jfrenz commited on
Commit
b5640b6
1 Parent(s): 91bce18

legal glue py - multieurlex fix

Browse files
Files changed (1) hide show
  1. legalglue.py +85 -28
legalglue.py CHANGED
@@ -8257,9 +8257,9 @@ class LegalGlueConfig(datasets.BuilderConfig):
8257
  if language != "all_languages":
8258
  self.languages = [language]
8259
  else:
8260
- if self.name == swissJudgmentPrediction:
8261
  self.languages = languages if languages is not None else SWISS_LANG
8262
- if self.name == multi_eurlex:
8263
  self.languages = languages if languages is not None else MULTI_EURLEX_LANGUAGES
8264
 
8265
 
@@ -8343,7 +8343,7 @@ class LegalGLUE(datasets.GeneratorBasedBuilder):
8343
  language = "pt"
8344
  ),
8345
  LegalGlueConfig(
8346
- name="multi_eurlex",
8347
  description=textwrap.dedent(
8348
  """\
8349
  MultiEURLEX comprises 65k EU laws in 23 official EU languages (some low-ish resource).
@@ -8353,6 +8353,7 @@ class LegalGLUE(datasets.GeneratorBasedBuilder):
8353
  """
8354
  ),
8355
  label_classes=EUROVOC_CONCEPTS,
 
8356
  multi_label=True,
8357
  homepage="https://github.com/nlpaueb/multi-eurlex",
8358
  data_url="https://zenodo.org/record/5363165/files/multi_eurlex.tar.gz",
@@ -8373,7 +8374,7 @@ class LegalGLUE(datasets.GeneratorBasedBuilder):
8373
  )
8374
  ),
8375
  LegalGlueConfig(
8376
- name="swissJudgmentPrediction",
8377
  description=textwrap.dedent(
8378
  """
8379
  Swiss-Judgment-Prediction is a multilingual, diachronic dataset of 85K Swiss Federal Supreme Court (FSCS) cases annotated with the respective binarized judgment outcome (approval/dismissal), posing a challenging text classification task. We also provide additional metadata, i.e., the publication year, the legal area and the canton of origin per case, to promote robustness and fairness studies on the critical area of legal NLP.
@@ -8381,6 +8382,7 @@ class LegalGLUE(datasets.GeneratorBasedBuilder):
8381
  ),
8382
  label_classes=[0, 1],
8383
  multi_label=False,
 
8384
  data_url="https://zenodo.org/record/5529712/files/",
8385
  data_files=["train.jsonl", "val.jsonl", "test.jsonl"],
8386
  homepage="https://github.com/JoelNiklaus/SwissCourtRulingCorpus",
@@ -8396,30 +8398,85 @@ class LegalGLUE(datasets.GeneratorBasedBuilder):
8396
  }"""
8397
  ),
8398
  ),
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8399
  ]
8400
 
8401
  def _info(self):
8402
  if self.config.name == "german_ler":
8403
- features = {
8404
- "id": datasets.Value("string"),
8405
- "tokens": datasets.Sequence(datasets.Value("string")),
8406
- "ner_tags": datasets.Sequence(
8407
- datasets.features.ClassLabel(
8408
- names=self.config.label_classes
 
 
8409
  )
8410
- )
8411
- }
8412
  elif self.config.name == "lener_br":
8413
- features = {
8414
- "id": datasets.Value("string"),
8415
- "tokens": datasets.Sequence(datasets.Value("string")),
8416
- "ner_tags": datasets.Sequence(
8417
- datasets.features.ClassLabel(
8418
- names=self.config.label_classes
8419
- )
8420
- )
8421
- }
8422
- elif self.config.name == "swissJudgmentPrediction":
 
 
8423
  features = datasets.Features(
8424
  {
8425
  "id": datasets.Value("int32"),
@@ -8432,7 +8489,7 @@ class LegalGLUE(datasets.GeneratorBasedBuilder):
8432
  "legal area": datasets.Value("string"),
8433
  }
8434
  )
8435
- elif self.config.name == "multi_eurlex":
8436
  if self.config.language == "all_languages":
8437
  features = datasets.Features(
8438
  {
@@ -8457,7 +8514,7 @@ class LegalGLUE(datasets.GeneratorBasedBuilder):
8457
  )
8458
  return datasets.DatasetInfo(
8459
  description=self.config.description,
8460
- features=datasets.Features(features),
8461
  homepage=self.config.homepage,
8462
  citation=self.config.citation,
8463
  )
@@ -8507,7 +8564,7 @@ class LegalGLUE(datasets.GeneratorBasedBuilder):
8507
  },
8508
  ),
8509
  ]
8510
- elif self.config.name == "swissJudgmentPrediction":
8511
  urls_to_download = {
8512
  "train": self.config.data_url + "train.jsonl",
8513
  "test": self.config.data_url + "test.jsonl",
@@ -8531,7 +8588,7 @@ class LegalGLUE(datasets.GeneratorBasedBuilder):
8531
  gen_kwargs={"filepath": data_dir["val"], "split": "dev"},
8532
  ),
8533
  ]
8534
- elif self.config.name == "multi_eurlex":
8535
  data_dir = dl_manager.download_and_extract(self.config.data_url)
8536
  return [
8537
  datasets.SplitGenerator(
@@ -8610,7 +8667,7 @@ class LegalGLUE(datasets.GeneratorBasedBuilder):
8610
  "ner_tags": ner_tags,
8611
  }
8612
 
