Datasets:
Updated script
Browse files- README.md +548 -1
- dataset_infos.json +1 -0
- iwslt2017_context.py +246 -0
- prepare_corpus.py +124 -0
README.md
CHANGED
@@ -1,3 +1,550 @@
|
|
1 |
---
|
2 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
annotations_creators:
|
3 |
+
- crowdsourced
|
4 |
+
language:
|
5 |
+
- ar
|
6 |
+
- de
|
7 |
+
- en
|
8 |
+
- fr
|
9 |
+
- it
|
10 |
+
- ja
|
11 |
+
- ko
|
12 |
+
- nl
|
13 |
+
- ro
|
14 |
+
- zh
|
15 |
+
language_creators:
|
16 |
+
- expert-generated
|
17 |
+
license:
|
18 |
+
- cc-by-nc-nd-4.0
|
19 |
+
multilinguality:
|
20 |
+
- translation
|
21 |
+
pretty_name: IWSLT 2017
|
22 |
+
size_categories:
|
23 |
+
- 1M<n<10M
|
24 |
+
source_datasets:
|
25 |
+
- original
|
26 |
+
task_categories:
|
27 |
+
- translation
|
28 |
+
task_ids: []
|
29 |
+
paperswithcode_id: iwslt-2017
|
30 |
+
dataset_info:
|
31 |
+
- config_name: iwslt2017-en-it
|
32 |
+
features:
|
33 |
+
- name: translation
|
34 |
+
dtype:
|
35 |
+
translation:
|
36 |
+
languages:
|
37 |
+
- en
|
38 |
+
- it
|
39 |
+
splits:
|
40 |
+
- name: train
|
41 |
+
num_bytes: 46647925
|
42 |
+
num_examples: 231619
|
43 |
+
- name: test
|
44 |
+
num_bytes: 305246
|
45 |
+
num_examples: 1566
|
46 |
+
- name: validation
|
47 |
+
num_bytes: 200023
|
48 |
+
num_examples: 929
|
49 |
+
download_size: 329391132
|
50 |
+
dataset_size: 47153194
|
51 |
+
- config_name: iwslt2017-en-nl
|
52 |
+
features:
|
53 |
+
- name: translation
|
54 |
+
dtype:
|
55 |
+
translation:
|
56 |
+
languages:
|
57 |
+
- en
|
58 |
+
- nl
|
59 |
+
splits:
|
60 |
+
- name: train
|
61 |
+
num_bytes: 42843933
|
62 |
+
num_examples: 237240
|
63 |
+
- name: test
|
64 |
+
num_bytes: 311646
|
65 |
+
num_examples: 1777
|
66 |
+
- name: validation
|
67 |
+
num_bytes: 197814
|
68 |
+
num_examples: 1003
|
69 |
+
download_size: 329391132
|
70 |
+
dataset_size: 43353393
|
71 |
+
- config_name: iwslt2017-en-ro
|
72 |
+
features:
|
73 |
+
- name: translation
|
74 |
+
dtype:
|
75 |
+
translation:
|
76 |
+
languages:
|
77 |
+
- en
|
78 |
+
- ro
|
79 |
+
splits:
|
80 |
+
- name: train
|
81 |
+
num_bytes: 44129950
|
82 |
+
num_examples: 220538
|
83 |
+
- name: test
|
84 |
+
num_bytes: 316790
|
85 |
+
num_examples: 1678
|
86 |
+
- name: validation
|
87 |
+
num_bytes: 205028
|
88 |
+
num_examples: 914
|
89 |
+
download_size: 329391132
|
90 |
+
dataset_size: 44651768
|
91 |
+
- config_name: iwslt2017-it-en
|
92 |
+
features:
|
93 |
+
- name: translation
|
94 |
+
dtype:
|
95 |
+
translation:
|
96 |
+
languages:
|
97 |
+
- it
|
98 |
+
- en
|
99 |
+
splits:
|
100 |
+
- name: train
|
101 |
+
num_bytes: 46647925
|
102 |
+
num_examples: 231619
|
103 |
+
- name: test
|
104 |
+
num_bytes: 305246
|
105 |
+
num_examples: 1566
|
106 |
+
- name: validation
|
107 |
+
num_bytes: 200023
|
108 |
+
num_examples: 929
|
109 |
+
download_size: 329391132
|
110 |
+
dataset_size: 47153194
|
111 |
+
- config_name: iwslt2017-it-nl
|
112 |
+
features:
|
113 |
+
- name: translation
|
114 |
+
dtype:
|
115 |
+
translation:
|
116 |
+
languages:
|
117 |
+
- it
|
118 |
+
- nl
|
119 |
+
splits:
|
120 |
+
- name: train
|
121 |
+
num_bytes: 43033168
|
122 |
+
num_examples: 233415
|
123 |
+
- name: test
|
124 |
+
num_bytes: 309725
|
125 |
+
num_examples: 1669
|
126 |
+
- name: validation
|
127 |
+
num_bytes: 197774
|
128 |
+
num_examples: 1001
|
129 |
+
download_size: 329391132
|
130 |
+
dataset_size: 43540667
|
131 |
+
- config_name: iwslt2017-it-ro
|
132 |
+
features:
|
133 |
+
- name: translation
|
134 |
+
dtype:
|
135 |
+
translation:
|
136 |
+
languages:
|
137 |
+
- it
|
138 |
+
- ro
|
139 |
+
splits:
|
140 |
+
- name: train
|
141 |
+
num_bytes: 44485169
|
142 |
+
num_examples: 217551
|
143 |
+
- name: test
|
144 |
+
num_bytes: 314974
|
145 |
+
num_examples: 1643
|
146 |
+
- name: validation
|
147 |
+
num_bytes: 204989
|
148 |
+
num_examples: 914
|
149 |
+
download_size: 329391132
|
150 |
+
dataset_size: 45005132
|
151 |
+
- config_name: iwslt2017-nl-en
|
152 |
+
features:
|
153 |
+
- name: translation
|
154 |
+
dtype:
|
155 |
+
translation:
|
156 |
+
languages:
|
157 |
+
- nl
|
158 |
+
- en
|
159 |
+
splits:
|
160 |
+
- name: train
|
161 |
+
num_bytes: 42843933
|
162 |
+
num_examples: 237240
|
163 |
+
- name: test
|
164 |
+
num_bytes: 311646
|
165 |
+
num_examples: 1777
|
166 |
+
- name: validation
|
167 |
+
num_bytes: 197814
|
168 |
+
num_examples: 1003
|
169 |
+
download_size: 329391132
|
170 |
+
dataset_size: 43353393
|
171 |
+
- config_name: iwslt2017-nl-it
|
172 |
+
features:
|
173 |
+
- name: translation
|
174 |
+
dtype:
|
175 |
+
translation:
|
176 |
+
languages:
|
177 |
+
- nl
|
178 |
+
- it
|
179 |
+
splits:
|
180 |
+
- name: train
|
181 |
+
num_bytes: 43033168
|
182 |
+
num_examples: 233415
|
183 |
+
- name: test
|
184 |
+
num_bytes: 309725
|
185 |
+
num_examples: 1669
|
186 |
+
- name: validation
|
187 |
+
num_bytes: 197774
|
188 |
+
num_examples: 1001
|
189 |
+
download_size: 329391132
|
190 |
+
dataset_size: 43540667
|
191 |
+
- config_name: iwslt2017-nl-ro
|
192 |
+
features:
|
193 |
+
- name: translation
|
194 |
+
dtype:
|
195 |
+
translation:
|
196 |
+
languages:
|
197 |
+
- nl
|
198 |
+
- ro
|
199 |
+
splits:
|
200 |
+
- name: train
|
201 |
+
num_bytes: 41338738
|
202 |
+
num_examples: 206920
|
203 |
+
- name: test
|
204 |
+
num_bytes: 320952
|
205 |
+
num_examples: 1680
|
206 |
+
- name: validation
|
207 |
+
num_bytes: 202380
|
208 |
+
num_examples: 913
|
209 |
+
download_size: 329391132
|
210 |
+
dataset_size: 41862070
|
211 |
+
- config_name: iwslt2017-ro-en
|
212 |
+
features:
|
213 |
+
- name: translation
|
214 |
+
dtype:
|
215 |
+
translation:
|
216 |
+
languages:
|
217 |
+
- ro
|
218 |
+
- en
|
219 |
+
splits:
|
220 |
+
- name: train
|
221 |
+
num_bytes: 44129950
|
222 |
+
num_examples: 220538
|
223 |
+
- name: test
|
224 |
+
num_bytes: 316790
|
225 |
+
num_examples: 1678
|
226 |
+
- name: validation
|
227 |
+
num_bytes: 205028
|
228 |
+
num_examples: 914
|
229 |
+
download_size: 329391132
|
230 |
+
dataset_size: 44651768
|
231 |
+
- config_name: iwslt2017-ro-it
|
232 |
+
features:
|
233 |
+
- name: translation
|
234 |
+
dtype:
|
235 |
+
translation:
|
236 |
+
languages:
|
237 |
+
- ro
|
238 |
+
- it
|
239 |
+
splits:
|
240 |
+
- name: train
|
241 |
+
num_bytes: 44485169
