Datasets:

Modalities:
Text
Libraries:
Datasets
License:
gabrielaltay commited on
Commit
41a7e13
1 Parent(s): 7cafdf8

upload hubscripts/muchmore_hub.py to hub from bigbio repo

Browse files
Files changed (1) hide show
  1. muchmore.py +738 -0
muchmore.py ADDED
@@ -0,0 +1,738 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # coding=utf-8
2
+ # Copyright 2022 The HuggingFace Datasets Authors and the current dataset script contributor.
3
+ #
4
+ # Licensed under the Apache License, Version 2.0 (the "License");
5
+ # you may not use this file except in compliance with the License.
6
+ # You may obtain a copy of the License at
7
+ #
8
+ # http://www.apache.org/licenses/LICENSE-2.0
9
+ #
10
+ # Unless required by applicable law or agreed to in writing, software
11
+ # distributed under the License is distributed on an "AS IS" BASIS,
12
+ # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13
+ # See the License for the specific language governing permissions and
14
+ # limitations under the License.
15
+
16
+ """
17
+ A dataset loader for the MuchMore Springer Bilingual Corpus
18
+
19
+ homepage
20
+
21
+ * https://muchmore.dfki.de/resources1.htm
22
+
23
+ description of annotation format
24
+
25
+ * https://muchmore.dfki.de/pubs/D4.1.pdf
26
+
27
+ Four files are distributed
28
+
29
+ * springer_english_train_plain.tar.gz (english plain text of abstracts)
30
+ * springer_german_train_plain.tar.gz (german plain text of abstracts)
31
+ * springer_english_train_V4.2.tar.gz (annotated xml in english)
32
+ * springer_german_train_V4.2.tar.gz (annotated xml in german)
33
+
34
+ Each tar file has one member file per abstract.
35
+ There are keys to join the english and german files
36
+ but there is not a 1-1 mapping between them (i.e. some
37
+ english files have no german counterpart and some german
38
+ files have no english counterpart). However, there is a 1-1
39
+ mapping between plain text and annotations for a given language
40
+ (i.e. an abstract in springer_english_train_plain.tar.gz will
41
+ also be found in springer_english_train_V4.2.tar.gz)
42
+
43
+ Counts,
44
+
45
+ * 15,631 total abstracts
46
+ * 7,823 english abstracts
47
+ * 7,808 german abstracts
48
+ * 6,374 matched (en/de) abstracts
49
+ * 1,449 english abstracts with no german
50
+ * 1,434 german abstracts with no english
51
+
52
+ Notes
53
+
54
+ * Arthroskopie.00130237.eng.abstr.chunkmorph.annotated.xml seems to be empty
55
+
56
+
57
+ * entity spans can overlap. an example from the first sample:
58
+
59
+ {'id': 'Arthroskopie.00130003.eng.abstr-s1-t1',
60
+ 'type': 'umlsterm',
61
+ 'text': ['posterior'],
62
+ 'offsets': [[4, 13]],
63
+ 'normalized': [{'db_name': 'UMLS', 'db_id': 'C0032009'}]},
64
+ {'id': 'Arthroskopie.00130003.eng.abstr-s1-t8',
65
+ 'type': 'umlsterm',
66
+ 'text': ['posterior cruciate ligament'],
67
+ 'offsets': [[4, 31]],
68
+ 'normalized': [{'db_name': 'UMLS', 'db_id': 'C0080039'}]},
69
+ {'id': 'Arthroskopie.00130003.eng.abstr-s1-t2',
70
+ 'type': 'umlsterm',
71
+ 'text': ['ligament'],
72
+ 'offsets': [[23, 31]],
