Datasets:
wmt
/

Modalities:
Text
Formats:
parquet
Libraries:
Datasets
Dask
License:
albertvillanova HF staff commited on
Commit
41d8a40
1 Parent(s): 62cb12f

Convert dataset to Parquet (#8)

Browse files

- Convert dataset to Parquet (53b63b33fa51d9b154b459781e36c3ca4bdf625b)
- Add de-en data files (c40135dc86c3fd218a79cb675ad3149aa68fc51a)
- Add fi-en data files (fae23fd9f29c31153f2f6f966d6376917c8a2d63)
- Add ro-en data files (92a91d8b49104121b3d503f869db0e4c27f50d33)
- Add ru-en data files (742f1e314d7a73ef2af55e57f39e9b72f1b1a2d7)
- Add tr-en data files (75bb1cfaae2aa014f64bc171de536538aa9aac19)
- Delete loading script (3923157a85dd2f15031c2b4657a7ba7193237544)
- Delete loading script auxiliary file (049ba6fb74fdfa56f8e3605f633ef35afc86cb10)

README.md CHANGED
@@ -25,8 +25,8 @@ source_datasets:
25
  task_categories:
26
  - translation
27
  task_ids: []
28
- pretty_name: WMT16
29
  paperswithcode_id: wmt-2016
 
30
  dataset_info:
31
  - config_name: cs-en
32
  features:
@@ -38,16 +38,16 @@ dataset_info:
38
  - en
39
  splits:
40
  - name: train
41
- num_bytes: 296006386
42
  num_examples: 997240
43
  - name: validation
44
- num_bytes: 572203
45
  num_examples: 2656
46
  - name: test
47
- num_bytes: 707870
48
  num_examples: 2999
49
- download_size: 1690726387
50
- dataset_size: 297286459
51
  - config_name: de-en
52
  features:
53
  - name: translation
@@ -58,16 +58,16 @@ dataset_info:
58
  - en
59
  splits:
60
  - name: train
61
- num_bytes: 1373123263
62
  num_examples: 4548885
63
  - name: validation
64
- num_bytes: 522989
65
  num_examples: 2169
66
  - name: test
67
- num_bytes: 735516
68
  num_examples: 2999
69
- download_size: 1690726387
70
- dataset_size: 1374381768
71
  - config_name: fi-en
72
  features:
73
  - name: translation
@@ -78,16 +78,16 @@ dataset_info:
78
  - en
79
  splits:
80
  - name: train
81
- num_bytes: 605146827
82
  num_examples: 2073394
83
  - name: validation
84
- num_bytes: 306335
85
  num_examples: 1370
86
  - name: test
87
- num_bytes: 1410515
88
  num_examples: 6000
89
- download_size: 273390220
90
- dataset_size: 606863677
91
  - config_name: ro-en
92
  features:
93
  - name: translation
@@ -98,16 +98,16 @@ dataset_info:
98
  - en
99
  splits:
100
  - name: train
101
- num_bytes: 188288211
102
  num_examples: 610320
103
  - name: validation
104
- num_bytes: 561799
105
  num_examples: 1999
106
  - name: test
107
- num_bytes: 539216
108
  num_examples: 1999
109
- download_size: 287363574
110
- dataset_size: 189389226
111
  - config_name: ru-en
112
  features:
113
  - name: translation
@@ -118,16 +118,16 @@ dataset_info:
118
  - en
119
  splits:
120
  - name: train
121
- num_bytes: 448338585
122
  num_examples: 1516162
123
  - name: validation
124
- num_bytes: 955972
125
  num_examples: 2818
126
  - name: test
127
- num_bytes: 1050677
128
  num_examples: 2998
129
- download_size: 1042119564
130
- dataset_size: 450345234
131
  - config_name: tr-en
132
  features:
133
  - name: translation
@@ -138,16 +138,65 @@ dataset_info:
138
  - en
139
  splits:
140
  - name: train
141
- num_bytes: 60416617
142
  num_examples: 205756
143
  - name: validation
144
- num_bytes: 240650
145
  num_examples: 1001
146
  - name: test
147
- num_bytes: 732436
148
  num_examples: 3000
149
- download_size: 62263061
150
- dataset_size: 61389703
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
151
  ---
152
 
153
  # Dataset Card for "wmt16"
 
25
  task_categories:
26
  - translation
27
  task_ids: []
 
28
  paperswithcode_id: wmt-2016
29
+ pretty_name: WMT16
30
  dataset_info:
31
  - config_name: cs-en
32
  features:
 
38
  - en
39
  splits:
40
  - name: train
41
+ num_bytes: 295995226
42
  num_examples: 997240
43
  - name: validation
44
+ num_bytes: 572195
45
  num_examples: 2656
46
  - name: test
47
+ num_bytes: 707862
48
  num_examples: 2999
49
+ download_size: 178250444
50
+ dataset_size: 297275283
51
  - config_name: de-en
52
  features:
53
  - name: translation
 
58
  - en
59
  splits:
60
  - name: train
61
+ num_bytes: 1373099816
62
  num_examples: 4548885
63
  - name: validation
64
+ num_bytes: 522981
65
  num_examples: 2169
66
  - name: test
67
+ num_bytes: 735508
68
  num_examples: 2999
69
+ download_size: 827152589
70
+ dataset_size: 1374358305
71
  - config_name: fi-en
72
  features:
73
  - name: translation
 
78
  - en
79
  splits:
80
  - name: train
81
+ num_bytes: 605145153
82
  num_examples: 2073394
83
  - name: validation
84
+ num_bytes: 306327
85
  num_examples: 1370
86
  - name: test
87
+ num_bytes: 1410507
88
  num_examples: 6000
89
+ download_size: 348306427
90
+ dataset_size: 606861987
91
  - config_name: ro-en
92
  features:
93
  - name: translation
 
98
  - en
99
  splits:
100
  - name: train
101
+ num_bytes: 188287711
102
  num_examples: 610320
103
  - name: validation
104
+ num_bytes: 561791
105
  num_examples: 1999
106
  - name: test
107
+ num_bytes: 539208
108
  num_examples: 1999
109
+ download_size: 108584039
110
+ dataset_size: 189388710
111
  - config_name: ru-en
112
  features:
113
  - name: translation
 
118
  - en
119
  splits:
120
  - name: train
121
+ num_bytes: 448322024
122
  num_examples: 1516162
123
  - name: validation
124
+ num_bytes: 955964
125
  num_examples: 2818
126
  - name: test
127
+ num_bytes: 1050669
128
  num_examples: 2998
129
+ download_size: 231557371
130
+ dataset_size: 450328657
131
  - config_name: tr-en
132
  features:
133
  - name: translation
 