|
242 |
+
num_examples: 217551
|
243 |
+
- name: test
|
244 |
+
num_bytes: 314974
|
245 |
+
num_examples: 1643
|
246 |
+
- name: validation
|
247 |
+
num_bytes: 204989
|
248 |
+
num_examples: 914
|
249 |
+
download_size: 329391132
|
250 |
+
dataset_size: 45005132
|
251 |
+
- config_name: iwslt2017-ro-nl
|
252 |
+
features:
|
253 |
+
- name: translation
|
254 |
+
dtype:
|
255 |
+
translation:
|
256 |
+
languages:
|
257 |
+
- ro
|
258 |
+
- nl
|
259 |
+
splits:
|
260 |
+
- name: train
|
261 |
+
num_bytes: 41338738
|
262 |
+
num_examples: 206920
|
263 |
+
- name: test
|
264 |
+
num_bytes: 320952
|
265 |
+
num_examples: 1680
|
266 |
+
- name: validation
|
267 |
+
num_bytes: 202380
|
268 |
+
num_examples: 913
|
269 |
+
download_size: 329391132
|
270 |
+
dataset_size: 41862070
|
271 |
+
- config_name: iwslt2017-ar-en
|
272 |
+
features:
|
273 |
+
- name: translation
|
274 |
+
dtype:
|
275 |
+
translation:
|
276 |
+
languages:
|
277 |
+
- ar
|
278 |
+
- en
|
279 |
+
splits:
|
280 |
+
- name: train
|
281 |
+
num_bytes: 56481059
|
282 |
+
num_examples: 231713
|
283 |
+
- name: test
|
284 |
+
num_bytes: 2014296
|
285 |
+
num_examples: 8583
|
286 |
+
- name: validation
|
287 |
+
num_bytes: 241206
|
288 |
+
num_examples: 888
|
289 |
+
download_size: 27748780
|
290 |
+
dataset_size: 58736561
|
291 |
+
- config_name: iwslt2017-de-en
|
292 |
+
features:
|
293 |
+
- name: translation
|
294 |
+
dtype:
|
295 |
+
translation:
|
296 |
+
languages:
|
297 |
+
- de
|
298 |
+
- en
|
299 |
+
splits:
|
300 |
+
- name: train
|
301 |
+
num_bytes: 42608380
|
302 |
+
num_examples: 206112
|
303 |
+
- name: test
|
304 |
+
num_bytes: 1608474
|
305 |
+
num_examples: 8079
|
306 |
+
- name: validation
|
307 |
+
num_bytes: 210975
|
308 |
+
num_examples: 888
|
309 |
+
download_size: 16758320
|
310 |
+
dataset_size: 44427829
|
311 |
+
- config_name: iwslt2017-en-ar
|
312 |
+
features:
|
313 |
+
- name: translation
|
314 |
+
dtype:
|
315 |
+
translation:
|
316 |
+
languages:
|
317 |
+
- en
|
318 |
+
- ar
|
319 |
+
splits:
|
320 |
+
- name: train
|
321 |
+
num_bytes: 56481059
|
322 |
+
num_examples: 231713
|
323 |
+
- name: test
|
324 |
+
num_bytes: 2014296
|
325 |
+
num_examples: 8583
|
326 |
+
- name: validation
|
327 |
+
num_bytes: 241206
|
328 |
+
num_examples: 888
|
329 |
+
download_size: 29333173
|
330 |
+
dataset_size: 58736561
|
331 |
+
- config_name: iwslt2017-en-de
|
332 |
+
features:
|
333 |
+
- name: translation
|
334 |
+
dtype:
|
335 |
+
translation:
|
336 |
+
languages:
|
337 |
+
- en
|
338 |
+
- de
|
339 |
+
splits:
|
340 |
+
- name: train
|
341 |
+
num_bytes: 42608380
|
342 |
+
num_examples: 206112
|
343 |
+
- name: test
|
344 |
+
num_bytes: 1608474
|
345 |
+
num_examples: 8079
|
346 |
+
- name: validation
|
347 |
+
num_bytes: 210975
|
348 |
+
num_examples: 888
|
349 |
+
download_size: 16758334
|
350 |
+
dataset_size: 44427829
|
351 |
+
- config_name: iwslt2017-en-fr
|
352 |
+
features:
|
353 |
+
- name: translation
|
354 |
+
dtype:
|
355 |
+
translation:
|
356 |
+
languages:
|
357 |
+
- en
|
358 |
+
- fr
|
359 |
+
splits:
|
360 |
+
- name: train
|
361 |
+
num_bytes: 49273286
|
362 |
+
num_examples: 232825
|
363 |
+
- name: test
|
364 |
+
num_bytes: 1767465
|
365 |
+
num_examples: 8597
|
366 |
+
- name: validation
|
367 |
+
num_bytes: 207579
|
368 |
+
num_examples: 890
|
369 |
+
download_size: 27699724
|
370 |
+
dataset_size: 51248330
|
371 |
+
- config_name: iwslt2017-en-ja
|
372 |
+
features:
|
373 |
+
- name: translation
|
374 |
+
dtype:
|
375 |
+
translation:
|
376 |
+
languages:
|
377 |
+
- en
|
378 |
+
- ja
|
379 |
+
splits:
|
380 |
+
- name: train
|
381 |
+
num_bytes: 48204987
|
382 |
+
num_examples: 223108
|
383 |
+
- name: test
|
384 |
+
num_bytes: 1809007
|
385 |
+
num_examples: 8469
|
386 |
+
- name: validation
|
387 |
+
num_bytes: 208124
|
388 |
+
num_examples: 871
|
389 |
+
download_size: 26983602
|
390 |
+
dataset_size: 50222118
|
391 |
+
- config_name: iwslt2017-en-ko
|
392 |
+
features:
|
393 |
+
- name: translation
|
394 |
+
dtype:
|
395 |
+
translation:
|
396 |
+
languages:
|
397 |
+
- en
|
398 |
+
- ko
|
399 |
+
splits:
|
400 |
+
- name: train
|
401 |
+
num_bytes: 51678043
|
402 |
+
num_examples: 230240
|
403 |
+
- name: test
|
404 |
+
num_bytes: 1869793
|
405 |
+
num_examples: 8514
|
406 |
+
- name: validation
|
407 |
+
num_bytes: 219295
|
408 |
+
num_examples: 879
|
409 |
+
download_size: 19364776
|
410 |
+
dataset_size: 53767131
|
411 |
+
- config_name: iwslt2017-en-zh
|
412 |
+
features:
|
413 |
+
- name: translation
|
414 |
+
dtype:
|
415 |
+
translation:
|
416 |
+
languages:
|
417 |
+
- en
|
418 |
+
- zh
|
419 |
+
splits:
|
420 |
+
- name: train
|
421 |
+
num_bytes: 44271004
|
422 |
+
num_examples: 231266
|
423 |
+
- name: test
|
424 |
+
num_bytes: 1605527
|
425 |
+
num_examples: 8549
|
426 |
+
- name: validation
|
427 |
+
num_bytes: 202537
|
428 |
+
num_examples: 879
|
429 |
+
download_size: 27597071
|
430 |
+
dataset_size: 46079068
|
431 |
+
- config_name: iwslt2017-fr-en
|
432 |
+
features:
|
433 |
+
- name: translation
|
434 |
+
dtype:
|
435 |
+
translation:
|
436 |
+
languages:
|
437 |
+
- fr
|
438 |
+
- en
|
439 |
+
splits:
|
440 |
+
- name: train
|
441 |
+
num_bytes: 49273286
|
442 |
+
num_examples: 232825
|
443 |
+
- name: test
|
444 |
+
num_bytes: 1767465
|
445 |
+
num_examples: 8597
|
446 |
+
- name: validation
|
447 |
+
num_bytes: 207579
|
448 |
+
num_examples: 890
|
449 |
+
download_size: 26880731
|
450 |
+
dataset_size: 51248330
|
451 |
+
- config_name: iwslt2017-ja-en
|
452 |
+
features:
|
453 |
+
- name: translation
|
454 |
+
dtype:
|
455 |
+
translation:
|
456 |
+
languages:
|
457 |
+
- ja
|
458 |
+
- en
|
459 |
+
splits:
|
460 |
+
- name: train
|
461 |
+
num_bytes: 48204987
|
462 |
+
num_examples: 223108
|
463 |
+
- name: test
|
464 |
+
num_bytes: 1809007
|
465 |
+
num_examples: 8469
|
466 |
+
- name: validation
|
467 |
+
num_bytes: 208124
|
468 |
+
num_examples: 871
|
469 |
+
download_size: 26190859
|
470 |
+
dataset_size: 50222118
|
471 |
+
- config_name: iwslt2017-ko-en
|
472 |
+
features:
|
473 |
+
- name: translation
|
474 |
+
dtype:
|
475 |
+
translation:
|
476 |
+
languages:
|
477 |
+
- ko
|
478 |
+
- en
|
479 |
+
splits:
|
480 |
+
- name: train
|
481 |
+
num_bytes: 51678043
|