73
+ 'normalized': [{'db_name': 'UMLS', 'db_id': 'C0023685'},
74
+ {'db_name': 'UMLS', 'db_id': 'C0023686'}]},
75
+
76
+
77
+ * semantic relations are defined beween concepts but entities can
78
+ have multiple concpets associated with them. in the bigbio
79
+ schema we skip relations between multiple concept of the
80
+ same entity. an example of a relation that is kept from the
81
+ source schema is below,
82
+
83
+ In [35]: dsd['train'][0]['sentences'][0]['tokens']
84
+ Out[35]:
85
+ [{'id': 'w1', 'pos': 'DT', 'lemma': 'the', 'text': 'The'},
86
+ {'id': 'w2', 'pos': 'JJ', 'lemma': 'posterior', 'text': 'posterior'},
87
+ {'id': 'w3', 'pos': 'JJ', 'lemma': 'cruciate', 'text': 'cruciate'},
88
+ {'id': 'w4', 'pos': 'NN', 'lemma': 'ligament', 'text': 'ligament'},
89
+ {'id': 'w5', 'pos': 'PUNCT', 'lemma': None, 'text': '('},
90
+ {'id': 'w6', 'pos': 'NN', 'lemma': None, 'text': 'PCL'},
91
+ {'id': 'w7', 'pos': 'PUNCT', 'lemma': None, 'text': ')'},
92
+ {'id': 'w8', 'pos': 'VBZ', 'lemma': 'be', 'text': 'is'},
93
+ {'id': 'w9', 'pos': 'DT', 'lemma': 'the', 'text': 'the'},
94
+ {'id': 'w10', 'pos': 'JJS', 'lemma': 'strong', 'text': 'strongest'},
95
+ {'id': 'w11', 'pos': 'NN', 'lemma': 'ligament', 'text': 'ligament'},
96
+ {'id': 'w12', 'pos': 'IN', 'lemma': 'of', 'text': 'of'},
97
+ {'id': 'w13', 'pos': 'DT', 'lemma': 'the', 'text': 'the'},
98
+ {'id': 'w14', 'pos': 'JJ', 'lemma': 'human', 'text': 'human'},
99
+ {'id': 'w15', 'pos': 'NN', 'lemma': 'knee', 'text': 'knee'},
100
+ {'id': 'w16', 'pos': 'JJ', 'lemma': 'joint', 'text': 'joint'},
101
+ {'id': 'w17', 'pos': 'PUNCT', 'lemma': None, 'text': '.'}]
102
+
103
+
104
+ In [36]: dsd['train'][0]['sentences'][0]['semrels'][0]
105
+ Out[36]: {'id': 'r1', 'term1': 't3.1', 'term2': 't6.1', 'reltype': 'surrounds'}
106
+
107
+ In [37]: dsd['train'][0]['sentences'][0]['umlsterms'][2]
108
+ Out[37]:
109
+ {'id': 't3',
110
+ 'from': 'w11',
111
+ 'to': 'w11',
112
+ 'concepts': [{'id': 't3.1',
113
+ 'cui': 'C0023685',
114
+ 'preferred': 'Ligaments',
115
+ 'tui': 'T024',
116
+ 'mshs': [{'code': 'A2.513'}]},
117
+ {'id': 't3.2',
118
+ 'cui': 'C0023686',
119
+ 'preferred': 'Articular ligaments',
120
+ 'tui': 'T023',
121
+ 'mshs': [{'code': 'A2.513.514'}, {'code': 'A2.835.583.512'}]}]}
122
+
123
+ In [38]: dsd['train'][0]['sentences'][0]['umlsterms'][5]
124
+ Out[38]:
125
+ {'id': 't6',
126
+ 'from': 'w16',
127
+ 'to': 'w16',
128
+ 'concepts': [{'id': 't6.1',
129
+ 'cui': 'C0022417',
130
+ 'preferred': 'Joints',
131
+ 'tui': 'T030',
132
+ 'mshs': [{'code': 'A2.835.583'}]}]}
133
+
134
+ """
135
+
136
+ import itertools
137
+ import os
138
+ import re
139
+ import tarfile
140
+ import xml.etree.ElementTree as ET
141
+ from collections import defaultdict
142
+ from typing import Dict, List
143
+ from xml.etree.ElementTree import Element
144
+
145
+ import datasets
146
+ from datasets import Features, Value
147
+
148
+ # TODO: home page has a list of publications but its not clear which to choose
149
+ # https://muchmore.dfki.de/papers1.htm
150
+ # to start, chose the one below.