8613
- elif self.config.name =="swissJudgmentPrediction":
8614
  if self.config.language == "all_languages":
8615
  with open(filepath, encoding="utf-8") as f:
8616
  for id_, row in enumerate(f):
@@ -8641,7 +8698,7 @@ class LegalGLUE(datasets.GeneratorBasedBuilder):
8641
  "legal area": data["legal area"],
8642
  }
8643
 
8644
- elif self.config.name =="multieurlex":
8645
  if self.config.language == "all_languages":
8646
  with open(filepath, encoding="utf-8") as f:
8647
  for id_, row in enumerate(f):
8257
  if language != "all_languages":
8258
  self.languages = [language]
8259
  else:
8260
+ if "swissJudgmentPrediction" in self.name:
8261
  self.languages = languages if languages is not None else SWISS_LANG
8262
+ if "multi_eurlex" in self.name:
8263
  self.languages = languages if languages is not None else MULTI_EURLEX_LANGUAGES
8264
 
8265
 
8343
  language = "pt"
8344
  ),
8345
  LegalGlueConfig(
8346
+ name="multi_eurlex_all_languages",
8347
  description=textwrap.dedent(
8348
  """\
8349
  MultiEURLEX comprises 65k EU laws in 23 official EU languages (some low-ish resource).
8353
  """
8354
  ),
8355
  label_classes=EUROVOC_CONCEPTS,
8356
+ language="all_languages",
8357
  multi_label=True,
8358
  homepage="https://github.com/nlpaueb/multi-eurlex",
8359
  data_url="https://zenodo.org/record/5363165/files/multi_eurlex.tar.gz",
8374
  )
8375
  ),
8376
  LegalGlueConfig(
8377
+ name="swissJudgmentPrediction_all_languages",
8378
  description=textwrap.dedent(
8379
  """
8380
  Swiss-Judgment-Prediction is a multilingual, diachronic dataset of 85K Swiss Federal Supreme Court (FSCS) cases annotated with the respective binarized judgment outcome (approval/dismissal), posing a challenging text classification task. We also provide additional metadata, i.e., the publication year, the legal area and the canton of origin per case, to promote robustness and fairness studies on the critical area of legal NLP.
8382
  ),
8383
  label_classes=[0, 1],
8384
  multi_label=False,
8385
+ language="all_languages",
8386
  data_url="https://zenodo.org/record/5529712/files/",
8387
  data_files=["train.jsonl", "val.jsonl", "test.jsonl"],
8388
  homepage="https://github.com/JoelNiklaus/SwissCourtRulingCorpus",
8398
  }"""
8399
  ),
8400
  ),
8401
+ ] + [
8402
+ LegalGlueConfig(
8403
+ name="multi_eurlex" + "_" + lang,
8404
+ description=f"Plain text import of MultiEURLEX for the {lang} language",
8405
+ label_classes=EUROVOC_CONCEPTS,
8406
+ language=lang,
8407
+ multi_label=True,
8408
+ homepage="https://github.com/nlpaueb/multi-eurlex",
8409
+ data_url="https://zenodo.org/record/5363165/files/multi_eurlex.tar.gz",
8410
+ data_files=["train.jsonl", "test.jsonl", "dev.jsonl"],
8411
+ citation=textwrap.dedent("""\
8412
+ @InProceedings{chalkidis-etal-2021-multieurlex,
8413
+ author = {Chalkidis, Ilias
8414
+ and Fergadiotis, Manos
8415
+ and Androutsopoulos, Ion},
8416
+ title = {MultiEURLEX -- A multi-lingual and multi-label legal document
8417
+ classification dataset for zero-shot cross-lingual transfer},
8418
+ booktitle = {Proceedings of the 2021 Conference on Empirical Methods
8419
+ in Natural Language Processing},
8420
+ year = {2021},
8421
+ publisher = {Association for Computational Linguistics},
8422
+ location = {Punta Cana, Dominican Republic},
8423
+ }"""
8424
+ )
8425
+ )for lang in MULTI_EURLEX_LANGUAGES
8426
+ ] + [
8427
+ LegalGlueConfig(
8428
+ name="swissJudgmentPrediction" + "_"+ lang,
8429
+ description=textwrap.dedent(
8430
+ """
8431
+ Swiss-Judgment-Prediction is a multilingual, diachronic dataset of 85K Swiss Federal Supreme Court (FSCS) cases annotated with the respective binarized judgment outcome (approval/dismissal), posing a challenging text classification task. We also provide additional metadata, i.e., the publication year, the legal area and the canton of origin per case, to promote robustness and fairness studies on the critical area of legal NLP.
8432
+ """
8433
+ ),
8434
+ label_classes=[0, 1],
8435
+ language=lang,
8436
+ multi_label=False,
8437
+ data_url="https://zenodo.org/record/5529712/files/",
8438
+ data_files=["train.jsonl", "val.jsonl", "test.jsonl"],
8439
+ homepage="https://github.com/JoelNiklaus/SwissCourtRulingCorpus",
8440
+ citation=textwrap.dedent("""\
8441
+ @InProceedings{niklaus-etal-2021-swiss,
8442
+ author = {Niklaus, Joel
8443
+ and Chalkidis, Ilias
8444
+ and Stürmer, Matthias},
8445
+ title = {Swiss-Court-Predict: A Multilingual Legal Judgment Prediction Benchmark},
8446
+ booktitle = {Proceedings of the 2021 Natural Legal Language Processing Workshop},
8447
+ year = {2021},
8448
+ location = {Punta Cana, Dominican Republic},
8449
+ }"""
8450
+ ),
8451
+ )for lang in SWISS_LANG
8452
  ]
8453
 