482 |
+
num_examples: 230240
|
483 |
+
- name: test
|
484 |
+
num_bytes: 1869793
|
485 |
+
num_examples: 8514
|
486 |
+
- name: validation
|
487 |
+
num_bytes: 219295
|
488 |
+
num_examples: 879
|
489 |
+
download_size: 19364733
|
490 |
+
dataset_size: 53767131
|
491 |
+
- config_name: iwslt2017-zh-en
|
492 |
+
features:
|
493 |
+
- name: translation
|
494 |
+
dtype:
|
495 |
+
translation:
|
496 |
+
languages:
|
497 |
+
- zh
|
498 |
+
- en
|
499 |
+
splits:
|
500 |
+
- name: train
|
501 |
+
num_bytes: 44271004
|
502 |
+
num_examples: 231266
|
503 |
+
- name: test
|
504 |
+
num_bytes: 1605527
|
505 |
+
num_examples: 8549
|
506 |
+
- name: validation
|
507 |
+
num_bytes: 202537
|
508 |
+
num_examples: 879
|
509 |
+
download_size: 26849290
|
510 |
+
dataset_size: 46079068
|
511 |
---
|
512 |
+
|
513 |
+
# Dataset Card for IWSLT 2017
|
514 |
+
|
515 |
+
## Table of Contents
|
516 |
+
- [Dataset Description](#dataset-description)
|
517 |
+
- [Dataset Summary](#dataset-summary)
|
518 |
+
- [Supported Tasks and Leaderboards](#supported-tasks-and-leaderboards)
|
519 |
+
- [Languages](#languages)
|
520 |
+
- [Dataset Structure](#dataset-structure)
|
521 |
+
- [Data Instances](#data-instances)
|
522 |
+
- [Data Fields](#data-fields)
|
523 |
+
- [Data Splits](#data-splits)
|
524 |
+
- [Dataset Creation](#dataset-creation)
|
525 |
+
- [Curation Rationale](#curation-rationale)
|
526 |
+
- [Source Data](#source-data)
|
527 |
+
- [Annotations](#annotations)
|
528 |
+
- [Personal and Sensitive Information](#personal-and-sensitive-information)
|
529 |
+
- [Considerations for Using the Data](#considerations-for-using-the-data)
|
530 |
+
- [Social Impact of Dataset](#social-impact-of-dataset)
|
531 |
+
- [Discussion of Biases](#discussion-of-biases)
|
532 |
+
- [Other Known Limitations](#other-known-limitations)
|
533 |
+
- [Additional Information](#additional-information)
|
534 |
+
- [Dataset Curators](#dataset-curators)
|
535 |
+
- [Licensing Information](#licensing-information)
|
536 |
+
- [Citation Information](#citation-information)
|
537 |
+
- [Contributions](#contributions)
|
538 |
+
|
539 |
+
## Dataset Description
|
540 |
+
|
541 |
+
- **Homepage:** [https://sites.google.com/site/iwsltevaluation2017/TED-tasks](https://sites.google.com/site/iwsltevaluation2017/TED-tasks)
|
542 |
+
- **Repository:** [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
543 |
+
- **Paper:** [Overview of the IWSLT 2017 Evaluation Campaign](https://aclanthology.org/2017.iwslt-1.1/)
|
544 |
+
- **Point of Contact:** [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
|
545 |
+
- **Size of downloaded dataset files:** 4.24 GB
|
546 |
+
- **Size of the generated dataset:** 1.14 GB
|
547 |
+
- **Total amount of disk used:** 5.38 GB
|
548 |
+
|
549 |
+
*This repository contain a modified version of the loading script used in the official [iwslt2017](https://huggingface.co/datasets/iwslt2017) repository updated to include document and segment information for all available sentence pairs, enabling their usage for document-level and context-aware MT applications. Refer to the original repository for additional information.*
|
550 |
+
|
dataset_infos.json
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
{"iwslt2017-en-it": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["en", "it"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-en-it", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 48662753, "num_examples": 233521, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 318698, "num_examples": 1577, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 208123, "num_examples": 937, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnmted/texts/DeEnItNlRo/DeEnItNlRo/DeEnItNlRo-DeEnItNlRo.zip": {"num_bytes": 329391132, "checksum": "21f9bb62d1358534fc8d97e1b89ad80e80f95cbba231155922af45e0e57a1a61"}}, "download_size": 329391132, "post_processing_size": null, "dataset_size": 49189574, "size_in_bytes": 378580706}, "iwslt2017-en-nl": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["en", "nl"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-en-nl", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 44895457, "num_examples": 239045, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 326786, "num_examples": 1788, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 206506, "num_examples": 1011, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnmted/texts/DeEnItNlRo/DeEnItNlRo/DeEnItNlRo-DeEnItNlRo.zip": {"num_bytes": 329391132, "checksum": "21f9bb62d1358534fc8d97e1b89ad80e80f95cbba231155922af45e0e57a1a61"}}, "download_size": 329391132, "post_processing_size": null, "dataset_size": 45428749, "size_in_bytes": 374819881}, "iwslt2017-en-ro": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["en", "ro"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-en-ro", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 46048386, "num_examples": 222350, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 331138, "num_examples": 1689, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 213008, "num_examples": 922, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnmted/texts/DeEnItNlRo/DeEnItNlRo/DeEnItNlRo-DeEnItNlRo.zip": {"num_bytes": 329391132, "checksum": "21f9bb62d1358534fc8d97e1b89ad80e80f95cbba231155922af45e0e57a1a61"}}, "download_size": 329391132, "post_processing_size": null, "dataset_size": 46592532, "size_in_bytes": 375983664}, "iwslt2017-it-en": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["it", "en"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-it-en", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 48662753, "num_examples": 233521, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 318698, "num_examples": 1577, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 208123, "num_examples": 937, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnmted/texts/DeEnItNlRo/DeEnItNlRo/DeEnItNlRo-DeEnItNlRo.zip": {"num_bytes": 329391132, "checksum": "21f9bb62d1358534fc8d97e1b89ad80e80f95cbba231155922af45e0e57a1a61"}}, "download_size": 329391132, "post_processing_size": null, "dataset_size": 49189574, "size_in_bytes": 