151
+ # Buitelaar, Paul / Declerck, Thierry / Sacaleanu, Bogdan / Vintar, Spela / Raileanu, Diana / Crispi, Claudia: A Multi-Layered, XML-Based Approach to the Integration of Linguistic and Semantic Annotations. In: Proceedings of EACL 2003 Workshop on Language Technology and the Semantic Web (NLPXML’03), Budapest, Hungary, April 2003.
152
+ from .bigbiohub import kb_features
153
+ from .bigbiohub import BigBioConfig
154
+ from .bigbiohub import Tasks
155
+
156
+ _LANGUAGES = ['English', 'German']
157
+ _PUBMED = True
158
+ _LOCAL = False
159
+ _CITATION = """\
160
+ @inproceedings{buitelaar2003multi,
161
+ title={A multi-layered, xml-based approach to the integration of linguistic and semantic annotations},
162
+ author={Buitelaar, Paul and Declerck, Thierry and Sacaleanu, Bogdan and Vintar, {\v{S}}pela and Raileanu, Diana and Crispi, Claudia},
163
+ booktitle={Proceedings of EACL 2003 Workshop on Language Technology and the Semantic Web (NLPXML'03), Budapest, Hungary},
164
+ year={2003}
165
+ }
166
+ """
167
+
168
+ _DESCRIPTION = """\
169
+ The corpus used in the MuchMore project is a parallel corpus of English-German scientific
170
+ medical abstracts obtained from the Springer Link web site. The corpus consists
171
+ approximately of 1 million tokens for each language. Abstracts are from 41 medical
172
+ journals, each of which constitutes a relatively homogeneous medical sub-domain (e.g.
173
+ Neurology, Radiology, etc.). The corpus of downloaded HTML documents is normalized in
174
+ various ways, in order to produce a clean, plain text version, consisting of a title, abstract
175
+ and keywords. Additionally, the corpus was aligned on the sentence level.
176
+
177
+ Automatic (!) annotation includes: Part-of-Speech; Morphology (inflection and
178
+ decomposition); Chunks; Semantic Classes (UMLS: Unified Medical Language System,
179
+ MeSH: Medical Subject Headings, EuroWordNet); Semantic Relations from UMLS.
180
+ """
181
+
182
+ _DATASETNAME = "muchmore"
183
+ _DISPLAYNAME = "MuchMore"
184
+
185
+ _HOMEPAGE = "https://muchmore.dfki.de/resources1.htm"
186
+
187
+ # TODO: website says the following, but don't see a specific license
188
+ # TODO: add to FAQs about what to do in this situation.
189
+
190
+ # "The cross-lingual information access prototype system for the medical domain
191
+ # will be made publicly accessible through the internet. It provides access to
192
+ # multilingual information on the basis of a domain ontology and classification.
193
+ # For the main task of multilingual domain modelling, the project will focus
194
+ # on German and English. "
195
+ _LICENSE = 'License information unavailable'
196
+ _URLs = {
197
+ "muchmore_source": [
198
+ "https://muchmore.dfki.de/pubs/springer_english_train_plain.tar.gz",
199
+ "https://muchmore.dfki.de/pubs/springer_english_train_V4.2.tar.gz",
200
+ "https://muchmore.dfki.de/pubs/springer_german_train_plain.tar.gz",
201
+ "https://muchmore.dfki.de/pubs/springer_german_train_V4.2.tar.gz",
202
+ ],
203
+ "muchmore_bigbio_kb": [
204
+ "https://muchmore.dfki.de/pubs/springer_english_train_V4.2.tar.gz",
205
+ "https://muchmore.dfki.de/pubs/springer_german_train_V4.2.tar.gz",
206
+ ],
