mathiascreutz commited on
Commit
d6a4efb
1 Parent(s): 04ebbb9

Minor modifications

Browse files
Files changed (1) hide show
  1. README.md +43 -13
README.md CHANGED
@@ -123,8 +123,8 @@ data = load_dataset("GEM/opusparcus", lang="de")
123
 
124
  The above command will download the validation and test sets for
125
  German. If additionally, you want to retrieve training data, you need
126
- to specify the level of quality you desire, such as "90% quality training data of
127
- French":
128
 
129
  ```
130
  data = load_dataset("GEM/opusparcus", lang="fr", quality=90)
@@ -283,55 +283,85 @@ and largest (`quality=60`) train configuration have been shown.
283
 
284
  ### Curation Rationale
285
 
286
- TBA
 
 
287
 
288
  ### Source Data
289
 
290
  #### Initial Data Collection and Normalization
291
 
292
- TBA
 
 
 
 
293
 
294
  #### Who are the source language producers?
295
 
296
- TBA
 
297
 
298
  ### Annotations
299
 
300
  #### Annotation process
301
 
302
- TBA
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
303
 
304
  #### Who are the annotators?
305
 
306
- TBA
 
307
 
308
  ### Personal and Sensitive Information
309
 
310
- [More Information Needed]
311
 
312
  ## Considerations for Using the Data
313
 
314
  ### Social Impact of Dataset
315
 
316
- [More Information Needed]
317
 
318
  ### Discussion of Biases
319
 
320
- [More Information Needed]
 
321
 
322
  ### Other Known Limitations
323
 
324
- [More Information Needed]
 
 
 
 
 
 
325
 
326
  ## Additional Information
327
 
328
  ### Dataset Curators
329
 
330
- TBA
331
 
332
  ### Licensing Information
333
 
334
- [More Information Needed]
335
 
336
  ### Citation Information
337
 
 
123
 
124
  The above command will download the validation and test sets for
125
  German. If additionally, you want to retrieve training data, you need
126
+ to specify the level of quality you desire, such as "French, with 90%
127
+ quality of the training data":
128
 
129
  ```
130
  data = load_dataset("GEM/opusparcus", lang="fr", quality=90)
 
283
 
284
  ### Curation Rationale
285
 
286
+ Opusparcus was created in order to produce a *sentential* paraphrase corpus
287
+ for multiple languages containing *colloquial* language (as opposed to
288
+ news or religious text, for instance).
289
 
290
  ### Source Data
291
 
292
  #### Initial Data Collection and Normalization
293
 
294
+ The data in Opusparcus has been extracted from
295
+ [OpenSubtitles2016](http://opus.nlpl.eu/OpenSubtitles2016.php), which
296
+ is in turn based on data from http://www.opensubtitles.org/.
297
+
298
+ The sentences have been tokenized.
299
 
300
  #### Who are the source language producers?
301
 
302
+ The texts consist of subtitles that have been produced using
303
+ crowdsourcing.
304
 
305
  ### Annotations
306
 
307
  #### Annotation process
308
 
309
+ The development and test sets consist of sentence
310
+ pairs that have been annotated manually; each set contains
311
+ approximately 1000 sentence pairs that have been verified to be
312
+ acceptable paraphrases by two indepedent annotators.
313
+
314
+ The `annot_score` field reflects the judgments made by the annotators.
315
+ If ´the annnotators fully agreed on the category (4.0: dark green,
316
+ 3.0: light green, 2.0: yellow, 1.0: red), the value of
317
+ `annot_score` is 4.0, 3.0, 2.0 or 1.0. If the two annotators
318
+ chose adjacent categories, the value in this field will be 3.5, 2.5 or
319
+ 1.5. For instance, a value of 2.5 means that one annotator gave a
320
+ score of 3 ("mostly good"), indicating a possible paraphrase pair,
321
+ whereas the other annotator scored this as a 2 ("mostly bad"), that
322
+ is, unlikely to be a paraphrase pair. If the annotators disagreed by
323
+ more than one category, the sentence pair was discarded and won't show
324
+ up in the datasets.
325
 
326
  #### Who are the annotators?
327
 
328
+ Students and staff at the University of Helsinki (native or very
329
+ proficient speakers of the target languages)
330
 
331
  ### Personal and Sensitive Information
332
 
333
+ The datasets do not contain any personal or sensitive information.
334
 
335
  ## Considerations for Using the Data
336
 
337
  ### Social Impact of Dataset
338
 
339
+ The goal of Opusparcus is to promote the support for colloquial language.
340
 
341
  ### Discussion of Biases
342
 
343
+ The data reflect the biases present in the movies and TV shows that
344
+ have been subtitled.
345
 
346
  ### Other Known Limitations
347
 
348
+ The sentence pairs in the validation and test sets have been selected
349
+ in such a manner that their Levenshtein distance (minimum edit
350
+ distance) exceeds a certain theshold. This guarantees that the manual
351
+ annotation effort focuses on "interesting" sentence pairs rather than
352
+ trivial variations (such as "It is good." vs. "It's good."). The
353
+ training sets, however, have not been prefiltered in this manner and
354
+ thus also contain highly similar sentences.
355
 
356
  ## Additional Information
357
 
358
  ### Dataset Curators
359
 
360
+ Mathias Creutz, University of Helsinki, Finland
361
 
362
  ### Licensing Information
363
 
364
+ CC-BY-NC 4.0
365
 
366
  ### Citation Information
367