378580706}, "iwslt2017-it-nl": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["it", "nl"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-it-nl", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 45051602, "num_examples": 235191, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 324001, "num_examples": 1680, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 206450, "num_examples": 1009, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnmted/texts/DeEnItNlRo/DeEnItNlRo/DeEnItNlRo-DeEnItNlRo.zip": {"num_bytes": 329391132, "checksum": "21f9bb62d1358534fc8d97e1b89ad80e80f95cbba231155922af45e0e57a1a61"}}, "download_size": 329391132, "post_processing_size": null, "dataset_size": 45582053, "size_in_bytes": 374973185}, "iwslt2017-it-ro": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["it", "ro"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-it-ro", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 46377735, "num_examples": 219340, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 329042, "num_examples": 1654, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 212969, "num_examples": 922, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnmted/texts/DeEnItNlRo/DeEnItNlRo/DeEnItNlRo-DeEnItNlRo.zip": {"num_bytes": 329391132, "checksum": "21f9bb62d1358534fc8d97e1b89ad80e80f95cbba231155922af45e0e57a1a61"}}, "download_size": 329391132, "post_processing_size": null, "dataset_size": 46919746, "size_in_bytes": 376310878}, "iwslt2017-nl-en": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["nl", "en"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-nl-en", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 44895457, "num_examples": 239045, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 326786, "num_examples": 1788, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 206506, "num_examples": 1011, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnmted/texts/DeEnItNlRo/DeEnItNlRo/DeEnItNlRo-DeEnItNlRo.zip": {"num_bytes": 329391132, "checksum": "21f9bb62d1358534fc8d97e1b89ad80e80f95cbba231155922af45e0e57a1a61"}}, "download_size": 329391132, "post_processing_size": null, "dataset_size": 45428749, "size_in_bytes": 374819881}, "iwslt2017-nl-it": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["nl", "it"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-nl-it", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 45051602, "num_examples": 235191, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 324001, "num_examples": 1680, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 206450, "num_examples": 1009, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnmted/texts/DeEnItNlRo/DeEnItNlRo/DeEnItNlRo-DeEnItNlRo.zip": {"num_bytes": 329391132, "checksum": "21f9bb62d1358534fc8d97e1b89ad80e80f95cbba231155922af45e0e57a1a61"}}, "download_size": 329391132, "post_processing_size": null, "dataset_size": 45582053, "size_in_bytes": 374973185}, "iwslt2017-nl-ro": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["nl", "ro"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-nl-ro", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 43140478, "num_examples": 208641, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 335316, "num_examples": 1691, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 210352, "num_examples": 921, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnmted/texts/DeEnItNlRo/DeEnItNlRo/DeEnItNlRo-DeEnItNlRo.zip": {"num_bytes": 329391132, "checksum": "21f9bb62d1358534fc8d97e1b89ad80e80f95cbba231155922af45e0e57a1a61"}}, "download_size": 329391132, "post_processing_size": null, "dataset_size": 43686146, "size_in_bytes": 373077278}, "iwslt2017-ro-en": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["ro", "en"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-ro-en", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 46048386, "num_examples": 222350, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 331138, "num_examples": 1689, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 213008, "num_examples": 922, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnmted/texts/DeEnItNlRo/DeEnItNlRo/DeEnItNlRo-DeEnItNlRo.zip": {"num_bytes": 329391132, "checksum": "21f9bb62d1358534fc8d97e1b89ad80e80f95cbba231155922af45e0e57a1a61"}}, "download_size": 329391132, "post_processing_size": null, "dataset_size": 46592532, "size_in_bytes": 375983664}, "iwslt2017-ro-it": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["ro", "it"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-ro-it", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 46377735, "num_examples": 219340, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 329042, "num_examples": 1654, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 212969, "num_examples": 922, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnmted/texts/DeEnItNlRo/DeEnItNlRo/DeEnItNlRo-DeEnItNlRo.zip": {"num_bytes": 329391132, "checksum": "21f9bb62d1358534fc8d97e1b89ad80e80f95cbba231155922af45e0e57a1a61"}}, "download_size": 329391132, "post_processing_size": null, "dataset_size": 46919746, "size_in_bytes": 376310878}, "iwslt2017-ro-nl": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["ro", "nl"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-ro-nl", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 43140478, "num_examples": 208641, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 335316, "num_examples": 1691, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 210352, "num_examples": 921, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnmted/texts/DeEnItNlRo/DeEnItNlRo/DeEnItNlRo-DeEnItNlRo.zip": {"num_bytes": 329391132, "checksum": "21f9bb62d1358534fc8d97e1b89ad80e80f95cbba231155922af45e0e57a1a61"}}, "download_size": 329391132, "post_processing_size": null, "dataset_size": 43686146, "size_in_bytes": 373077278}, "iwslt2017-ar-en": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["ar", "en"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-ar-en", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 58497073, "num_examples": 233620, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 2090592, "num_examples": 8672, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 248978, "num_examples": 896, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnted/texts/ar/en/ar-en.zip": {"num_bytes": 