207
+ "muchmore_en_bigbio_kb": "https://muchmore.dfki.de/pubs/springer_english_train_V4.2.tar.gz",
208
+ "muchmore_de_bigbio_kb": "https://muchmore.dfki.de/pubs/springer_german_train_V4.2.tar.gz",
209
+ "plain": [
210
+ "https://muchmore.dfki.de/pubs/springer_english_train_plain.tar.gz",
211
+ "https://muchmore.dfki.de/pubs/springer_german_train_plain.tar.gz",
212
+ ],
213
+ "plain_en": "https://muchmore.dfki.de/pubs/springer_english_train_plain.tar.gz",
214
+ "plain_de": "https://muchmore.dfki.de/pubs/springer_german_train_plain.tar.gz",
215
+ "muchmore_bigbio_t2t": [
216
+ "https://muchmore.dfki.de/pubs/springer_english_train_plain.tar.gz",
217
+ "https://muchmore.dfki.de/pubs/springer_german_train_plain.tar.gz",
218
+ ],
219
+ }
220
+
221
+ # took version from annotated file names
222
+ _SOURCE_VERSION = "4.2.0"
223
+ _BIGBIO_VERSION = "1.0.0"
224
+ _SUPPORTED_TASKS = [
225
+ Tasks.TRANSLATION,
226
+ Tasks.NAMED_ENTITY_RECOGNITION,
227
+ Tasks.NAMED_ENTITY_DISAMBIGUATION,
228
+ Tasks.RELATION_EXTRACTION,
229
+ ]
230
+
231
+ NATIVE_ENCODING = "ISO-8859-1"
232
+ FILE_NAME_PATTERN = r"^(.+?)\.(eng|ger)\.abstr(\.chunkmorph\.annotated\.xml)?$"
233
+ LANG_MAP = {"eng": "en", "ger": "de"}
234
+
235
+
236
+ class MuchMoreDataset(datasets.GeneratorBasedBuilder):
237
+ """MuchMore Springer Bilingual Corpus"""
238
+
239
+ DEFAULT_CONFIG_NAME = "muchmore_source"
240
+ SOURCE_VERSION = datasets.Version(_SOURCE_VERSION)
241
+ BIGBIO_VERSION = datasets.Version(_BIGBIO_VERSION)
242
+
243
+ BUILDER_CONFIGS = [
244
+ BigBioConfig(
245
+ name="muchmore_source",
246
+ version=SOURCE_VERSION,
247
+ description="MuchMore source schema",
248
+ schema="source",
249
+ subset_id="muchmore",
250
+ ),
251
+ BigBioConfig(
252
+ name="muchmore_bigbio_kb",
253
+ version=BIGBIO_VERSION,
254
+ description="MuchMore simplified BigBio kb schema",
255
+ schema="bigbio_kb",
256
+ subset_id="muchmore",
257
+ ),
258
+ BigBioConfig(
259
+ name="muchmore_en_bigbio_kb",
260
+ version=BIGBIO_VERSION,
261
+ description="MuchMore simplified BigBio kb schema",
262
+ schema="bigbio_kb",
263
+ subset_id="muchmore_en",
264
+ ),
265
+ BigBioConfig(
266
+ name="muchmore_de_bigbio_kb",
267
+ version=BIGBIO_VERSION,
268
+ description="MuchMore simplified BigBio kb schema",
269
+ schema="bigbio_kb",
270
+ subset_id="muchmore_de",
271
+ ),
272
+ BigBioConfig(
273
+ name="muchmore_bigbio_t2t",
274
+ version=BIGBIO_VERSION,
275
+ description="MuchMore simplified BigBio translation schema",
276
+ schema="bigbio_t2t",
277
+ subset_id="muchmore",
278
+ ),
279
+ ]
280
+
281
+ # default config produces english annotations at the moment
282
+ def _info(self):
283
+
284
+ if self.config.schema == "source":
285
+ features = Features(
286
+ {
287
+ "sample_id": Value("string"),
288
+ "corresp": Value("string"),
289
+ "language": Value("string"),
290
+ "abstract": Value("string"),
291
+ "sentences": [
292
+ {
293
+ "id": Value("string"),
294
+ "corresp": Value("string"),
295
+ "umlsterms": [
296
+ {
297
+ "id": Value("string"),
298
+ "from": Value("string"),
299
+ "to": Value("string"),
300
+ "concepts": [
301
+ {
302
+ "id": Value("string"),
303
+ "cui": Value("string"),
304
+ "preferred": Value("string"),
305
