3S",
144
+ "ARGAB1P",
145
+ "ARGAB2P",
146
+ "ARGAB3P",
147
+ "ARGER1S",
148
+ "ARGER2S",
149
+ "ARGER3S",
150
+ "ARGER1P",
151
+ "ARGER2P",
152
+ "ARGER3P",
153
+ "ARGDA1S",
154
+ "ARGDA2S",
155
+ "ARGDA3S",
156
+ "ARGDA1P",
157
+ "ARGDA2P",
158
+ "ARGDA3P",
159
+ "ARGBE1S",
160
+ "ARGBE2S",
161
+ "ARGBE3S",
162
+ "ARGBE1P",
163
+ "ARGBE2P",
164
+ "ARGBE3P",
165
+ ],
166
+ "Aspect": ["IPFV", "PFV", "PRF", "PROG", "PROSP", "ITER", "HAB"],
167
+ "Case": [
168
+ "NOM",
169
+ "ACC",
170
+ "ERG",
171
+ "ABS",
172
+ "NOMS",
173
+ "DAT",
174
+ "BEN",
175
+ "PRP",
176
+ "GEN",
177
+ "REL",
178
+ "PRT",
179
+ "INS",
180
+ "COM",
181
+ "VOC",
182
+ "COMPV",
183
+ "EQTV",
184
+ "PRIV",
185
+ "PROPR",
186
+ "AVR",
187
+ "FRML",
188
+ "TRANS",
189
+ "BYWAY",
190
+ "INTER",
191
+ "AT",
192
+ "POST",
193
+ "IN",
194
+ "CIRC",
195
+ "ANTE",
196
+ "APUD",
197
+ "ON",
198
+ "ONHR",
199
+ "ONVR",
200
+ "SUB",
201
+ "REM",
202
+ "PROXM",
203
+ "ESS",
204
+ "ALL",
205
+ "ABL",
206
+ "APPRX",
207
+ "TERM",
208
+ ],
209
+ "Comparison": ["CMPR", "SPRL", "AB", "RL", "EQT"],
210
+ "Definiteness": ["DEF", "INDF", "SPEC", "NSPEC"],
211
+ "Deixis": ["PROX", "MED", "REMT", "REF1", "REF2", "NOREF", "PHOR", "VIS", "NVIS", "ABV", "EVEN", "BEL"],
212
+ "Evidentiality": ["FH", "DRCT", "SEN", "VISU", "NVSEN", "AUD", "NFH", "QUOT", "RPRT", "HRSY", "INFER", "ASSUM"],
213
+ "Finiteness": ["FIN", "NFIN"],
214
+ "Gender": [
215
+ "MASC",
216
+ "FEM",
217
+ "NEUT",
218
+ "NAKH1",
219
+ "NAKH2",
220
+ "NAKH3",
221
+ "NAKH4",
222
+ "NAKH5",
223
+ "NAKH6",
224
+ "NAKH7",
225
+ "NAKH8",
226
+ "BANTU1",
227
+ "BANTU2",
228
+ "BANTU3",
229
+ "BANTU4",
230
+ "BANTU5",
231
+ "BANTU6",
232
+ "BANTU7",
233
+ "BANTU8",
234
+ "BANTU9",
235
+ "BANTU10",
236
+ "BANTU11",
237
+ "BANTU12",
238
+ "BANTU13",
239
+ "BANTU14",
240
+ "BANTU15",
241
+ "BANTU16",
242
+ "BANTU17",
243
+ "BANTU18",
244
+ "BANTU19",
245
+ "BANTU20",
246
+ "BANTU21",
247
+ "BANTU22",
248
+ "BANTU23",
249
+ ],
250
+ "Information_Structure": ["TOP", "FOC"],
251
+ "Interrogativity": ["DECL", "INT"],
252
+ "Language_Specific": [
253
+ "LGSPEC1",
254
+ "LGSPEC2",
255
+ "LGSPEC3",
256
+ "LGSPEC4",
257
+ "LGSPEC5",
258
+ "LGSPEC6",
259
+ "LGSPEC7",
260
+ "LGSPEC8",
261
+ "LGSPEC9",
262
+ "LGSPEC10",
263
+ ],
264
+ "Mood": [
265
+ "IND",
266
+ "SBJV",
267
+ "REAL",
268
+ "IRR",
269
+ "AUPRP",
270
+ "AUNPRP",
271
+ "IMP",
272
+ "COND",
273
+ "PURP",
274
+ "INTEN",
275
+ "POT",
276
+ "LKLY",
277
+ "ADM",
278
+ "OBLIG",
279
+ "DEB",
280
+ "PERM",
281
+ "DED",
282
+ "SIM",
283
+ "OPT",
284
+ ],
285
+ "Number": ["SG", "PL", "GRPL", "DU", "TRI", "PAUC", "GRPAUC", "INVN"],
286
+ "Part_Of_Speech": [
287
+ "N",