27748780, "checksum": "9536bee4f28b4870cc045aa2d9460f0647f3161f874e74c171e818674e9dae1a"}}, "download_size": 27748780, "post_processing_size": null, "dataset_size": 60836643, "size_in_bytes": 88585423}, "iwslt2017-de-en": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["de", "en"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-de-en", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 44402244, "num_examples": 207817, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 1680394, "num_examples": 8164, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 218747, "num_examples": 896, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnted/texts/de/en/de-en.zip": {"num_bytes": 16758320, "checksum": "a2d55063175c732c5e6569af32afa528a72e14ddf07cbbe5e2ef11435b909dce"}}, "download_size": 16758320, "post_processing_size": null, "dataset_size": 46301385, "size_in_bytes": 63059705}, "iwslt2017-en-ar": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["en", "ar"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-en-ar", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 58497073, "num_examples": 233620, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 2090592, "num_examples": 8672, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 248978, "num_examples": 896, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnted/texts/en/ar/en-ar.zip": {"num_bytes": 29333173, "checksum": "05790036a1bfb74a27452005b2274764c268fd6d9f7aa3d7a64fc7b5cc774368"}}, "download_size": 29333173, "post_processing_size": null, "dataset_size": 60836643, "size_in_bytes": 90169816}, "iwslt2017-en-de": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["en", "de"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-en-de", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 44402244, "num_examples": 207817, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 1680394, "num_examples": 8164, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 218747, "num_examples": 896, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnted/texts/en/de/en-de.zip": {"num_bytes": 16758334, "checksum": "d64c16ea33b6d49c44ff5f11768f44a382ed1c0cba351a00c776f227fad80e39"}}, "download_size": 16758334, "post_processing_size": null, "dataset_size": 46301385, "size_in_bytes": 63059719}, "iwslt2017-en-fr": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["en", "fr"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-en-fr", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 51298794, "num_examples": 234739, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 1843873, "num_examples": 8686, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 215367, "num_examples": 898, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnted/texts/en/fr/en-fr.zip": {"num_bytes": 27699724, "checksum": "aa5b3b622e9389dfebc9b99ad22d7b21cdccb49be4c67e3e0a9f22573a1b500d"}}, "download_size": 27699724, "post_processing_size": null, "dataset_size": 53358034, "size_in_bytes": 81057758}, "iwslt2017-en-ja": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["en", "ja"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-en-ja", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 50148433, "num_examples": 224971, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 1884391, "num_examples": 8558, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 215760, "num_examples": 879, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnted/texts/en/ja/en-ja.zip": {"num_bytes": 26983602, "checksum": "3ec1dec974c3a96202997a00cf593a32ca3b411e1f57740323c14bfdb4ce87a7"}}, "download_size": 26983602, "post_processing_size": null, "dataset_size": 52248584, "size_in_bytes": 79232186}, "iwslt2017-en-ko": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["en", "ko"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-en-ko", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 53683393, "num_examples": 232160, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 1945537, "num_examples": 8603, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 226995, "num_examples": 887, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnted/texts/en/ko/en-ko.zip": {"num_bytes": 19364776, "checksum": "cb0cae0959c859ae09c64d0623ef0e87a759c22d83e06b9c7ad05f1bbbbbb6cc"}}, "download_size": 19364776, "post_processing_size": null, "dataset_size": 55855925, "size_in_bytes": 75220701}, "iwslt2017-en-zh": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["en", "zh"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-en-zh", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 46283352, "num_examples": 233172, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 1681551, "num_examples": 8638, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 210237, "num_examples": 887, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnted/texts/en/zh/en-zh.zip": {"num_bytes": 27597071, "checksum": "33bec34a7234b5fb5faf09977b19569072c454a4fe79f5e7210d04b3d0e3184f"}}, "download_size": 27597071, "post_processing_size": null, "dataset_size": 48175140, "size_in_bytes": 75772211}, "iwslt2017-fr-en": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["fr", "en"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-fr-en", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 51298794, "num_examples": 234739, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 1843873, "num_examples": 8686, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 215367, "num_examples": 898, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnted/texts/fr/en/fr-en.zip": {"num_bytes": 26880731, "checksum": "44dbb42bae10ca94880f4b3fe5be33f6b5f2390d217311edfb06abadb8fd16c2"}}, "download_size": 26880731, "post_processing_size": null, "dataset_size": 53358034, "size_in_bytes": 80238765}, "iwslt2017-ja-en": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["ja", "en"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-ja-en", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 50148433, "num_examples": 224971, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 1884391, "num_examples": 8558, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 215760, "num_examples": 879, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnted/texts/ja/en/ja-en.zip": {"num_bytes": 26190859, "checksum": "a923cdaa5632e55a94e799d31468c58fd2eab290a34f8b950a562f6ff20046b6"}}, "download_size": 26190859, "post_processing_size": null, "dataset_size": 52248584, "size_in_bytes": 78439443}, "iwslt2017-ko-en": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["ko", "en"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-ko-en", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 53683393, "num_examples": 232160, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 1945537, "num_examples": 8603, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 226995, "num_examples": 887, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnted/texts/ko/en/ko-en.zip": {"num_bytes": 19364733, "checksum": "63257a22a2b21ed816d2677ea426898f4788041f9573e0e7bd1a5ceb8ad63bbb"}}, "download_size": 19364733, "post_processing_size": null, "dataset_size": 55855925, "size_in_bytes": 75220658}, "iwslt2017-zh-en": {"description": "The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.\n", "citation": "@inproceedings{cettolo-etal-2017-overview,\n title = \"Overview of the {IWSLT} 2017 Evaluation Campaign\",\n author = {Cettolo, Mauro and\n Federico, Marcello and\n Bentivogli, Luisa and\n Niehues, Jan and\n St{\\\"u}ker, Sebastian and\n Sudoh, Katsuhito and\n Yoshino, Koichiro and\n Federmann, Christian},\n booktitle = \"Proceedings of the 14th International Conference on Spoken Language Translation\",\n month = dec # \" 14-15\",\n year = \"2017\",\n address = \"Tokyo, Japan\",\n publisher = \"International Workshop on Spoken Language Translation\",\n url = \"https://aclanthology.org/2017.iwslt-1.1\",\n pages = \"2--14\",\n}\n", "homepage": "https://sites.google.com/site/iwsltevaluation2017/TED-tasks", "license": "", "features": {"doc_id": {"dtype": "int32", "id": null, "_type": "Value"}, "seg_id": {"dtype": "int32", "id": null, "_type": "Value"}, "translation": {"languages": ["zh", "en"], "id": null, "_type": "Translation"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "iwslt2017_context", "config_name": "iwslt2017-zh-en", "version": {"version_str": "1.0.0", "description": null, "major": 1, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 46283352, "num_examples": 233172, "dataset_name": "iwslt2017_context"}, "test": {"name": "test", "num_bytes": 1681551, "num_examples": 8638, "dataset_name": "iwslt2017_context"}, "validation": {"name": "validation", "num_bytes": 210237, "num_examples": 887, "dataset_name": "iwslt2017_context"}}, "download_checksums": {"https://huggingface.co/datasets/iwslt2017/resolve/main/data/2017-01-trnted/texts/zh/en/zh-en.zip": {"num_bytes": 26849290, "checksum": "32849811afa74a69eeb70b4baa17e7b7c6a6626227c874d3f15630811f8b0104"}}, "download_size": 26849290, "post_processing_size": null, "dataset_size": 48175140, "size_in_bytes": 75024430}}
|
iwslt2017_context.py
ADDED
@@ -0,0 +1,246 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# coding=utf-8
|
2 |
+
# Copyright 2020 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 |
+
"""IWSLT 2017 dataset """
|
16 |
+
|
17 |
+
|
18 |
+
import os
|
19 |
+
import re
|
20 |
+
|
21 |
+
import datasets
|
22 |
+
|
23 |
+
|
24 |
+
_HOMEPAGE = "https://sites.google.com/site/iwsltevaluation2017/TED-tasks"
|
25 |
+
|
26 |
+
_DESCRIPTION = """\
|
27 |
+
The IWSLT 2017 Multilingual Task addresses text translation, including zero-shot translation, with a single MT system across all directions including English, German, Dutch, Italian and Romanian. As unofficial task, conventional bilingual text translation is offered between English and Arabic, French, Japanese, Chinese, German and Korean.
|
28 |
+
"""
|
29 |
+
|
30 |
+
_CITATION = """\
|
31 |
+
@inproceedings{cettolo-etal-2017-overview,
|
32 |
+
title = "Overview of the {IWSLT} 2017 Evaluation Campaign",
|
33 |
+
author = {Cettolo, Mauro and
|
34 |
+
Federico, Marcello and
|
35 |
+
Bentivogli, Luisa and
|
36 |
+
Niehues, Jan and
|
37 |
+
St{\\"u}ker, Sebastian and
|
38 |
+
Sudoh, Katsuhito and
|
39 |
+
Yoshino, Koichiro and
|
40 |
+
Federmann, Christian},
|
41 |
+
booktitle = "Proceedings of the 14th International Conference on Spoken Language Translation",
|
42 |
+
month = dec # " 14-15",
|
43 |
+
year = "2017",
|
44 |
+
address = "Tokyo, Japan",
|
45 |
+
publisher = "International Workshop on Spoken Language Translation",
|
46 |
+
url = "https://aclanthology.org/2017.iwslt-1.1",
|
47 |
+
pages = "2--14",
|
48 |
+
}
|
49 |
+
"""
|
50 |
+
|
51 |
+
REPO_URL = "https://huggingface.co/datasets/iwslt2017/resolve/main/"
|
52 |
+
MULTI_URL = REPO_URL + "data/2017-01-trnmted/texts/DeEnItNlRo/DeEnItNlRo/DeEnItNlRo-DeEnItNlRo.zip"
|
53 |
+
BI_URL = REPO_URL + "data/2017-01-trnted/texts/{source}/{target}/{source}-{target}.zip"
|
54 |
+
|
55 |
+
DOCID_PATTERN = r'<doc docid="(\d+)"'
|
56 |
+
SEG_PATTERN = r'<seg id="(\d+)">(.*)</seg>'
|
57 |
+
|
58 |
+
|
59 |
+
class IWSLT2017Config(datasets.BuilderConfig):
|
60 |
+
"""BuilderConfig for NewDataset"""
|
61 |
+
|
62 |
+
def __init__(self, pair, is_multilingual, **kwargs):
|
63 |
+
"""
|
64 |
+
|
65 |
+
Args:
|
66 |
+
pair: the language pair to consider
|
67 |
+
is_multilingual: Is this pair in the multilingual dataset (download source is different)
|
68 |
+
**kwargs: keyword arguments forwarded to super.
|
69 |
+
"""
|
70 |
+
self.pair = pair
|
71 |
+
self.is_multilingual = is_multilingual
|
72 |
+
super().__init__(**kwargs)
|
73 |
+
|
74 |
+
|
75 |
+
# XXX: Artificially removed DE from here, as it also exists within bilingual data
|
76 |
+
MULTI_LANGUAGES = ["en", "it", "nl", "ro"]
|
77 |
+
BI_LANGUAGES = ["ar", "de", "en", "fr", "ja", "ko", "zh"]
|
78 |
+
MULTI_PAIRS = [f"{source}-{target}" for source in MULTI_LANGUAGES for target in MULTI_LANGUAGES if source != target]
|
79 |
+
BI_PAIRS = [
|
80 |
+
f"{source}-{target}"
|
81 |
+
for source in BI_LANGUAGES
|
82 |
+
for target in BI_LANGUAGES
|
83 |
+
if source != target and (source == "en" or target == "en")
|
84 |
+
]
|
85 |
+
|
86 |
+
PAIRS = MULTI_PAIRS + BI_PAIRS
|
87 |
+
|
88 |
+
|
89 |
+
class IWSLT217(datasets.GeneratorBasedBuilder):
|
90 |
+
"""The IWSLT 2017 Evaluation Campaign includes a multilingual TED Talks MT task."""