+ "tui": Value("string"),
306
+ "mshs": [
307
+ {
308
+ "code": Value("string"),
309
+ }
310
+ ],
311
+ }
312
+ ],
313
+ }
314
+ ],
315
+ "ewnterms": [
316
+ {
317
+ "id": Value("string"),
318
+ "to": Value("string"),
319
+ "from": Value("string"),
320
+ "senses": [
321
+ {
322
+ "offset": Value("string"),
323
+ }
324
+ ],
325
+ }
326
+ ],
327
+ "semrels": [
328
+ {
329
+ "id": Value("string"),
330
+ "term1": Value("string"),
331
+ "term2": Value("string"),
332
+ "reltype": Value("string"),
333
+ }
334
+ ],
335
+ "chunks": [
336
+ {
337
+ "id": Value("string"),
338
+ "to": Value("string"),
339
+ "from": Value("string"),
340
+ "type": Value("string"),
341
+ }
342
+ ],
343
+ "tokens": [
344
+ {
345
+ "id": Value("string"),
346
+ "pos": Value("string"),
347
+ "lemma": Value("string"),
348
+ "text": Value("string"),
349
+ }
350
+ ],
351
+ }
352
+ ],
353
+ }
354
+ )
355
+
356
+ elif self.config.schema == "bigbio_kb":
357
+ features = kb_features
358
+
359
+ elif self.config.name in ("plain", "plain_en", "plain_de"):
360
+ features = Features(
361
+ {
362
+ "sample_id": Value("string"),
363
+ "sample_id_prefix": Value("string"),
364
+ "language": Value("string"),
365
+ "abstract": Value("string"),
366
+ }
367
+ )
368
+
369
+ elif self.config.schema == "bigbio_t2t":
370
+ features = text2text_features
371
+
372
+ return datasets.DatasetInfo(
373
+ description=_DESCRIPTION,
374
+ features=features,
375
+ supervised_keys=None,
376
+ homepage=_HOMEPAGE,
377
+ license=str(_LICENSE),
378
+ citation=_CITATION,
379
+ )
380
+
381
+ def _split_generators(self, dl_manager):
382
+ """Returns SplitGenerators."""
383
+ my_urls = _URLs[self.config.name]
384
+ data_dirs = dl_manager.download(my_urls)
385
+ # ensure that data_dirs is always a list of string paths
386
+ if isinstance(data_dirs, str):
387
+ data_dirs = [data_dirs]
388
+
389
+ return [
390
+ datasets.SplitGenerator(
391
+ name=datasets.Split.TRAIN,
392
+ gen_kwargs={
393
+ "file_names_and_pointers": itertools.chain(
394
+ *[dl_manager.iter_archive(data_dir) for data_dir in data_dirs]
395
+ ),
396
+ "split": "train",
397
+ },
398
+ ),
399
+ ]
400
+
401
+ @staticmethod
402
+ def _get_umlsterms_from_xsent(xsent: Element) -> List:
403
+ xumlsterms = xsent.find("./umlsterms")
404
+
405
+ umlsterms = []
406
+ for xumlsterm in xumlsterms.findall("./umlsterm"):
407
+
408
+ concepts = []
409
+ for xconcept in xumlsterm.findall("./concept"):
410
+
411
+ mshs = [
412
+ {"code": xmsh.get("code")} for xmsh in xconcept.findall("./msh")
413
+ ]
414
+
415
+ concept = {
416
+ "id": xconcept.get("id"),
417
+ "cui": xconcept.get("cui"),
418
+ "preferred": xconcept.get("preferred"),
419
+ "tui": xconcept.get("tui"),
420
+ "mshs": mshs,
421
+ }
422
+ concepts.append(concept)
423
+
424
+ umlsterm = {
425
+ "id": xumlsterm.get("id"),
426
+ "from": xumlsterm.get("from"),
427
+ "to": xumlsterm.get("to"),
428
+ "concepts": concepts,
429
+ }
430
+ umlsterms.append(umlsterm)
431
+
432
+ return umlsterms
433
+
434
+ @staticmethod
435
+ def _get_ewnterms_from_xsent(xsent: Element) -> List:
436
+ xewnterms = xsent.find("./ewnterms")
437
+
438
+ ewnterms = []
439
+ for xewnterm in xewnterms.findall("./ewnterm"):
440
+
441
+ senses = [