8454
  def _info(self):
8455
  if self.config.name == "german_ler":
8456
+ features = datasets.Features(
8457
+ {
8458
+ "id": datasets.Value("string"),
8459
+ "tokens": datasets.Sequence(datasets.Value("string")),
8460
+ "ner_tags": datasets.Sequence(
8461
+ datasets.features.ClassLabel(
8462
+ names=self.config.label_classes
8463
+ )
8464
  )
8465
+ }
8466
+ )
8467
  elif self.config.name == "lener_br":
8468
+ features = datasets.Features(
8469
+ {
8470
+ "id": datasets.Value("string"),
8471
+ "tokens": datasets.Sequence(datasets.Value("string")),
8472
+ "ner_tags": datasets.Sequence(
8473
+ datasets.features.ClassLabel(
8474
+ names=self.config.label_classes
8475
+ )
8476
+ )
8477
+ }
8478
+ )
8479
+ elif "swissJudgmentPrediction" in self.config.name:
8480
  features = datasets.Features(
8481
  {
8482
  "id": datasets.Value("int32"),
8489
  "legal area": datasets.Value("string"),
8490
  }
8491
  )
8492
+ elif "multi_eurlex" in self.config.name :
8493
  if self.config.language == "all_languages":
8494
  features = datasets.Features(
8495
  {
8514
  )
8515
  return datasets.DatasetInfo(
8516
  description=self.config.description,
8517
+ features=features,
8518
  homepage=self.config.homepage,
8519
  citation=self.config.citation,
8520
  )
8564
  },
8565
  ),
8566
  ]
8567
+ elif "swissJudgmentPrediction" in self.config.name:
8568
  urls_to_download = {
8569
  "train": self.config.data_url + "train.jsonl",
8570
  "test": self.config.data_url + "test.jsonl",
8588
  gen_kwargs={"filepath": data_dir["val"], "split": "dev"},
8589
  ),
8590
  ]
8591
+ elif "multi_eurlex" in self.config.name:
8592
  data_dir = dl_manager.download_and_extract(self.config.data_url)
8593
  return [
8594
  datasets.SplitGenerator(
8667
  "ner_tags": ner_tags,
8668
  }
8669
 
8670
+ elif "swissJudgmentPrediction" in self.config.name:
8671
  if self.config.language == "all_languages":
8672
  with open(filepath, encoding="utf-8") as f:
8673
  for id_, row in enumerate(f):
8698
  "legal area": data["legal area"],
8699
  }
8700
 
8701
+ elif "multi_eurlex" in self.config.name:
8702
  if self.config.language == "all_languages":
8703
  with open(filepath, encoding="utf-8") as f:
8704
  for id_, row in enumerate(f):