288
+ "PROPN",
289
+ "ADJ",
290
+ "PRO",
291
+ "CLF",
292
+ "ART",
293
+ "DET",
294
+ "V",
295
+ "ADV",
296
+ "AUX",
297
+ "V.PTCP",
298
+ "V.MSDR",
299
+ "V.CVB",
300
+ "ADP",
301
+ "COMP",
302
+ "CONJ",
303
+ "NUM",
304
+ "PART",
305
+ "INTJ",
306
+ ],
307
+ "Person": ["0", "1", "2", "3", "4", "INCL", "EXCL", "PRX", "OBV"],
308
+ "Polarity": ["POS", "NEG"],
309
+ "Politeness": [
310
+ "INFM",
311
+ "FORM",
312
+ "ELEV",
313
+ "HUMB",
314
+ "POL",
315
+ "AVOID",
316
+ "LOW",
317
+ "HIGH",
318
+ "STELEV",
319
+ "STSUPR",
320
+ "LIT",
321
+ "FOREG",
322
+ "COL",
323
+ ],
324
+ "Possession": [
325
+ "ALN",
326
+ "NALN",
327
+ "PSS1S",
328
+ "PSS2S",
329
+ "PSS2SF",
330
+ "PSS2SM",
331
+ "PSS2SINFM",
332
+ "PSS2SFORM",
333
+ "PSS3S",
334
+ "PSS3SF",
335
+ "PSS3SM",
336
+ "PSS1D",
337
+ "PSS1DI",
338
+ "PSS1DE",
339
+ "PSS2D",
340
+ "PSS2DM",
341
+ "PSS2DF",
342
+ "PSS3D",
343
+ "PSS3DF",
344
+ "PSS3DM",
345
+ "PSS1P",
346
+ "PSS1PI",
347
+ "PSS1PE",
348
+ "PSS2P",
349
+ "PSS2PF",
350
+ "PSS2PM",
351
+ "PSS3PF",
352
+ "PSS3PM",
353
+ ],
354
+ "Switch_Reference": ["SS", "SSADV", "DS", "DSADV", "OR", "SIMMA", "SEQMA", "LOG"],
355
+ "Tense": ["PRS", "PST", "FUT", "IMMED", "HOD", "1DAY", "RCT", "RMT"],
356
+ "Valency": ["IMPRS", "INTR", "TR", "DITR", "REFL", "RECP", "CAUS", "APPL"],
357
+ "Voice": ["ACT", "MID", "PASS", "ANTIP", "DIR", "INV", "AGFOC", "PFOC", "LFOC", "BFOC", "ACFOC", "IFOC", "CFOC"],
358
+ }
359
+
360
+ TAG_TO_CAT = dict([(tag, cat) for cat, tags in CLASS_CATEGORIES.items() for tag in tags])
361
+ CLASS_LABELS = [feat for _, category in CLASS_CATEGORIES.items() for feat in category]
362
+
363
+ def _info(self) -> datasets.DatasetInfo:
364
+ if self.config.schema == "source":
365
+ features = datasets.Features(
366
+ {
367
+ "lemma": datasets.Value("string"),
368
+ "forms": datasets.Sequence(
369
+ dict(
370
+ [("word", datasets.Value("string"))]
371
+ + [(cat, datasets.Sequence(datasets.ClassLabel(names=tasks))) for cat, tasks in self.CLASS_CATEGORIES.items()]
372
+ + [("Other", datasets.Sequence(datasets.Value("string")))] # for misspecified tags
373
+ )
374
+ ),
375
+ }
376
+ )
377
+
378
+ if self.config.schema == f"seacrowd_{self.SEACROWD_SCHEMA_NAME}":
379
+ all_features = [feat for _, category in self.CLASS_CATEGORIES.items() for feat in category]
380
+ features = schemas.pairs_multi_features(label_names=self.CLASS_LABELS)
381
+
382
+ return datasets.DatasetInfo(description=_DESCRIPTION, features=features, homepage=_HOMEPAGE, license=_LICENSE, citation=_CITATION)
383
+
384
+ def _split_generators(self, dl_manager: DownloadManager) -> List[datasets.SplitGenerator]:
385
+ """Return SplitGenerators."""