442
+ {"offset": xsense.get("offset")}
443
+ for xsense in xewnterm.findall("./sense")
444
+ ]
445
+
446
+ ewnterm = {
447
+ "id": xewnterm.get("id"),
448
+ "from": xewnterm.get("from"),
449
+ "to": xewnterm.get("to"),
450
+ "senses": senses,
451
+ }
452
+ ewnterms.append(ewnterm)
453
+
454
+ return ewnterms
455
+
456
+ @staticmethod
457
+ def _get_semrels_from_xsent(xsent: Element) -> List[Dict[str, str]]:
458
+ xsemrels = xsent.find("./semrels")
459
+ return [
460
+ {
461
+ "id": xsemrel.get("id"),
462
+ "term1": xsemrel.get("term1"),
463
+ "term2": xsemrel.get("term2"),
464
+ "reltype": xsemrel.get("reltype"),
465
+ }
466
+ for xsemrel in xsemrels.findall("./semrel")
467
+ ]
468
+
469
+ @staticmethod
470
+ def _get_chunks_from_xsent(xsent: Element) -> List[Dict[str, str]]:
471
+ xchunks = xsent.find("./chunks")
472
+ return [
473
+ {
474
+ "id": xchunk.get("id"),
475
+ "to": xchunk.get("to"),
476
+ "from": xchunk.get("from"),
477
+ "type": xchunk.get("type"),
478
+ }
479
+ for xchunk in xchunks.findall("./chunk")
480
+ ]
481
+
482
+ @staticmethod
483
+ def _get_tokens_from_xsent(xsent: Element) -> List[Dict[str, str]]:
484
+ xtext = xsent.find("./text")
485
+ return [
486
+ {
487
+ "id": xtoken.get("id"),
488
+ "pos": xtoken.get("pos"),
489
+ "lemma": xtoken.get("lemma"),
490
+ "text": xtoken.text,
491
+ }
492
+ for xtoken in xtext.findall("./token")
493
+ ]
494
+
495
+ def _generate_original_examples(self, file_names_and_pointers):
496
+ """Generate something close to the original dataset.
497
+
498
+ This will yield one sample per abstract with the plaintext
499
+ and the annotations combined into one object. If an abstract
500
+ is available in both english and german each language version
501
+ will be a distinct example.
502
+ """
503
+ abstracts = {}
504
+ samples = {}
505
+ for file_name, fp in file_names_and_pointers:
506
+
507
+ if file_name.endswith(".abstr"):
508
+ sample_id = file_name
509
+ abstracts[sample_id] = fp.read().decode(NATIVE_ENCODING)
510
+
511
+ elif file_name.endswith(".abstr.chunkmorph.annotated.xml"):
512
+ content_bytes = fp.read()
513
+ content_str = content_bytes.decode(NATIVE_ENCODING)
514
+ if content_str == "":
515
+ continue
516
+
517
+ xroot = ET.fromstring(content_str)
518
+
519
+ sentences = []
520
+ for xsent in xroot.findall("./"):
521
+ sentence = {
522
+ "id": xsent.get("id"),
523
+ "corresp": xsent.get("corresp"),
524
+ "umlsterms": self._get_umlsterms_from_xsent(xsent),
525
+ "ewnterms": self._get_ewnterms_from_xsent(xsent),
526
+ "semrels": self._get_semrels_from_xsent(xsent),
527
+ "chunks": self._get_chunks_from_xsent(xsent),
528
+ "tokens": self._get_tokens_from_xsent(xsent),
529
+ }
530
+ sentences.append(sentence)
531
+
532
+ sample_id = xroot.get("id")
533
+ samples[sample_id] = {
534
+ "sample_id": sample_id,
535
+ "corresp": xroot.get("corresp"),
536
+ "language": xroot.get("lang"),
537
+ "sentences": sentences,
538
+ }
539
+
540
+ for _id, (sample_id, sample) in enumerate(samples.items()):
541
+ sample["abstract"] = abstracts[sample_id]
542
+ yield _id, sample
543
+
544
+ def _generate_bigbio_kb_examples(self, file_names_and_pointers):
545
+ """Generate big science biomedical kb examples."""