138
  - en
139
  splits:
140
  - name: train
141
+ num_bytes: 60416449
142
  num_examples: 205756
143
  - name: validation
144
+ num_bytes: 240642
145
  num_examples: 1001
146
  - name: test
147
+ num_bytes: 732428
148
  num_examples: 3000
149
+ download_size: 37389436
150
+ dataset_size: 61389519
151
+ configs:
152
+ - config_name: cs-en
153
+ data_files:
154
+ - split: train
155
+ path: cs-en/train-*
156
+ - split: validation
157
+ path: cs-en/validation-*
158
+ - split: test
159
+ path: cs-en/test-*
160
+ - config_name: de-en
161
+ data_files:
162
+ - split: train
163
+ path: de-en/train-*
164
+ - split: validation
165
+ path: de-en/validation-*
166
+ - split: test
167
+ path: de-en/test-*
168
+ - config_name: fi-en
169
+ data_files:
170
+ - split: train
171
+ path: fi-en/train-*
172
+ - split: validation
173
+ path: fi-en/validation-*
174
+ - split: test
175
+ path: fi-en/test-*
176
+ - config_name: ro-en
177
+ data_files:
178
+ - split: train
179
+ path: ro-en/train-*
180
+ - split: validation
181
+ path: ro-en/validation-*
182
+ - split: test
183
+ path: ro-en/test-*
184
+ - config_name: ru-en
185
+ data_files:
186
+ - split: train
187
+ path: ru-en/train-*
188
+ - split: validation
189
+ path: ru-en/validation-*
190
+ - split: test
191
+ path: ru-en/test-*
192
+ - config_name: tr-en
193
+ data_files:
194
+ - split: train
195
+ path: tr-en/train-*
196
+ - split: validation
197
+ path: tr-en/validation-*
198
+ - split: test
199
+ path: tr-en/test-*
200
  ---
201
 
