Datasets:

ArXiv:
License:
Files changed (1) hide show
  1. IN22-Conv.py +29 -36
IN22-Conv.py CHANGED
@@ -1,5 +1,5 @@
1
  # coding=utf-8
2
- """The IN-22 Conv Evaluation Benchmark for evaluation of Machine Translation for Indic Languages."""
3
 
4
  import os
5
  import sys
@@ -19,8 +19,7 @@ _CITATION = """
19
 
20
  _DESCRIPTION = """\
21
  IN-22 is a newly created comprehensive benchmark for evaluating machine translation performance in multi-domain, n-way parallel contexts across 22 Indic languages.
22
- IN22-Conv is the conversation domain subset of IN22. It is designed to assess translation quality in typical day-to-day conversational-style applications.
23
- Currently, we use it for sentence-level evaluation of MT systems but can be repurposed for document translation evaluation as well.
24
  """
25
 
26
  _HOMEPAGE = "https://github.com/AI4Bharat/IndicTrans2"
@@ -40,7 +39,7 @@ _LANGUAGES = [
40
 
41
  _URL = "https://indictrans2-public.objectstore.e2enetworks.net/IN22_benchmark.tar.gz"
42
 
43
- _SPLITS = ["conv"]
44
 
45
  _SENTENCES_PATHS = {
46
  lang: {
@@ -64,8 +63,8 @@ def _pairings(iterable, r=2):
64
  yield p
65
 
66
 
67
- class IN22ConvConfig(datasets.BuilderConfig):
68
- """BuilderConfig for the IN-22 Conv evaluation subset."""
69
  def __init__(self, lang: str, lang2: str = None, **kwargs):
70
  """
71
  Args:
@@ -76,24 +75,24 @@ class IN22ConvConfig(datasets.BuilderConfig):
76
  self.lang2 = lang2
77
 
78
 
79
- class IN22Conv(datasets.GeneratorBasedBuilder):
80
- """IN-22 Conv evaluation subset."""
81
 
82
  BUILDER_CONFIGS = [
83
- IN22ConvConfig(
84
  name=lang,
85
  description=f"IN-22: {lang} subset.",
86
  lang=lang
87
  )
88
  for lang in _LANGUAGES
89
  ] + [
90
- IN22ConvConfig(
91
  name="all",
92
  description=f"IN-22: all language pairs",
93
  lang=None
94
  )
95
  ] + [
96
- IN22ConvConfig(
97
  name=f"{l1}-{l2}",
98
  description=f"IN-22: {l1}-{l2} aligned subset.",
99
  lang=l1,
@@ -104,14 +103,12 @@ class IN22Conv(datasets.GeneratorBasedBuilder):
104
  def _info(self):
105
  features = {
106
  "id": datasets.Value("int32"),
107
- "doc_id": datasets.Value("int32"),
108
- "sent_id": datasets.Value("int32"),
109
- "topic": datasets.Value("string"),
110
  "domain": datasets.Value("string"),
111
- "prompt": datasets.Value("string"),
112
- "scenario": datasets.Value("string"),
113
- "speaker": datasets.Value("int32"),
114
- "turn": datasets.Value("int32")
115
  }
116
  if self.config.name != "all" and "-" not in self.config.name:
117
  features["sentence"] = datasets.Value("string")
@@ -153,8 +150,8 @@ class IN22Conv(datasets.GeneratorBasedBuilder):
153
  def _generate_examples(self, sentence_paths: Union[str, List[str]], metadata_path: str, langs: Optional[List[str]] = None):
154
  """Yields examples as (key, example) tuples."""
155
  if isinstance(sentence_paths, str):
156
- with open(sentence_paths, "r") as sentences_file:
157
- with open(metadata_path, "r") as metadata_file:
158
  metadata_lines = [l.strip() for l in metadata_file.readlines()[1:]]
159
  for id_, (sentence, metadata) in enumerate(
160
  zip(sentences_file, metadata_lines)
@@ -164,14 +161,12 @@ class IN22Conv(datasets.GeneratorBasedBuilder):
164
  yield id_, {
165
  "id": id_ + 1,
166
  "sentence": sentence,
167
- "doc_id": metadata[0],
168
- "sent_id": metadata[1],
169
- "topic": metadata[2],
170
  "domain": metadata[3],
171
- "prompt": metadata[4],
172
- "scenario": metadata[5],
173
- "speaker": metadata[6],
174
- "turn": metadata[7]
175
  }
176
  else:
177
  sentences = {}
@@ -180,23 +175,21 @@ class IN22Conv(datasets.GeneratorBasedBuilder):
180
  else:
181
  langs = [self.config.lang, self.config.lang2]
182
  for path, lang in zip(sentence_paths, langs):
183
- with open(path, "r") as sent_file:
184
  sentences[lang] = [l.strip() for l in sent_file.readlines()]
185
- with open(metadata_path, "r") as metadata_file:
186
  metadata_lines = [l.strip() for l in metadata_file.readlines()[1:]]
187
  for id_, metadata in enumerate(metadata_lines):
188
  metadata = metadata.split("\t")
189
  yield id_, {
190
  **{
191
  "id": id_ + 1,
192
- "doc_id": metadata[0],
193
- "sent_id": metadata[1],
194
- "topic": metadata[2],
195
  "domain": metadata[3],
196
- "prompt": metadata[4],
197
- "scenario": metadata[5],
198
- "speaker": metadata[6],
199
- "turn": metadata[7]
200
  }, **{
201
  f"sentence_{lang}": sentences[lang][id_]
202
  for lang in langs
 