|
91 |
+
|
92 |
+
VERSION = datasets.Version("1.0.0")
|
93 |
+
|
94 |
+
# This is an example of a dataset with multiple configurations.
|
95 |
+
# If you don't want/need to define several sub-sets in your dataset,
|
96 |
+
# just remove the BUILDER_CONFIG_CLASS and the BUILDER_CONFIGS attributes.
|
97 |
+
BUILDER_CONFIG_CLASS = IWSLT2017Config
|
98 |
+
BUILDER_CONFIGS = [
|
99 |
+
IWSLT2017Config(
|
100 |
+
name="iwslt2017-" + pair,
|
101 |
+
description="A small dataset",
|
102 |
+
version=datasets.Version("1.0.0"),
|
103 |
+
pair=pair,
|
104 |
+
is_multilingual=pair in MULTI_PAIRS,
|
105 |
+
)
|
106 |
+
for pair in PAIRS
|
107 |
+
]
|
108 |
+
|
109 |
+
def _info(self):
|
110 |
+
return datasets.DatasetInfo(
|
111 |
+
description=_DESCRIPTION,
|
112 |
+
features=datasets.Features(
|
113 |
+
{
|
114 |
+
"doc_id": datasets.features.Value("int32"),
|
115 |
+
"seg_id": datasets.features.Value("int32"),
|
116 |
+
"translation": datasets.features.Translation(languages=self.config.pair.split("-"))
|
117 |
+
|
118 |
+
}
|
119 |
+
),
|
120 |
+
homepage=_HOMEPAGE,
|
121 |
+
citation=_CITATION,
|
122 |
+
)
|
123 |
+
|
124 |
+
def _split_generators(self, dl_manager):
|
125 |
+
"""Returns SplitGenerators."""
|
126 |
+
source, target = self.config.pair.split("-")
|
127 |
+
if self.config.is_multilingual:
|
128 |
+
dl_dir = dl_manager.download_and_extract(MULTI_URL)
|
129 |
+
data_dir = os.path.join(dl_dir, "DeEnItNlRo-DeEnItNlRo")
|
130 |
+
years = [2010]
|
131 |
+
else:
|
132 |
+
bi_url = BI_URL.format(source=source, target=target)
|
133 |
+
dl_dir = dl_manager.download_and_extract(bi_url)
|
134 |
+
data_dir = os.path.join(dl_dir, f"{source}-{target}")
|
135 |
+
years = [2010, 2011, 2012, 2013, 2014, 2015]
|
136 |
+
return [
|
137 |
+
datasets.SplitGenerator(
|
138 |
+
name=datasets.Split.TRAIN,
|
139 |
+
gen_kwargs={
|
140 |
+
"split": "train",
|
141 |
+
"source_files": [
|
142 |
+
os.path.join(
|
143 |
+
data_dir,
|
144 |
+
f"train.tags.{self.config.pair}.{source}",
|
145 |
+
)
|
146 |
+
],
|
147 |
+
"target_files": [
|
148 |
+
os.path.join(
|
149 |
+
data_dir,
|
150 |
+
f"train.tags.{self.config.pair}.{target}",
|
151 |
+
)
|
152 |
+
],
|
153 |
+
},
|
154 |
+
),
|
155 |
+
datasets.SplitGenerator(
|
156 |
+
name=datasets.Split.TEST,
|
157 |
+
gen_kwargs={
|
158 |
+
"split": "test",
|
159 |
+
"source_files": [
|
160 |
+
os.path.join(
|
161 |
+
data_dir,
|
162 |
+
f"IWSLT17.TED.tst{year}.{self.config.pair}.{source}.xml",
|
163 |
+
)
|
164 |
+
for year in years
|
165 |
+
],
|
166 |
+
"target_files": [
|
167 |
+
os.path.join(
|
168 |
+
data_dir,
|
169 |
+
f"IWSLT17.TED.tst{year}.{self.config.pair}.{target}.xml",
|
170 |
+
)
|
171 |
+
for year in years
|
172 |
+
],
|
173 |
+
},
|
174 |
+
),
|
175 |
+
datasets.SplitGenerator(
|
176 |
+
name=datasets.Split.VALIDATION,
|
177 |
+
gen_kwargs={
|
178 |
+
"split": "validation",
|
179 |
+
"source_files": [
|
180 |
+
os.path.join(
|
181 |
+
data_dir,
|
182 |
+
f"IWSLT17.TED.dev2010.{self.config.pair}.{source}.xml",
|
183 |
+
)
|
184 |
+
],
|
185 |
+
"target_files": [
|
186 |
+
os.path.join(
|
187 |
+
data_dir,
|
188 |
+
f"IWSLT17.TED.dev2010.{self.config.pair}.{target}.xml",
|
189 |
+
)
|
190 |
+
],
|
191 |
+
},
|
192 |
+
),
|
193 |
+
]
|
194 |
+
|
195 |
+
def _generate_examples(self, split, source_files, target_files):
|
196 |
+
"""Yields examples."""
|
197 |
+
id_ = 0
|
198 |
+
source, target = self.config.pair.split("-")
|
199 |
+
doc_id = 0
|
200 |
+
seg_id = 0
|
201 |
+
for source_file, target_file in zip(source_files, target_files):
|
202 |
+
with open(source_file, "r", encoding="utf-8") as sf:
|
203 |
+
with open(target_file, "r", encoding="utf-8") as tf:
|
204 |
+
if split in ["validation", "test"]:
|
205 |
+
for source_row, target_row in zip(sf, tf):
|
206 |
+
source_row = source_row.strip()
|
207 |
+
target_row = target_row.strip()
|
208 |
+
|
209 |
+
if source_row.startswith("<doc"):
|
210 |
+
doc_id = int(re.findall(DOCID_PATTERN, source_row)[0])
|
211 |
+
elif source_row.startswith("<seg"):
|
212 |
+
seg_id_source, source_seg = re.findall(SEG_PATTERN, source_row)[0]
|
213 |
+
_, target_seg = re.findall(SEG_PATTERN, target_row)[0]
|
214 |
+
seg_id = int(seg_id_source)
|
215 |
+
|
216 |
+
source_row = source_seg.strip()
|
217 |
+
target_row = target_seg.strip()
|
218 |
+
else:
|
219 |
+
continue
|
220 |
+
|
221 |
+
yield id_, {
|
222 |
+
"doc_id": doc_id,
|
223 |
+
"seg_id": seg_id,
|
224 |
+
"translation": {source: source_row, target: target_row}
|
225 |
+
}
|
226 |
+
id_ += 1
|
227 |
+
else: # train
|
228 |
+
for source_row, target_row in zip(sf, tf):
|
229 |
+
source_row = source_row.strip()
|
230 |
+
target_row = target_row.strip()
|
231 |
+
|
232 |
+
if source_row.startswith("<"):
|
233 |
+
if source_row.startswith("<doc"):
|
234 |
+
doc_id = int(re.findall(DOCID_PATTERN, source_row)[0])
|
235 |
+
seg_id = 0
|
236 |
+
else:
|
237 |
+
continue