202
  # Dataset Card for "wmt16"
cs-en/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0c19c35a0d4932c738a4a02e080cc473f1b5544374d97bdfbed7815d247cd172
3
+ size 468787
cs-en/train-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:320b8b34f7520e558122cbc04516429b92ec726544f4522abf21f4f363af3d0b
3
+ size 177391346
cs-en/validation-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e5f3e29f5b49bc5c896e8fa9653a52cf2ad2b6b95d8b45dab793f8f15d06817a
3
+ size 390311
de-en/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bb87426aded7472ee24408639dae68ab6eb27bb2f495353b406564e72e87fd26
3
+ size 475031
de-en/train-00000-of-00003.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0a037ece172b3f42c7bf06bb50ff5f718f62720d17998d4e6cc4223c541667cf
3
+ size 282014647
de-en/train-00001-of-00003.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:96cf09052f9bc879e8a70029f7a49868fa3f4ba6e5c31254a1c51ab14a72eb2e
3
+ size 267375046
de-en/train-00002-of-00003.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d9991b85a44e6e5691cd3fac98e401f6f1b9c4c8208a77bd82234e9041b2b285
3
+ size 276945174
de-en/validation-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dca6c5c1a432f7064665323555352e809b5c7e2ed620e61bbb6a27dcf2c9235e
3
+ size 342691
fi-en/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:10ac26e7da9da054054391d5dbe786b81bc9ed18b44946dff4a8a12768f52ad1
3
+ size 904438
fi-en/train-00000-of-00002.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4d4f5a98fd6a407e0a4e21049d8eecc6b5d2a6e745a53ad3993218fc1ac60713
3
+ size 187478084
fi-en/train-00001-of-00002.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d439321116c69a3e21f1436fcd09f39ef95f6232bbb0f444036d6f008de04fe9
3
+ size 159720579
fi-en/validation-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2693bb05ad4d1c2739486a29a380696c88d8495eb6b55e7e7cacf96050cb0c1a
3
+ size 203326
ro-en/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:21742f54d534024c59e0d94f92b9b8e5bf4c871027a866566275cc3ab582b218
3
+ size 342174
ro-en/train-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0a3b7151736e9fc0c7cae9c6f2e96493d929e910372f5c6c29d41f716dd956a7
3
+ size 107880264
ro-en/validation-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9dff36e7a899121682d47f0238de1aa307f3b03f087ad411d1a2846977326082
3
+ size 361601
ru-en/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a01917bda325d1a287dd09061cf9e8727f79c3f95fd29ce8ac0ec675ec7a227f
3
+ size 586531
ru-en/train-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a74a159275942c6b6ec11fef24f0f9b364b9e5784ff0fb84952cc7b1593b5943
3
+ size 230432840
ru-en/validation-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:80bd5e16b06852680fb6670d99e50bbe014a4fb8777722387a7558f775d98bf6
3
+ size 538000
tr-en/test-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5c2226d153ed115b81901eb32396a2b8b2c1b5f6f99035e4b1fb9e4a28c81f9f
3
+ size 465730
tr-en/train-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0195fe43ea5ed06b7d243112459a021f33780351543d853b0d6420ab19759def
3
+ size 36771180
tr-en/validation-00000-of-00001.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:76f7f6c9b43d09ee013ee9494009d0cd1835d773a258bc847c22f3d3ee29fa31
3
+ size 152526
wmt16.py DELETED
@@ -1,72 +0,0 @@
1
- # coding=utf-8
2
- # Copyright 2020 The TensorFlow Datasets Authors and the HuggingFace Datasets Authors.
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
- # Lint as: python3
17
- """WMT16: Translate dataset."""
18
-
19
- import datasets
20
-
21
- from .wmt_utils import Wmt, WmtConfig
22
-
23
-
24
- _URL = "http://www.statmt.org/wmt16/translation-task.html"
25
- _CITATION = """
26
- @InProceedings{bojar-EtAl:2016:WMT1,
27
- author = {Bojar, Ond\v{r}ej and Chatterjee, Rajen and Federmann, Christian and Graham, Yvette and Haddow, Barry and Huck, Matthias and Jimeno Yepes, Antonio and Koehn, Philipp and Logacheva, Varvara and Monz, Christof and Negri, Matteo and Neveol, Aurelie and Neves, Mariana and Popel, Martin and Post, Matt and Rubino, Raphael and Scarton, Carolina and Specia, Lucia and Turchi, Marco and Verspoor, Karin and Zampieri, Marcos},
28
- title = {Findings of the 2016 Conference on Machine Translation},
29
- booktitle = {Proceedings of the First Conference on Machine Translation},
30
- month = {August},
31
- year = {2016},
32
- address = {Berlin, Germany},
33
- publisher = {Association for Computational Linguistics},
34
- pages = {131--198},
35
- url = {http://www.aclweb.org/anthology/W/W16/W16-2301}
36
- }
37
- """
38
-
39
- _LANGUAGE_PAIRS = [(lang, "en") for lang in ["cs", "de", "fi", "ro", "ru", "tr"]]
40
-
41
-
42
- class Wmt16(Wmt):
43
- """WMT 16 translation datasets for all {xx, "en"} language pairs."""
44
-
45
- BUILDER_CONFIGS = [
46
- WmtConfig( # pylint:disable=g-complex-comprehension
47
- description="WMT 2016 %s-%s translation task dataset." % (l1, l2),
48
- url=_URL,
49
- citation=_CITATION,
50
- language_pair=(l1, l2),
51
- version=datasets.Version("1.0.0"),
52
- )
53
- for l1, l2 in _LANGUAGE_PAIRS
54
- ]
55
-
56
- @property
57
- def _subsets(self):
58
- return {
59
- datasets.Split.TRAIN: [
60
- "europarl_v7",
61
- "europarl_v8_16",
62
- "commoncrawl",
63
- "newscommentary_v11",
64
- "czeng_16pre",
65
- "yandexcorpus",
66
- "wikiheadlines_fi",
67
- "wikiheadlines_ru",
68
- "setimes_2",
69
- ],
70
- datasets.Split.VALIDATION: ["newsdev2016", "newstest2015"],
71
- datasets.Split.TEST: ["newstest2016", "newstestB2016"],
72
- }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
wmt_utils.py DELETED
@@ -1,1025 +0,0 @@
1
- # coding=utf-8
2
- # Copyright 2020 The TensorFlow Datasets Authors and the HuggingFace Datasets Authors.
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
- # Lint as: python3
17
- """WMT: Translate dataset."""
18
-
19
-
20
- import codecs
21
- import functools
22
- import glob
23
- import gzip
24
- import itertools
25
- import os
26
- import re
27
- import xml.etree.cElementTree as ElementTree
28
-
29
- import datasets
30
-
31
-
32
- logger = datasets.logging.get_logger(__name__)
33
-
34
-
35
- _DESCRIPTION = """\
36
- Translation dataset based on the data from statmt.org.
37
-
38
- Versions exist for different years using a combination of data
39
- sources. The base `wmt` allows you to create a custom dataset by choosing
40
- your own data/language pair. This can be done as follows:
41
-
42
- ```python
43
- from datasets import inspect_dataset, load_dataset_builder
44
-
45
- inspect_dataset("wmt16", "path/to/scripts")
46
- builder = load_dataset_builder(
47
- "path/to/scripts/wmt_utils.py",
48
- language_pair=("fr", "de"),
49
- subsets={
50
- datasets.Split.TRAIN: ["commoncrawl_frde"],
51
- datasets.Split.VALIDATION: ["euelections_dev2019"],
52
- },
53
- )
54
-
55
- # Standard version
56
- builder.download_and_prepare()
57
- ds = builder.as_dataset()
58
-
59
- # Streamable version
60
- ds = builder.as_streaming_dataset()
61
- ```
62
-
63
- """
64
-
65
-
66
- CWMT_SUBSET_NAMES = ["casia2015", "casict2011", "casict2015", "datum2015", "datum2017", "neu2017"]
67
-
68
-
69
- class SubDataset:
70
- """Class to keep track of information on a sub-dataset of WMT."""
71
-
72
- def __init__(self, name, target, sources, url, path, manual_dl_files=None):
73
- """Sub-dataset of WMT.
