Sebastian Gehrmann commited on
Commit
337c8dc
1 Parent(s): 5e50336

detokenize target

Browse files
Files changed (2) hide show
  1. RotoWire_English-German.py +19 -2
  2. dataset_infos.json +1 -1
RotoWire_English-German.py CHANGED
@@ -56,6 +56,22 @@ player_line = "<PLAYER> %s <TEAM> %s <POS> %s <RANK> %s <MIN> %d <PTS> %d <FG> %
56
  team_line = "%s <TEAM> %s <CITY> %s <TEAM-RESULT> %s <TEAM-PTS> %d <WINS-LOSSES> %d %d <QTRS> %d %d %d %d " \
57
  "<TEAM-AST> %d <3PT> %d <TEAM-FG> %d <TEAM-FT> %d <TEAM-REB> %d <TEAM-TO> %d"
58
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
59
  class RotowireEnglishGerman(datasets.GeneratorBasedBuilder):
60
  """Dataset for WNGT2019 shared task on Document-level Generation and Translation."""
61
 
@@ -478,7 +494,8 @@ class RotowireEnglishGerman(datasets.GeneratorBasedBuilder):
478
  with open(filepath, encoding="utf-8") as f:
479
  all_data = json.load(f)
480
  for id_, data in enumerate(all_data):
 
481
  data['linearized_input'] = self.linearize_input(data)
482
- data['target'] = data['summary_de'][0]
483
- data['references'] = data['summary_de']
484
  yield id_, data
56
  team_line = "%s <TEAM> %s <CITY> %s <TEAM-RESULT> %s <TEAM-PTS> %d <WINS-LOSSES> %d %d <QTRS> %d %d %d %d " \
57
  "<TEAM-AST> %d <3PT> %d <TEAM-FG> %d <TEAM-FT> %d <TEAM-REB> %d <TEAM-TO> %d"
58
 
59
+ def detokenize(text):
60
+ """
61
+ Untokenizing a text undoes the tokenizing operation, restoring
62
+ punctuation and spaces to the places that people expect them to be.
63
+ Ideally, `untokenize(tokenize(text))` should be identical to `text`,
64
+ except for line breaks.
65
+ """
66
+ step1 = text.replace("`` ", '"').replace(" ''", '"').replace('. . .', '...')
67
+ step2 = step1.replace(" ( ", " (").replace(" ) ", ") ")
68
+ step3 = re.sub(r' ([.,:;?!%]+)([ \'"`])', r"\1\2", step2)
69
+ step4 = re.sub(r' ([.,:;?!%]+)$', r"\1", step3)
70
+ step5 = step4.replace(" '", "'").replace(" n't", "n't").replace(
71
+ "can not", "cannot").replace(" 've", "'ve")
72
+ step6 = step5.replace(" ` ", " '")
73
+ return step6.strip()
74
+
75
  class RotowireEnglishGerman(datasets.GeneratorBasedBuilder):
76
  """Dataset for WNGT2019 shared task on Document-level Generation and Translation."""
77
 
494
  with open(filepath, encoding="utf-8") as f:
495
  all_data = json.load(f)
496
  for id_, data in enumerate(all_data):
497
+ detok_target = detokenize(" ".join(data['summary_de']))
498
  data['linearized_input'] = self.linearize_input(data)
499
+ data['target'] = detok_target
500
+ data['references'] = [detok_target]
501
  yield id_, data
dataset_infos.json CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ecd0484e5d6450c3f5532d9b5821e18c5c15df516d21bebf3708f74c18a4f476
3
  size 40690
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f4a8df1f5677f21148d65ab7b5fb40074f3e5c0fb52bffbf3ebcc551ba5d66ad
3
  size 40690