1
  # coding=utf-8
2
+ """The IN-22 Gen Evaluation Benchmark for evaluation of Machine Translation for Indic Languages."""
3
 
4
  import os
5
  import sys
 
19
 
20
  _DESCRIPTION = """\
21
  IN-22 is a newly created comprehensive benchmark for evaluating machine translation performance in multi-domain, n-way parallel contexts across 22 Indic languages.
22
+ IN22-Gen is a general-purpose multi-domain evaluation subset of IN22. It has been created from two sources: Wikipedia and Web Sources offering diverse content spanning news, entertainment, culture, legal, and India-centric topics.
 
23
  """
24
 
25
  _HOMEPAGE = "https://github.com/AI4Bharat/IndicTrans2"
 
39
 
40
  _URL = "https://indictrans2-public.objectstore.e2enetworks.net/IN22_benchmark.tar.gz"
41
 
42
+ _SPLITS = ["gen"]
43
 
44
  _SENTENCES_PATHS = {
45
  lang: {
 
63
  yield p
64
 
65
 
66
+ class IN22GenConfig(datasets.BuilderConfig):
67
+ """BuilderConfig for the IN-22 Gen evaluation subset."""
68
  def __init__(self, lang: str, lang2: str = None, **kwargs):
69
  """
70
  Args:
 
75
  self.lang2 = lang2
76
 
77
 
78
+ class IN22Gen(datasets.GeneratorBasedBuilder):
79
+ """IN-22 Gen evaluation subset."""
80
 
81
  BUILDER_CONFIGS = [
82
+ IN22GenConfig(
83
  name=lang,
84
  description=f"IN-22: {lang} subset.",
85
  lang=lang
86
  )
87
  for lang in _LANGUAGES
88
  ] + [
89
+ IN22GenConfig(
90
  name="all",
91
  description=f"IN-22: all language pairs",
92
  lang=None
93
  )
94
  ] + [
95
+ IN22GenConfig(
96
  name=f"{l1}-{l2}",
97
  description=f"IN-22: {l1}-{l2} aligned subset.",
98
  lang=l1,
 
103
  def _info(self):
104
  features = {
105
  "id": datasets.Value("int32"),
106
+ "context": datasets.Value("string"),
107
+ "source": datasets.Value("string"),
108
+ "url": datasets.Value("string"),
109
  "domain": datasets.Value("string"),
110
+ "num_words": datasets.Value("int32"),
111
+ "bucket": datasets.Value("string")
 
 
112
  }
113
  if self.config.name != "all" and "-" not in self.config.name:
114
  features["sentence"] = datasets.Value("string")
 
150
  def _generate_examples(self, sentence_paths: Union[str, List[str]], metadata_path: str, langs: Optional[List[str]] = None):
151
  """Yields examples as (key, example) tuples."""
152
  if isinstance(sentence_paths, str):
153
+ with open(sentence_paths, "r",encoding="utf-8") as sentences_file:
154
+ with open(metadata_path, "r",encoding="utf-8") as metadata_file:
155
  metadata_lines = [l.strip() for l in metadata_file.readlines()[1:]]
156
  for id_, (sentence, metadata) in enumerate(
157
  zip(sentences_file, metadata_lines)
 
161
  yield id_, {
162
  "id": id_ + 1,
163
  "sentence": sentence,
164
+ "context": metadata[0],
165
+ "source": metadata[1],
166
+ "url": metadata[2],
167
  "domain": metadata[3],
168
+ "num_words": metadata[4],
169
+ "bucket": metadata[5]
 
 
170
  }
171
  else:
172
  sentences = {}
 
175
  else:
176
  langs = [self.config.lang, self.config.lang2]
177
  for path, lang in zip(sentence_paths, langs):
178
+ with open(path, "r",encoding="utf-8") as sent_file:
179
  sentences[lang] = [l.strip() for l in sent_file.readlines()]
180
+ with open(metadata_path, "r",encoding="utf-8") as metadata_file:
181
  metadata_lines = [l.strip() for l in metadata_file.readlines()[1:]]
182
  for id_, metadata in enumerate(metadata_lines):
183
  metadata = metadata.split("\t")
184
  yield id_, {
185
  **{
186
  "id": id_ + 1,
187
+ "context": metadata[0],
188
+ "source": metadata[1],
189
+ "url": metadata[2],
190
  "domain": metadata[3],
191
+ "num_words": metadata[4],
192
+ "bucket": metadata[5]
 
 
193
  }, **{
194
  f"sentence_{lang}": sentences[lang][id_]
195
  for lang in langs