74
-
75
- Args:
76
- name: `string`, a unique dataset identifier.
77
- target: `string`, the target language code.
78
- sources: `set<string>`, the set of source language codes.
79
- url: `string` or `(string, string)`, URL(s) or URL template(s) specifying
80
- where to download the raw data from. If two strings are provided, the
81
- first is used for the source language and the second for the target.
82
- Template strings can either contain '{src}' placeholders that will be
83
- filled in with the source language code, '{0}' and '{1}' placeholders
84
- that will be filled in with the source and target language codes in
85
- alphabetical order, or all 3.
86
- path: `string` or `(string, string)`, path(s) or path template(s)
87
- specifing the path to the raw data relative to the root of the
88
- downloaded archive. If two strings are provided, the dataset is assumed
89
- to be made up of parallel text files, the first being the source and the
90
- second the target. If one string is provided, both languages are assumed
91
- to be stored within the same file and the extension is used to determine
92
- how to parse it. Template strings should be formatted the same as in
93
- `url`.
94
- manual_dl_files: `<list>(string)` (optional), the list of files that must
95
- be manually downloaded to the data directory.
96
- """
97
- self._paths = (path,) if isinstance(path, str) else path
98
- self._urls = (url,) if isinstance(url, str) else url
99
- self._manual_dl_files = manual_dl_files if manual_dl_files else []
100
- self.name = name
101
- self.target = target
102
- self.sources = set(sources)
103
-
104
- def _inject_language(self, src, strings):
105
- """Injects languages into (potentially) template strings."""
106
- if src not in self.sources:
107
- raise ValueError(f"Invalid source for '{self.name}': {src}")
108
-
109
- def _format_string(s):
110
- if "{0}" in s and "{1}" and "{src}" in s:
111
- return s.format(*sorted([src, self.target]), src=src)
112
- elif "{0}" in s and "{1}" in s:
113
- return s.format(*sorted([src, self.target]))
114
- elif "{src}" in s:
115
- return s.format(src=src)
116
- else:
117
- return s
118
-
119
- return [_format_string(s) for s in strings]
120
-
121
- def get_url(self, src):
122
- return self._inject_language(src, self._urls)
123
-
124
- def get_manual_dl_files(self, src):
125
- return self._inject_language(src, self._manual_dl_files)
126
-
127
- def get_path(self, src):
128
- return self._inject_language(src, self._paths)
129
-
130
-
131
- # Subsets used in the training sets for various years of WMT.
132
- _TRAIN_SUBSETS = [
133
- # pylint:disable=line-too-long
134
- SubDataset(
135
- name="commoncrawl",
136
- target="en", # fr-de pair in commoncrawl_frde
137
- sources={"cs", "de", "es", "fr", "ru"},
138
- url="https://huggingface.co/datasets/wmt/wmt13/resolve/main-zip/training-parallel-commoncrawl.zip",
139
- path=("commoncrawl.{src}-en.{src}", "commoncrawl.{src}-en.en"),
140
- ),
141
- SubDataset(
142
- name="commoncrawl_frde",
143
- target="de",
144
- sources={"fr"},
145
- url=(
146
- "https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/fr-de/bitexts/commoncrawl.fr.gz",
147
- "https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/fr-de/bitexts/commoncrawl.de.gz",
148
- ),
149
- path=("", ""),
150
- ),
151
- SubDataset(
152
- name="czeng_10",
153
- target="en",
154
- sources={"cs"},
155
- url="http://ufal.mff.cuni.cz/czeng/czeng10",
156
- manual_dl_files=["data-plaintext-format.%d.tar" % i for i in range(10)],
157
- # Each tar contains multiple files, which we process specially in
158
- # _parse_czeng.
159
- path=("data.plaintext-format/??train.gz",) * 10,
160
- ),
161
- SubDataset(
162
- name="czeng_16pre",
163
- target="en",
164
- sources={"cs"},
165
- url="http://ufal.mff.cuni.cz/czeng/czeng16pre",
166
- manual_dl_files=["czeng16pre.deduped-ignoring-sections.txt.gz"],
167
- path="",
168
- ),
169
- SubDataset(
170
- name="czeng_16",
171
- target="en",
172
- sources={"cs"},
173
- url="http://ufal.mff.cuni.cz/czeng",
174
- manual_dl_files=["data-plaintext-format.%d.tar" % i for i in range(10)],
175
- # Each tar contains multiple files, which we process specially in
176
- # _parse_czeng.
177
- path=("data.plaintext-format/??train.gz",) * 10,
178
- ),
179
- SubDataset(
180
- # This dataset differs from the above in the filtering that is applied
181
- # during parsing.
182
- name="czeng_17",
183
- target="en",
184
- sources={"cs"},
185
- url="http://ufal.mff.cuni.cz/czeng",
186
- manual_dl_files=["data-plaintext-format.%d.tar" % i for i in range(10)],
187
- # Each tar contains multiple files, which we process specially in
188
- # _parse_czeng.
189
- path=("data.plaintext-format/??train.gz",) * 10,
190
- ),
191
- SubDataset(
192
- name="dcep_v1",
193
- target="en",
194
- sources={"lv"},
195
- url="https://huggingface.co/datasets/wmt/wmt17/resolve/main-zip/translation-task/dcep.lv-en.v1.zip",
196
- path=("dcep.en-lv/dcep.lv", "dcep.en-lv/dcep.en"),
197
- ),
198
- SubDataset(
199
- name="europarl_v7",
200
- target="en",
201
- sources={"cs", "de", "es", "fr"},
202
- url="https://huggingface.co/datasets/wmt/wmt13/resolve/main-zip/training-parallel-europarl-v7.zip",
203
- path=("training/europarl-v7.{src}-en.{src}", "training/europarl-v7.{src}-en.en"),
204
- ),
205
- SubDataset(
206
- name="europarl_v7_frde",
207
- target="de",
208
- sources={"fr"},
209
- url=(
210
- "https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/fr-de/bitexts/europarl-v7.fr.gz",
211
- "https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/fr-de/bitexts/europarl-v7.de.gz",
212
- ),
213
- path=("", ""),
214
- ),
215
- SubDataset(
216
- name="europarl_v8_18",
217
- target="en",
218
- sources={"et", "fi"},
219
- url="https://huggingface.co/datasets/wmt/wmt18/resolve/main-zip/translation-task/training-parallel-ep-v8.zip",
220
- path=("training/europarl-v8.{src}-en.{src}", "training/europarl-v8.{src}-en.en"),
221
- ),
222
- SubDataset(
223
- name="europarl_v8_16",
224
- target="en",
225
- sources={"fi", "ro"},
226
- url="https://huggingface.co/datasets/wmt/wmt16/resolve/main-zip/translation-task/training-parallel-ep-v8.zip",
227
- path=("training-parallel-ep-v8/europarl-v8.{src}-en.{src}", "training-parallel-ep-v8/europarl-v8.{src}-en.en"),
228
- ),
229
- SubDataset(
230
- name="europarl_v9",
231
- target="en",
232
- sources={"cs", "de", "fi", "lt"},
233
- url="https://huggingface.co/datasets/wmt/europarl/resolve/main/v9/training/europarl-v9.{src}-en.tsv.gz",
234
- path="",
235
- ),
236
- SubDataset(
237
- name="gigafren",
238
- target="en",
239
- sources={"fr"},
240
- url="https://huggingface.co/datasets/wmt/wmt10/resolve/main-zip/training-giga-fren.zip",
241
- path=("giga-fren.release2.fixed.fr.gz", "giga-fren.release2.fixed.en.gz"),
242
- ),
243
- SubDataset(
244
- name="hindencorp_01",
245
- target="en",
246
- sources={"hi"},
247
- url="http://ufallab.ms.mff.cuni.cz/~bojar/hindencorp",
248
- manual_dl_files=["hindencorp0.1.gz"],
249
- path="",
250
- ),
251
- SubDataset(
252
- name="leta_v1",
253
- target="en",
254
- sources={"lv"},
255
- url="https://huggingface.co/datasets/wmt/wmt17/resolve/main-zip/translation-task/leta.v1.zip",
256
- path=("LETA-lv-en/leta.lv", "LETA-lv-en/leta.en"),
257
- ),
258