386
+ source_data = []
387
+
388
+ lang = self.config.name.split("_")[1]
389
+ if lang in _LANGUAGES:
390
+ # Load data per language
391
+ source_data.append(dl_manager.download_and_extract(_URL + f"{lang}/main/{lang}"))
392
+ else:
393
+ # Load examples from all languages at once.
394
+ for lang in _LANGUAGES:
395
+ source_data.append(dl_manager.download_and_extract(_URL + f"{lang}/main/{lang}"))
396
+
397
+ return [
398
+ datasets.SplitGenerator(
399
+ name=datasets.Split.TRAIN,
400
+ gen_kwargs={
401
+ "filepaths": source_data,
402
+ },
403
+ )
404
+ ]
405
+
406
+ def _generate_examples(self, filepaths: List[Path]) -> Tuple[int, Dict]:
407
+ """Yield examples as (key, example) tuples"""
408
+
409
+ all_forms: Dict[str, List[Dict[str, Any]]] = {}
410
+ for source_file in filepaths:
411
+ with open(source_file, encoding="utf-8") as file:
412
+ for row in file:
413
+ if row.strip() == "" or row.strip().startswith("#"):
414
+ continue
415
+ lemma, word, tags = row.strip().split("\t")
416
+ all_forms[lemma] = all_forms.get(lemma, [])
417
+ tag_list = tags.replace("NDEF", "INDF").split(";")
418
+ form = dict([("word", word), ("Other", [])] + [(cat, []) for cat, tasks in self.CLASS_CATEGORIES.items()])
419
+ for tag_pre in tag_list:
420
+ tag = tag_pre.split("+")
421
+ if tag[0] in self.TAG_TO_CAT:
422
+ form[self.TAG_TO_CAT[tag[0]]] = tag
423
+ else:
424
+ form["Other"] += tag
425
+ all_forms[lemma] += [form]
426
+
427
+ if self.config.schema == "source":
428
+ for id_, (lemma, forms) in enumerate(all_forms.items()):
429
+ res = {"lemma": lemma, "forms": {}}
430
+ for k in ["word", "Other"] + list(self.CLASS_CATEGORIES.keys()):
431
+ res["forms"][k] = [form[k] for form in forms]
432
+ yield id_, res
433
+
434
+ if self.config.schema == f"seacrowd_{self.SEACROWD_SCHEMA_NAME}":
435
+ idx = 0
436
+ for lemma, forms in all_forms.items():
437
+ for form in forms:
438
+ inflection = form.pop("word")
439
+ feats = [feat[0] for feat in list(form.values()) if feat and feat[0] in self.CLASS_LABELS]
440
+ example = {
441
+ "id": idx,
442
+ "text_1": lemma,
443
+ "text_2": inflection,
444
+ "label": feats,
445
+ }
446
+ idx += 1
447
+ yield idx, example