546
+
547
+ def snippets_tokens_from_sents(sentences):
548
+ snippets = []
549
+ for sentence in sentences:
550
+ snippet = [el["text"] for el in sentence["tokens"]]
551
+ snippets.append(snippet)
552
+ return snippets
553
+
554
+ def sid_to_text_off(sid, snip_txts_lens):
555
+ ii_sid = int(sid[1:])
556
+ start = sum(snip_txts_lens[: ii_sid - 1]) + (ii_sid - 1)
557
+ end = start + snip_txts_lens[ii_sid - 1]
558
+ return start, end
559
+
560
+ def sid_wid_to_text_off(sid, wid, snip_txts_lens, snip_toks_lens):
561
+ s_start, s_end = sid_to_text_off(sid, snip_txts_lens)
562
+ ii_sid = int(sid[1:])
563
+ ii_wid = int(wid[1:])
564
+ w_start = sum(snip_toks_lens[ii_sid - 1][: ii_wid - 1]) + (ii_wid - 1)
565
+ start = s_start + w_start
566
+ end = start + snip_toks_lens[ii_sid - 1][ii_wid - 1]
567
+ return start, end
568
+
569
+ for _id, (file_name, fp) in enumerate(file_names_and_pointers):
570
+
571
+ content_bytes = fp.read()
572
+ content_str = content_bytes.decode(NATIVE_ENCODING)
573
+ if content_str == "":
574
+ continue
575
+
576
+ xroot = ET.fromstring(content_str)
577
+
578
+ sentences = []
579
+ for xsent in xroot.findall("./"):
580
+ sentence = {
581
+ "id": xsent.get("id"),
582
+ "corresp": xsent.get("corresp"),
583
+ "umlsterms": self._get_umlsterms_from_xsent(xsent),
584
+ "ewnterms": self._get_ewnterms_from_xsent(xsent),
585
+ "semrels": self._get_semrels_from_xsent(xsent),
586
+ "chunks": self._get_chunks_from_xsent(xsent),
587
+ "tokens": self._get_tokens_from_xsent(xsent),
588
+ }
589
+ sentences.append(sentence)
590
+
591
+ snip_toks = snippets_tokens_from_sents(sentences)
592
+ snip_txts = [" ".join(snip_tok) for snip_tok in snip_toks]
593
+ snip_txts_lens = [len(el) for el in snip_txts]
594
+ snip_toks_lens = [[len(tok) for tok in snip] for snip in snip_toks]
595
+ text = " ".join(snip_txts)
596
+ passages = [
597
+ {
598
+ "id": "{}-passage-0".format(xroot.get("id")),
599
+ "type": "abstract",
600
+ "text": [text],
601
+ "offsets": [(0, len(text))],
602
+ }
603
+ ]
604
+
605
+ entities = []
606
+ rel_map = {}
607
+ for sentence in sentences:
608
+ sid = sentence["id"]
609
+ ii_sid = int(sid[1:])
610
+
611
+ for umlsterm in sentence["umlsterms"]:
612
+ umlsterm_id = umlsterm["id"]
613
+ entity_id = f"{sid}-{umlsterm_id}"
614
+ wid_from = umlsterm["from"]
615
+ wid_to = umlsterm["to"]
616
+ ii_wid_from = int(wid_from[1:])
617
+ ii_wid_to = int(wid_to[1:])
618
+
619
+ tok_text = " ".join(
620
+ snip_toks[ii_sid - 1][ii_wid_from - 1 : ii_wid_to]
621
+ )
622
+ w_from_start, w_from_end = sid_wid_to_text_off(
623
+ sid, wid_from, snip_txts_lens, snip_toks_lens
624
+ )
625
+ w_to_start, w_to_end = sid_wid_to_text_off(
626
+ sid, wid_to, snip_txts_lens, snip_toks_lens
627
+ )
628
+
629
+ offsets = [(w_from_start, w_to_end)]
630
+ main_text = text[w_from_start:w_to_end]
631
+ umls_cuis = [el["cui"] for el in umlsterm["concepts"]]
632
+ for concept in umlsterm["concepts"]:
633
+ rel_map[concept["id"]] = entity_id
634
+
635
+ entity = {
636
+ "id": "{}-{}".format(xroot.get("id"), entity_id),
637
+ "offsets": offsets,
638
+ "text": [tok_text],
639
+ "type": "umlsterm",
640
+ "normalized": [
641
+ {"db_name": "UMLS", "db_id": cui} for cui in umls_cuis
642
+ ],
643
+ }
644
+ entities.append(entity)
645