- SubDataset(
259
- name="multiun",
260
- target="en",
261
- sources={"es", "fr"},
262
- url="https://huggingface.co/datasets/wmt/wmt13/resolve/main-zip/training-parallel-un.zip",
263
- path=("un/undoc.2000.{src}-en.{src}", "un/undoc.2000.{src}-en.en"),
264
- ),
265
- SubDataset(
266
- name="newscommentary_v9",
267
- target="en",
268
- sources={"cs", "de", "fr", "ru"},
269
- url="https://huggingface.co/datasets/wmt/wmt14/resolve/main-zip/training-parallel-nc-v9.zip",
270
- path=("training/news-commentary-v9.{src}-en.{src}", "training/news-commentary-v9.{src}-en.en"),
271
- ),
272
- SubDataset(
273
- name="newscommentary_v10",
274
- target="en",
275
- sources={"cs", "de", "fr", "ru"},
276
- url="https://huggingface.co/datasets/wmt/wmt15/resolve/main-zip/training-parallel-nc-v10.zip",
277
- path=("news-commentary-v10.{src}-en.{src}", "news-commentary-v10.{src}-en.en"),
278
- ),
279
- SubDataset(
280
- name="newscommentary_v11",
281
- target="en",
282
- sources={"cs", "de", "ru"},
283
- url="https://huggingface.co/datasets/wmt/wmt16/resolve/main-zip/translation-task/training-parallel-nc-v11.zip",
284
- path=(
285
- "training-parallel-nc-v11/news-commentary-v11.{src}-en.{src}",
286
- "training-parallel-nc-v11/news-commentary-v11.{src}-en.en",
287
- ),
288
- ),
289
- SubDataset(
290
- name="newscommentary_v12",
291
- target="en",
292
- sources={"cs", "de", "ru", "zh"},
293
- url="https://huggingface.co/datasets/wmt/wmt17/resolve/main-zip/translation-task/training-parallel-nc-v12.zip",
294
- path=("training/news-commentary-v12.{src}-en.{src}", "training/news-commentary-v12.{src}-en.en"),
295
- ),
296
- SubDataset(
297
- name="newscommentary_v13",
298
- target="en",
299
- sources={"cs", "de", "ru", "zh"},
300
- url="https://huggingface.co/datasets/wmt/wmt18/resolve/main-zip/translation-task/training-parallel-nc-v13.zip",
301
- path=(
302
- "training-parallel-nc-v13/news-commentary-v13.{src}-en.{src}",
303
- "training-parallel-nc-v13/news-commentary-v13.{src}-en.en",
304
- ),
305
- ),
306
- SubDataset(
307
- name="newscommentary_v14",
308
- target="en", # fr-de pair in newscommentary_v14_frde
309
- sources={"cs", "de", "kk", "ru", "zh"},
310
- url="http://data.statmt.org/news-commentary/v14/training/news-commentary-v14.{0}-{1}.tsv.gz",
311
- path="",
312
- ),
313
- SubDataset(
314
- name="newscommentary_v14_frde",
315
- target="de",
316
- sources={"fr"},
317
- url="http://data.statmt.org/news-commentary/v14/training/news-commentary-v14.de-fr.tsv.gz",
318
- path="",
319
- ),
320
- SubDataset(
321
- name="onlinebooks_v1",
322
- target="en",
323
- sources={"lv"},
324
- url="https://huggingface.co/datasets/wmt/wmt17/resolve/main-zip/translation-task/books.lv-en.v1.zip",
325
- path=("farewell/farewell.lv", "farewell/farewell.en"),
326
- ),
327
- SubDataset(
328
- name="paracrawl_v1",
329
- target="en",
330
- sources={"cs", "de", "et", "fi", "ru"},
331
- url="https://s3.amazonaws.com/web-language-models/paracrawl/release1/paracrawl-release1.en-{src}.zipporah0-dedup-clean.tgz", # TODO(QL): use zip for streaming
332
- path=(
333
- "paracrawl-release1.en-{src}.zipporah0-dedup-clean.{src}",
334
- "paracrawl-release1.en-{src}.zipporah0-dedup-clean.en",
335
- ),
336
- ),
337
- SubDataset(
338
- name="paracrawl_v1_ru",
339
- target="en",
340
- sources={"ru"},
341
- url="https://s3.amazonaws.com/web-language-models/paracrawl/release1/paracrawl-release1.en-ru.zipporah0-dedup-clean.tgz", # TODO(QL): use zip for streaming
342
- path=(
343
- "paracrawl-release1.en-ru.zipporah0-dedup-clean.ru",
344
- "paracrawl-release1.en-ru.zipporah0-dedup-clean.en",
345
- ),
346
- ),
347
- SubDataset(
348
- name="paracrawl_v3",
349
- target="en", # fr-de pair in paracrawl_v3_frde
350
- sources={"cs", "de", "fi", "lt"},
351
- url="https://s3.amazonaws.com/web-language-models/paracrawl/release3/en-{src}.bicleaner07.tmx.gz",
352
- path="",
353
- ),
354
- SubDataset(
355
- name="paracrawl_v3_frde",
356
- target="de",
357
- sources={"fr"},
358
- url=(
359
- "https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/fr-de/bitexts/de-fr.bicleaner07.de.gz",
360
- "https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/fr-de/bitexts/de-fr.bicleaner07.fr.gz",
361
- ),
362
- path=("", ""),
363
- ),
364
- SubDataset(
365
- name="rapid_2016",
366
- target="en",
367
- sources={"de", "et", "fi"},
368
- url="https://huggingface.co/datasets/wmt/wmt18/resolve/main-zip/translation-task/rapid2016.zip",
369
- path=("rapid2016.{0}-{1}.{src}", "rapid2016.{0}-{1}.en"),
370
- ),
371
- SubDataset(
372
- name="rapid_2016_ltfi",
373
- target="en",
374
- sources={"fi", "lt"},
375
- url="https://tilde-model.s3-eu-west-1.amazonaws.com/rapid2016.en-{src}.tmx.zip",
376
- path="rapid2016.en-{src}.tmx",
377
- ),
378
- SubDataset(
379
- name="rapid_2019",
380
- target="en",
381
- sources={"de"},
382
- url="https://s3-eu-west-1.amazonaws.com/tilde-model/rapid2019.de-en.zip",
383
- path=("rapid2019.de-en.de", "rapid2019.de-en.en"),
384
- ),
385
- SubDataset(
386
- name="setimes_2",
387
- target="en",
388
- sources={"ro", "tr"},
389
- url="https://object.pouta.csc.fi/OPUS-SETIMES/v2/tmx/en-{src}.tmx.gz",
390
- path="",
391
- ),
392
- SubDataset(
393
- name="uncorpus_v1",
394
- target="en",
395
- sources={"ru", "zh"},
396
- url="https://huggingface.co/datasets/wmt/uncorpus/resolve/main-zip/UNv1.0.en-{src}.zip",
397
- path=("en-{src}/UNv1.0.en-{src}.{src}", "en-{src}/UNv1.0.en-{src}.en"),
398
- ),
399
- SubDataset(
400
- name="wikiheadlines_fi",
401
- target="en",
402
- sources={"fi"},
403
- url="https://huggingface.co/datasets/wmt/wmt15/resolve/main-zip/wiki-titles.zip",
404
- path="wiki/fi-en/titles.fi-en",
405
- ),
406
- SubDataset(
407
- name="wikiheadlines_hi",
408
- target="en",
409
- sources={"hi"},
410
- url="https://huggingface.co/datasets/wmt/wmt14/resolve/main-zip/wiki-titles.zip",
411
- path="wiki/hi-en/wiki-titles.hi-en",
412
- ),
413
- SubDataset(
414
- # Verified that wmt14 and wmt15 files are identical.
415
- name="wikiheadlines_ru",
416
- target="en",
417
- sources={"ru"},
418
- url="https://huggingface.co/datasets/wmt/wmt15/resolve/main-zip/wiki-titles.zip",
419
- path="wiki/ru-en/wiki.ru-en",
420
- ),
421
- SubDataset(
422
- name="wikititles_v1",
423
- target="en",
424
- sources={"cs", "de", "fi", "gu", "kk", "lt", "ru", "zh"},
425
- url="https://huggingface.co/datasets/wmt/wikititles/resolve/main/v1/wikititles-v1.{src}-en.tsv.gz",
426
- path="",
427
- ),
428
- SubDataset(
429
- name="yandexcorpus",
430
- target="en",
431
- sources={"ru"},
432
- url="https://translate.yandex.ru/corpus?lang=en",
433
- manual_dl_files=["1mcorpus.zip"],
434
- path=("corpus.en_ru.1m.ru", "corpus.en_ru.1m.en"),
435
- ),
436
- # pylint:enable=line-too-long
437
- ] + [
438
- SubDataset( # pylint:disable=g-complex-comprehension
439
- name=ss,
440
- target="en",
441
- sources={"zh"},
442
- url="https://huggingface.co/datasets/wmt/wmt18/resolve/main/cwmt-wmt/%s.zip" % ss,
443
- path=("%s/*_c[hn].txt" % ss, "%s/*_en.txt" % ss),
444
- )
445
- for ss in CWMT_SUBSET_NAMES
446
- ]
447
-
448
- _DEV_SUBSETS = [
449
- SubDataset(
450
- name="euelections_dev2019",
451
- target="de",
452
- sources={"fr"},
453
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
454
- path=("dev/euelections_dev2019.fr-de.src.fr", "dev/euelections_dev2019.fr-de.tgt.de"),