|
238 |
+
yield id_, {
|
239 |
+
"doc_id": doc_id,
|
240 |
+
"seg_id": seg_id,
|
241 |
+
"translation": {source: source_row, target: target_row}
|
242 |
+
}
|
243 |
+
id_ += 1
|
244 |
+
seg_id += 1
|
245 |
+
|
246 |
+
|
prepare_corpus.py
ADDED
@@ -0,0 +1,124 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import argparse
|
2 |
+
import os
|
3 |
+
import xmltodict
|
4 |
+
import glob
|
5 |
+
|
6 |
+
|
7 |
+
def extract_train(train_inp, train_out, docids_out=None):
|
8 |
+
with open(train_inp, "r", encoding="utf-8") as f:
|
9 |
+
all_docs = f.read()
|
10 |
+
docs_xml = [f"{d}</doc>" for d in all_docs.split("</doc>")[:-1]]
|
11 |
+
docs = [xmltodict.parse(doc) for doc in docs_xml]
|
12 |
+
|
13 |
+
train_out_f = open(train_out, "w", encoding="utf-8")
|
14 |
+
if docids_out is not None:
|
15 |
+
docids_out_f = open(docids_out, "w", encoding="utf-8")
|
16 |
+
|
17 |
+
for doc in docs:
|
18 |
+
if "#text" not in doc["doc"]:
|
19 |
+
continue
|
20 |
+
doc_id = doc["doc"]["@docid"]
|
21 |
+
for line in doc["doc"]["#text"].split("\n"):
|
22 |
+
print(line.strip(), file=train_out_f)
|
23 |
+
if docids_out is not None:
|
24 |
+
print(doc_id, file=docids_out_f)
|
25 |
+
|
26 |
+
|
27 |
+
def extract_eval(eval_inp, eval_out, docids_out=None):
|
28 |
+
with open(eval_inp, "r", encoding="utf-8") as f:
|
29 |
+
all_docs = f.read()
|
30 |
+
docs_xml = xmltodict.parse(all_docs)
|
31 |
+
docs = (
|
32 |
+
docs_xml["mteval"]["srcset"]["doc"]
|
33 |
+
if "srcset" in docs_xml["mteval"]
|
34 |
+
else docs_xml["mteval"]["refset"]["doc"]
|
35 |
+
)
|
36 |
+
|
37 |
+
eval_out_f = open(eval_out, "a")
|
38 |
+
if docids_out is not None:
|
39 |
+
docids_out_f = open(docids_out, "a")
|
40 |
+
|
41 |
+
for doc in docs:
|
42 |
+
doc_id = doc["@docid"]
|
43 |
+
if not doc["seg"]:
|
44 |
+
continue
|
45 |
+
for seg in doc["seg"]:
|
46 |
+
line = seg["#text"]
|
47 |
+
print(line.strip(), file=eval_out_f)
|
48 |
+
if docids_out is not None:
|
49 |
+
print(doc_id, file=docids_out_f)
|
50 |
+
|
51 |
+
|
52 |
+
if __name__ == "__main__":
|
53 |
+
parser = argparse.ArgumentParser("parser for iwslt2017 data")
|
54 |
+
|
55 |
+
parser.add_argument("raw_data")
|
56 |
+
parser.add_argument("out_data")
|
57 |
+
parser.add_argument("-s", "--source-lang", required=True, type=str)
|
58 |
+
parser.add_argument("-t", "--target-lang", required=True, type=str)
|
59 |
+
parser.add_argument(
|
60 |
+
"--devsets", nargs="*", default=["tst2011", "tst2012", "tst2013", "tst2014"]
|
61 |
+
)
|
62 |
+
parser.add_argument("--testsets", nargs="*", default=["tst2015"])
|
63 |
+
args = parser.parse_args()
|
64 |
+
|
65 |
+
first_l = (
|
66 |
+
args.source_lang
|
67 |
+
if glob.glob(f"{args.raw_data}/*.{args.source_lang}-{args.target_lang}.*")
|
68 |
+
else args.target_lang
|
69 |
+
)
|
70 |
+
second_l = (
|
71 |
+
args.target_lang
|
72 |
+
if glob.glob(f"{args.raw_data}/*.{args.source_lang}-{args.target_lang}.*")
|
73 |
+
else args.source_lang
|
74 |
+
)
|
75 |
+
|
76 |
+
train_inp_prefix = os.path.join(args.raw_data, f"train.tags.{first_l}-{second_l}")
|
77 |
+
train_out_prefix = os.path.join(
|
78 |
+
args.out_data, f"train.{args.source_lang}-{args.target_lang}"
|
79 |
+
)
|
80 |
+
extract_train(
|
81 |
+
f"{train_inp_prefix}.{args.source_lang}",
|
82 |
+
f"{train_out_prefix}.{args.source_lang}",
|
83 |
+
docids_out=f"{train_out_prefix}.docids",
|
84 |
+
)
|
85 |
+
extract_train(
|
86 |
+
f"{train_inp_prefix}.{args.target_lang}",
|
87 |
+
f"{train_out_prefix}.{args.target_lang}",
|
88 |
+
)
|
89 |
+
|
90 |
+
# generate valid set based on devsets paseed
|
91 |
+
valid_out_prefix = os.path.join(
|
92 |
+
args.out_data, f"valid.{args.source_lang}-{args.target_lang}"
|
93 |
+
)
|
94 |
+
for devset in args.devsets:
|
95 |
+
valid_inp_prefix = os.path.join(
|
96 |
+
args.raw_data, f"IWSLT17.TED.{devset}.{first_l}-{second_l}"
|
97 |
+
)
|
98 |
+
extract_eval(
|
99 |
+
f"{valid_inp_prefix}.{args.source_lang}.xml",
|
100 |
+
f"{valid_out_prefix}.{args.source_lang}",
|
101 |
+
docids_out=f"{valid_out_prefix}.docids",
|
102 |
+
)
|
103 |
+
extract_eval(
|
104 |
+
f"{valid_inp_prefix}.{args.target_lang}.xml",
|
105 |
+
f"{valid_out_prefix}.{args.target_lang}",
|
106 |
+
)
|
107 |
+
|
108 |
+
# generate test set based on testsets paseed
|
109 |
+
test_out_prefix = os.path.join(
|
110 |
+
args.out_data, f"test.{args.source_lang}-{args.target_lang}"
|
111 |
+
)
|
112 |
+
for testset in args.testsets:
|
113 |
+
test_inp_prefix = os.path.join(
|
114 |
+
args.raw_data, f"IWSLT17.TED.{testset}.{first_l}-{second_l}"
|
115 |
+
)
|
116 |
+
extract_eval(
|
117 |
+
f"{test_inp_prefix}.{args.source_lang}.xml",
|
118 |
+
f"{test_out_prefix}.{args.source_lang}",
|
119 |
+
docids_out=f"{test_out_prefix}.docids",
|
120 |
+
)
|
121 |
+
extract_eval(
|
122 |
+
f"{test_inp_prefix}.{args.target_lang}.xml",
|
123 |
+
f"{test_out_prefix}.{args.target_lang}",
|
124 |
+
)
|