+
646
+ relations = []
647
+ for sentence in sentences:
648
+ sid = sentence["id"]
649
+ for semrel in sentence["semrels"]:
650
+ semrel_id = semrel["id"]
651
+ rel_id = "{}-{}-{}-{}".format(
652
+ sid, semrel_id, semrel["term1"], semrel["term2"],
653
+ )
654
+ arg1_id = "{}-{}".format(xroot.get("id"), rel_map[semrel["term1"]])
655
+ arg2_id = "{}-{}".format(xroot.get("id"), rel_map[semrel["term2"]])
656
+ # some semrels are between multiple normalizations of
657
+ # a single entity. we skip these. see docstring at top
658
+ # of module for more complete description
659
+ if arg1_id == arg2_id:
660
+ continue
661
+ relation = {
662
+ "id": "{}-{}".format(xroot.get("id"), rel_id),
663
+ "type": semrel["reltype"],
664
+ "arg1_id": arg1_id,
665
+ "arg2_id": arg2_id,
666
+ "normalized": []
667
+ }
668
+ relations.append(relation)
669
+
670
+ yield _id, {
671
+ "id": xroot.get("id"),
672
+ "document_id": xroot.get("id"),
673
+ "passages": passages,
674
+ "entities": entities,
675
+ "coreferences": [],
676
+ "events": [],
677
+ "relations": relations,
678
+ }
679
+
680
+ def _generate_plain_examples(self, file_names_and_pointers):
681
+ """Generate plain text abstract examples."""
682
+ for _id, (file_name, fp) in enumerate(file_names_and_pointers):
683
+ match = re.match(FILE_NAME_PATTERN, file_name)
684
+ yield _id, {
685
+ "sample_id_prefix": match.group(1),
686
+ "sample_id": file_name,
687
+ "language": LANG_MAP[match.group(2)],
688
+ "abstract": fp.read().decode(NATIVE_ENCODING),
689
+ }
690
+
691
+ def _generate_translation_examples(self, file_names_and_pointers):
692
+ sample_map = defaultdict(list)
693
+ for file_name, fp in file_names_and_pointers:
694
+ if file_name.endswith("eng.abstr"):
695
+ language = "en"
696
+ elif file_name.endswith("ger.abstr"):
697
+ language = "de"
698
+ else:
699
+ raise ValueError()
700
+ sample_id_prefix = re.sub(".(eng|ger).abstr$", "", file_name)
701
+ sample_id = file_name
702
+ abstract = fp.read().decode(NATIVE_ENCODING)
703
+ sample_map[sample_id_prefix].append(
704
+ {"language": language, "sample_id": sample_id, "abstract": abstract}
705
+ )
706
+
707
+ _id = 0
708
+ for sample_id_prefix, sample_pair in sample_map.items():
709
+ if len(sample_pair) != 2:
710
+ continue
711
+ en_idx = 0 if sample_pair[0]["language"] == "en" else 1
712
+ de_idx = 0 if en_idx == 1 else 1
713
+ yield _id, {
714
+ "id": sample_id_prefix,
715
+ "document_id": sample_id_prefix,
716
+ "text_1": sample_pair[en_idx]["abstract"],
717
+ "text_2": sample_pair[de_idx]["abstract"],
718
+ "text_1_name": "en",
719
+ "text_2_name": "de",
720
+ }
721
+ _id += 1
722
+
723
+ def _generate_examples(self, file_names_and_pointers, split):
724
+
725
+ if self.config.schema == "source":
726
+ genny = self._generate_original_examples(file_names_and_pointers)
727
+
728
+ elif self.config.schema == "bigbio_kb":
729
+ genny = self._generate_bigbio_kb_examples(file_names_and_pointers)
730
+
731
+ elif self.config.name in ("plain", "plain_en", "plain_de"):
732
+ genny = self._generate_plain_examples(file_names_and_pointers)
733
+
734
+ elif self.config.schema == "bigbio_t2t":
735
+ genny = self._generate_translation_examples(file_names_and_pointers)
736
+
737
+ for _id, sample in genny:
738
+ yield _id, sample