455
- ),
456
- SubDataset(
457
- name="newsdev2014",
458
- target="en",
459
- sources={"hi"},
460
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
461
- path=("dev/newsdev2014.hi", "dev/newsdev2014.en"),
462
- ),
463
- SubDataset(
464
- name="newsdev2015",
465
- target="en",
466
- sources={"fi"},
467
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
468
- path=("dev/newsdev2015-fien-src.{src}.sgm", "dev/newsdev2015-fien-ref.en.sgm"),
469
- ),
470
- SubDataset(
471
- name="newsdiscussdev2015",
472
- target="en",
473
- sources={"ro", "tr"},
474
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
475
- path=("dev/newsdiscussdev2015-{src}en-src.{src}.sgm", "dev/newsdiscussdev2015-{src}en-ref.en.sgm"),
476
- ),
477
- SubDataset(
478
- name="newsdev2016",
479
- target="en",
480
- sources={"ro", "tr"},
481
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
482
- path=("dev/newsdev2016-{src}en-src.{src}.sgm", "dev/newsdev2016-{src}en-ref.en.sgm"),
483
- ),
484
- SubDataset(
485
- name="newsdev2017",
486
- target="en",
487
- sources={"lv", "zh"},
488
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
489
- path=("dev/newsdev2017-{src}en-src.{src}.sgm", "dev/newsdev2017-{src}en-ref.en.sgm"),
490
- ),
491
- SubDataset(
492
- name="newsdev2018",
493
- target="en",
494
- sources={"et"},
495
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
496
- path=("dev/newsdev2018-{src}en-src.{src}.sgm", "dev/newsdev2018-{src}en-ref.en.sgm"),
497
- ),
498
- SubDataset(
499
- name="newsdev2019",
500
- target="en",
501
- sources={"gu", "kk", "lt"},
502
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
503
- path=("dev/newsdev2019-{src}en-src.{src}.sgm", "dev/newsdev2019-{src}en-ref.en.sgm"),
504
- ),
505
- SubDataset(
506
- name="newsdiscussdev2015",
507
- target="en",
508
- sources={"fr"},
509
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
510
- path=("dev/newsdiscussdev2015-{src}en-src.{src}.sgm", "dev/newsdiscussdev2015-{src}en-ref.en.sgm"),
511
- ),
512
- SubDataset(
513
- name="newsdiscusstest2015",
514
- target="en",
515
- sources={"fr"},
516
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
517
- path=("dev/newsdiscusstest2015-{src}en-src.{src}.sgm", "dev/newsdiscusstest2015-{src}en-ref.en.sgm"),
518
- ),
519
- SubDataset(
520
- name="newssyscomb2009",
521
- target="en",
522
- sources={"cs", "de", "es", "fr"},
523
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
524
- path=("dev/newssyscomb2009.{src}", "dev/newssyscomb2009.en"),
525
- ),
526
- SubDataset(
527
- name="newstest2008",
528
- target="en",
529
- sources={"cs", "de", "es", "fr", "hu"},
530
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
531
- path=("dev/news-test2008.{src}", "dev/news-test2008.en"),
532
- ),
533
- SubDataset(
534
- name="newstest2009",
535
- target="en",
536
- sources={"cs", "de", "es", "fr"},
537
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
538
- path=("dev/newstest2009.{src}", "dev/newstest2009.en"),
539
- ),
540
- SubDataset(
541
- name="newstest2010",
542
- target="en",
543
- sources={"cs", "de", "es", "fr"},
544
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
545
- path=("dev/newstest2010.{src}", "dev/newstest2010.en"),
546
- ),
547
- SubDataset(
548
- name="newstest2011",
549
- target="en",
550
- sources={"cs", "de", "es", "fr"},
551
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
552
- path=("dev/newstest2011.{src}", "dev/newstest2011.en"),
553
- ),
554
- SubDataset(
555
- name="newstest2012",
556
- target="en",
557
- sources={"cs", "de", "es", "fr", "ru"},
558
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
559
- path=("dev/newstest2012.{src}", "dev/newstest2012.en"),
560
- ),
561
- SubDataset(
562
- name="newstest2013",
563
- target="en",
564
- sources={"cs", "de", "es", "fr", "ru"},
565
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
566
- path=("dev/newstest2013.{src}", "dev/newstest2013.en"),
567
- ),
568
- SubDataset(
569
- name="newstest2014",
570
- target="en",
571
- sources={"cs", "de", "es", "fr", "hi", "ru"},
572
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
573
- path=("dev/newstest2014-{src}en-src.{src}.sgm", "dev/newstest2014-{src}en-ref.en.sgm"),
574
- ),
575
- SubDataset(
576
- name="newstest2015",
577
- target="en",
578
- sources={"cs", "de", "fi", "ru"},
579
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
580
- path=("dev/newstest2015-{src}en-src.{src}.sgm", "dev/newstest2015-{src}en-ref.en.sgm"),
581
- ),
582
- SubDataset(
583
- name="newsdiscusstest2015",
584
- target="en",
585
- sources={"fr"},
586
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
587
- path=("dev/newsdiscusstest2015-{src}en-src.{src}.sgm", "dev/newsdiscusstest2015-{src}en-ref.en.sgm"),
588
- ),
589
- SubDataset(
590
- name="newstest2016",
591
- target="en",
592
- sources={"cs", "de", "fi", "ro", "ru", "tr"},
593
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
594
- path=("dev/newstest2016-{src}en-src.{src}.sgm", "dev/newstest2016-{src}en-ref.en.sgm"),
595
- ),
596
- SubDataset(
597
- name="newstestB2016",
598
- target="en",
599
- sources={"fi"},
600
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
601
- path=("dev/newstestB2016-enfi-ref.{src}.sgm", "dev/newstestB2016-enfi-src.en.sgm"),
602
- ),
603
- SubDataset(
604
- name="newstest2017",
605
- target="en",
606
- sources={"cs", "de", "fi", "lv", "ru", "tr", "zh"},
607
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
608
- path=("dev/newstest2017-{src}en-src.{src}.sgm", "dev/newstest2017-{src}en-ref.en.sgm"),
609
- ),
610
- SubDataset(
611
- name="newstestB2017",
612
- target="en",
613
- sources={"fi"},
614
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
615
- path=("dev/newstestB2017-fien-src.fi.sgm", "dev/newstestB2017-fien-ref.en.sgm"),
616
- ),
617
- SubDataset(
618
- name="newstest2018",
619
- target="en",
620
- sources={"cs", "de", "et", "fi", "ru", "tr", "zh"},
621
- url="https://huggingface.co/datasets/wmt/wmt19/resolve/main-zip/translation-task/dev.zip",
622
- path=("dev/newstest2018-{src}en-src.{src}.sgm", "dev/newstest2018-{src}en-ref.en.sgm"),
623
- ),
624
- ]
625
-
626
- DATASET_MAP = {dataset.name: dataset for dataset in _TRAIN_SUBSETS + _DEV_SUBSETS}
627
-
628
- _CZENG17_FILTER = SubDataset(
629
- name="czeng17_filter",
630
- target="en",
631
- sources={"cs"},
632
- url="http://ufal.mff.cuni.cz/czeng/download.php?f=convert_czeng16_to_17.pl.zip",
633
- path="convert_czeng16_to_17.pl",
634
- )
635
-
636
-
637
- class WmtConfig(datasets.BuilderConfig):
638
- """BuilderConfig for WMT."""
639
-
640
- def __init__(self, url=None, citation=None, description=None, language_pair=(None, None), subsets=None, **kwargs):
641
- """BuilderConfig for WMT.
642
-
643
- Args:
644
- url: The reference URL for the dataset.
645
- citation: The paper citation for the dataset.
646
- description: The description of the dataset.
647
- language_pair: pair of languages that will be used for translation. Should
648
- contain 2 letter coded strings. For example: ("en", "de").
649
- configuration for the `datasets.features.text.TextEncoder` used for the
650
- `datasets.features.text.Translation` features.
651
- subsets: Dict[split, list[str]]. List of the subset to use for each of the
652
- split. Note that WMT subclasses overwrite this parameter.
653
- **kwargs: keyword arguments forwarded to super.
654
- """
655
- name = "%s-%s" % (language_pair[0], language_pair[1])
656
- if "name" in kwargs: # Add name suffix for custom configs
657
- name += "." + kwargs.pop("name")
658
-
659
- super(WmtConfig, self).__init__(name=name, description=description, **kwargs)
660
-
661
- self.url = url or "http://www.statmt.org"
662
- self.citation = citation
663
- self.language_pair = language_pair
664
- self.subsets = subsets
665
-
666
- # TODO(PVP): remove when manual dir works
667
- # +++++++++++++++++++++
668
- if language_pair[1] in ["cs", "hi", "ru"]:
669
- assert NotImplementedError(f"The dataset for {language_pair[1]}-en is currently not fully supported.")
670
- # +++++++++++++++++++++
671
-
672
-
673
- class Wmt(datasets.GeneratorBasedBuilder):
674
- """WMT translation dataset."""
675
-
676
- BUILDER_CONFIG_CLASS = WmtConfig
677
-
678
- def __init__(self, *args, **kwargs):
679
- super(Wmt, self).__init__(*args, **kwargs)
680
-
681
- @property
682
- def _subsets(self):
683
- """Subsets that make up each split of the dataset."""
684
- raise NotImplementedError("This is a abstract method")
685
-
686
- @property
687
- def subsets(self):
688
- """Subsets that make up each split of the dataset for the language pair."""
689
- source, target = self.config.language_pair
690
- filtered_subsets = {}
691
- subsets = self._subsets if self.config.subsets is None else self.config.subsets
692
- for split, ss_names in subsets.items():
693
- filtered_subsets[split] = []
694
- for ss_name in ss_names:
695
- dataset = DATASET_MAP[ss_name]
696
- if dataset.target != target or source not in dataset.sources:
697
- logger.info("Skipping sub-dataset that does not include language pair: %s", ss_name)
698
- else:
699
- filtered_subsets[split].append(ss_name)
700
- logger.info("Using sub-datasets: %s", filtered_subsets)
701
- return filtered_subsets
702
-
703
- def _info(self):
704
- src, target = self.config.language_pair
705
- return datasets.DatasetInfo(
706
- description=_DESCRIPTION,
707
- features=datasets.Features(
708
- {"translation": datasets.features.Translation(languages=self.config.language_pair)}
709
- ),
710
- supervised_keys=(src, target),
711
- homepage=self.config.url,
712
- citation=self.config.citation,
713
- )
714
-
715
- def _vocab_text_gen(self, split_subsets, extraction_map, language):
716
- for _, ex in self._generate_examples(split_subsets, extraction_map, with_translation=False):
717
- yield ex[language]
718
-
719
- def _split_generators(self, dl_manager):
720
- source, _ = self.config.language_pair
721
- manual_paths_dict = {}
722
- urls_to_download = {}
723
- for ss_name in itertools.chain.from_iterable(self.subsets.values()):
724
- if ss_name == "czeng_17":
725
- # CzEng1.7 is CzEng1.6 with some blocks filtered out. We must download
726
- # the filtering script so we can parse out which blocks need to be
727
- # removed.
728
- urls_to_download[_CZENG17_FILTER.name] = _CZENG17_FILTER.get_url(source)
729
-
730
- # get dataset
731
- dataset = DATASET_MAP[ss_name]
732
- if dataset.get_manual_dl_files(source):
733
- # TODO(PVP): following two lines skip configs that are incomplete for now
734
- # +++++++++++++++++++++
735
- logger.info("Skipping {dataset.name} for now. Incomplete dataset for {self.config.name}")
736
- continue
737
- # +++++++++++++++++++++
738
-
739
- manual_dl_files = dataset.get_manual_dl_files(source)
740
- manual_paths = [
741
- os.path.join(os.path.abspath(os.path.expanduser(dl_manager.manual_dir)), fname)
742
- for fname in manual_dl_files
743
- ]
744
- assert all(
745
- os.path.exists(path) for path in manual_paths
746
- ), f"For {dataset.name}, you must manually download the following file(s) from {dataset.get_url(source)} and place them in {dl_manager.manual_dir}: {', '.join(manual_dl_files)}"
747
-
748
- # set manual path for correct subset
749
- manual_paths_dict[ss_name] = manual_paths
750
- else:
751
- urls_to_download[ss_name] = dataset.get_url(source)
752
-
753
- # Download and extract files from URLs.
754
- downloaded_files = dl_manager.download_and_extract(urls_to_download)
755
- # Extract manually downloaded files.
756
- manual_files = dl_manager.extract(manual_paths_dict)
757
- extraction_map = dict(downloaded_files, **manual_files)
758
-
759
- for language in self.config.language_pair:
760
- self._vocab_text_gen(self.subsets[datasets.Split.TRAIN], extraction_map, language)
761
-
762
- return [
763
- datasets.SplitGenerator( # pylint:disable=g-complex-comprehension
764
- name=split, gen_kwargs={"split_subsets": split_subsets, "extraction_map": extraction_map}
765
- )
766
- for split, split_subsets in self.subsets.items()
767
- ]
768
-
769
- def _generate_examples(self, split_subsets, extraction_map, with_translation=True):
770
- """Returns the examples in the raw (text) form."""
771
- source, _ = self.config.language_pair
772
-
773
- def _get_local_paths(dataset, extract_dirs):
774
- rel_paths = dataset.get_path(source)
775
- if len(extract_dirs) == 1:
776
- extract_dirs = extract_dirs * len(rel_paths)
777
- return [
778
- os.path.join(ex_dir, rel_path) if rel_path else ex_dir
779
- for ex_dir, rel_path in zip(extract_dirs, rel_paths)
780
- ]
781
-
782
- def _get_filenames(dataset):
783
- rel_paths = dataset.get_path(source)
784
- urls = dataset.get_url(source)
785
- if len(urls) == 1:
786
- urls = urls * len(rel_paths)
787
- return [rel_path if rel_path else os.path.basename(url) for url, rel_path in zip(urls, rel_paths)]
788
-
789
- for ss_name in split_subsets:
790
- # TODO(PVP) remove following five lines when manual data works
791
- # +++++++++++++++++++++
792
- dataset = DATASET_MAP[ss_name]
793
- source, _ = self.config.language_pair
794
- if dataset.get_manual_dl_files(source):
795
- logger.info(f"Skipping {dataset.name} for now. Incomplete dataset for {self.config.name}")
796
- continue
797
- # +++++++++++++++++++++
798
-
799
- logger.info("Generating examples from: %s", ss_name)
800
- dataset = DATASET_MAP[ss_name]
801
- extract_dirs = extraction_map[ss_name]
802
- files = _get_local_paths(dataset, extract_dirs)
803
- filenames = _get_filenames(dataset)
804
-
805
- sub_generator_args = tuple(files)
806
-
807
- if ss_name.startswith("czeng"):
808
- if ss_name.endswith("16pre"):
809
- sub_generator = functools.partial(_parse_tsv, language_pair=("en", "cs"))
810
- sub_generator_args += tuple(filenames)
811
- elif ss_name.endswith("17"):
812
- filter_path = _get_local_paths(_CZENG17_FILTER, extraction_map[_CZENG17_FILTER.name])[0]
813
- sub_generator = functools.partial(_parse_czeng, filter_path=filter_path)
814
- else:
815
- sub_generator = _parse_czeng
816
- elif ss_name == "hindencorp_01":
817
- sub_generator = _parse_hindencorp
818
- elif len(files) == 2:
819
- if ss_name.endswith("_frde"):
820
- sub_generator = _parse_frde_bitext
821
- else:
822
- sub_generator = _parse_parallel_sentences
823
- sub_generator_args += tuple(filenames)
824
- elif len(files) == 1:
825
- fname = filenames[0]
826
- # Note: Due to formatting used by `download_manager`, the file
827
- # extension may not be at the end of the file path.
828
- if ".tsv" in fname:
829
- sub_generator = _parse_tsv
830
- sub_generator_args += tuple(filenames)
831
- elif (
832
- ss_name.startswith("newscommentary_v14")
833
- or ss_name.startswith("europarl_v9")
834
- or ss_name.startswith("wikititles_v1")
835
- ):
836
- sub_generator = functools.partial(_parse_tsv, language_pair=self.config.language_pair)
837
- sub_generator_args += tuple(filenames)
838
- elif "tmx" in fname or ss_name.startswith("paracrawl_v3"):
839
- sub_generator = _parse_tmx
840
- elif ss_name.startswith("wikiheadlines"):
841
- sub_generator = _parse_wikiheadlines
842
- else:
843
- raise ValueError("Unsupported file format: %s" % fname)
844
- else:
845
- raise ValueError("Invalid number of files: %d" % len(files))
846
-
847
- for sub_key, ex in sub_generator(*sub_generator_args):
848
- if not all(ex.values()):
849
- continue
850
- # TODO(adarob): Add subset feature.
851
- # ex["subset"] = subset
852
- key = f"{ss_name}/{sub_key}"
853
- if with_translation is True:
854
- ex = {"translation": ex}
855
- yield key, ex
856
-
857
-
858
- def _parse_parallel_sentences(f1, f2, filename1, filename2):
859
- """Returns examples from parallel SGML or text files, which may be gzipped."""
860
-
861
- def _parse_text(path, original_filename):
862
- """Returns the sentences from a single text file, which may be gzipped."""
863
- split_path = original_filename.split(".")
864
-
865
- if split_path[-1] == "gz":
866
- lang = split_path[-2]
867
-
868
- def gen():
869
- with open(path, "rb") as f, gzip.GzipFile(fileobj=f) as g:
870
- for line in g:
871
- yield line.decode("utf-8").rstrip()
872
-
873
- return gen(), lang
874
-
875
- if split_path[-1] == "txt":
876
- # CWMT
877
- lang = split_path[-2].split("_")[-1]
878
- lang = "zh" if lang in ("ch", "cn", "c[hn]") else lang
879
- else:
880
- lang = split_path[-1]
881
-
882
- def gen():
883
- with open(path, "rb") as f:
884
- for line in f:
885
- yield line.decode("utf-8").rstrip()
886
-
887
- return gen(), lang
888
-
889
- def _parse_sgm(path, original_filename):
890
- """Returns sentences from a single SGML file."""
891
- lang = original_filename.split(".")[-2]
892
- # Note: We can't use the XML parser since some of the files are badly
893
- # formatted.
894
- seg_re = re.compile(r"<seg id=\"\d+\">(.*)</seg>")
895
-
896
- def gen():
897
- with open(path, encoding="utf-8") as f:
898
- for line in f:
899
- seg_match = re.match(seg_re, line)
900
- if seg_match:
901
- assert len(seg_match.groups()) == 1
902
- yield seg_match.groups()[0]
903
-
904
- return gen(), lang
905
-
906
- parse_file = _parse_sgm if os.path.basename(f1).endswith(".sgm") else _parse_text
907
-
908
- # Some datasets (e.g., CWMT) contain multiple parallel files specified with
909
- # a wildcard. We sort both sets to align them and parse them one by one.
910
- f1_files = sorted(glob.glob(f1))
911
- f2_files = sorted(glob.glob(f2))
912
-
913
- assert f1_files and f2_files, "No matching files found: %s, %s." % (f1, f2)
914
- assert len(f1_files) == len(f2_files), "Number of files do not match: %d vs %d for %s vs %s." % (
915
- len(f1_files),
916
- len(f2_files),
917
- f1,
918
- f2,
919
- )
920
-
921
- for f_id, (f1_i, f2_i) in enumerate(zip(sorted(f1_files), sorted(f2_files))):
922
- l1_sentences, l1 = parse_file(f1_i, filename1)
923
- l2_sentences, l2 = parse_file(f2_i, filename2)
924
-
925
- for line_id, (s1, s2) in enumerate(zip(l1_sentences, l2_sentences)):
926
- key = f"{f_id}/{line_id}"
927
- yield key, {l1: s1, l2: s2}
928
-
929
-
930
- def _parse_frde_bitext(fr_path, de_path):
931
- with open(fr_path, encoding="utf-8") as fr_f:
932
- with open(de_path, encoding="utf-8") as de_f:
933
- for line_id, (s1, s2) in enumerate(zip(fr_f, de_f)):
934
- yield line_id, {"fr": s1.rstrip(), "de": s2.rstrip()}
935
-
936
-
937
- def _parse_tmx(path):
938
- """Generates examples from TMX file."""
939
-
940
- def _get_tuv_lang(tuv):
941
- for k, v in tuv.items():
942
- if k.endswith("}lang"):
943
- return v
944
- raise AssertionError("Language not found in `tuv` attributes.")
945
-
946
- def _get_tuv_seg(tuv):
947
- segs = tuv.findall("seg")
948
- assert len(segs) == 1, "Invalid number of segments: %d" % len(segs)
949
- return segs[0].text
950
-
951
- with open(path, "rb") as f:
952
- # Workaround due to: https://github.com/tensorflow/tensorflow/issues/33563
953
- utf_f = codecs.getreader("utf-8")(f)
954
- for line_id, (_, elem) in enumerate(ElementTree.iterparse(utf_f)):
955
- if elem.tag == "tu":
956
- yield line_id, {_get_tuv_lang(tuv): _get_tuv_seg(tuv) for tuv in elem.iterfind("tuv")}
957
- elem.clear()
958
-
959
-
960
- def _parse_tsv(path, filename, language_pair=None):
961
- """Generates examples from TSV file."""
962
- if language_pair is None:
963
- lang_match = re.match(r".*\.([a-z][a-z])-([a-z][a-z])\.tsv", filename)
964
- assert lang_match is not None, "Invalid TSV filename: %s" % filename
965
- l1, l2 = lang_match.groups()
966
- else:
967
- l1, l2 = language_pair
968
- with open(path, encoding="utf-8") as f:
969
- for j, line in enumerate(f):
970
- cols = line.split("\t")
971
- if len(cols) != 2:
972
- logger.warning("Skipping line %d in TSV (%s) with %d != 2 columns.", j, path, len(cols))
973
- continue
974
- s1, s2 = cols
975
- yield j, {l1: s1.strip(), l2: s2.strip()}
976
-
977
-
978
- def _parse_wikiheadlines(path):
979
- """Generates examples from Wikiheadlines dataset file."""
980
- lang_match = re.match(r".*\.([a-z][a-z])-([a-z][a-z])$", path)
981
- assert lang_match is not None, "Invalid Wikiheadlines filename: %s" % path
982
- l1, l2 = lang_match.groups()
983
- with open(path, encoding="utf-8") as f:
984
- for line_id, line in enumerate(f):
985
- s1, s2 = line.split("|||")
986
- yield line_id, {l1: s1.strip(), l2: s2.strip()}
987
-
988
-
989
- def _parse_czeng(*paths, **kwargs):
990
- """Generates examples from CzEng v1.6, with optional filtering for v1.7."""
991
- filter_path = kwargs.get("filter_path", None)
992
- if filter_path:
993
- re_block = re.compile(r"^[^-]+-b(\d+)-\d\d[tde]")
994
- with open(filter_path, encoding="utf-8") as f:
995
- bad_blocks = {blk for blk in re.search(r"qw{([\s\d]*)}", f.read()).groups()[0].split()}
996
- logger.info("Loaded %d bad blocks to filter from CzEng v1.6 to make v1.7.", len(bad_blocks))
997
-
998
- for path in paths:
999
- for gz_path in sorted(glob.glob(path)):
1000
- with open(gz_path, "rb") as g, gzip.GzipFile(fileobj=g) as f:
1001
- filename = os.path.basename(gz_path)
1002
- for line_id, line in enumerate(f):
1003
- line = line.decode("utf-8") # required for py3
1004
- if not line.strip():
1005
- continue
1006
- id_, unused_score, cs, en = line.split("\t")
1007
- if filter_path:
1008
- block_match = re.match(re_block, id_)
1009
- if block_match and block_match.groups()[0] in bad_blocks:
1010
- continue
1011
- sub_key = f"{filename}/{line_id}"
1012
- yield sub_key, {
1013
- "cs": cs.strip(),
1014
- "en": en.strip(),
1015
- }
1016
-
1017
-
1018
- def _parse_hindencorp(path):
1019
- with open(path, encoding="utf-8") as f:
1020
- for line_id, line in enumerate(f):
1021
- split_line = line.split("\t")
1022
- if len(split_line) != 5:
1023
- logger.warning("Skipping invalid HindEnCorp line: %s", line)
1024
- continue
1025
- yield line_id, {"translation": {"en": split_line[3].strip(), "hi": split_line[4].strip()}}