Create sts22-crosslingual-sts.py

sts22-crosslingual-sts.py

@@ -0,0 +1,117 @@
+import json
+import datasets
+import os
+
+_CITATION = """\
+"""
+
+_LICENSE = """\
+"""
+
+_DESCRIPTION = """\
+SemEval 2022 Task 8: Multilingual News Article Similarity
+"""
+
+_LANGUAGES_TRAIN = {
+ "en": "en",
+ "de": "de",
+ "es": "es",
+ "pl": "pl",
+ "tr": "tr",
+ "fr": "fr",
+ "ar": "ar",
+ "de-en": "de-en",
+}
+_LANGUAGES_TEST = {
+ "en": "en",
+ "de": "de",
+ "es": "es",
+ "pl": "pl",
+ "tr": "tr",
+ "ar": "ar",
+ "ru": "ru",
+ "zh": "zh",
+ "fr": "fr",
+ "de-en": "de-en",
+ "es-en": "es-en",
+ "it": "it",
+ "pl-en": "pl-en",
+ "zh-en": "zh-en",
+ "es-it": "es-it",
+ "de-fr": "de-fr",
+ "de-pl": "de-pl",
+ "fr-pl": "fr-pl",
+}
+_LANGUAGES = {**_LANGUAGES_TRAIN, **_LANGUAGES_TEST}
+
+_ALL_LANGUAGES = "all_languages"
+_HOMEPAGE_URL = "https://competitions.codalab.org/competitions/33835"
+_DOWNLOAD_URL = "{lang}/{split}.jsonl"
+_VERSION = "1.0.0"
+
+
+class STS22Task8Config(datasets.BuilderConfig):
+ """BuilderConfig for STS22Task8Config."""
+
+ def __init__(self, languages=None, **kwargs):
+ super(STS22Task8Config, self).__init__(version=datasets.Version(_VERSION, ""), **kwargs),
+ self.languages = languages
+
+
+class AmazonReviewsMulti(datasets.GeneratorBasedBuilder):
+
+ """Multilingual News Article Similarity"""
+
+ BUILDER_CONFIGS = [
+ STS22Task8Config(
+ name=_ALL_LANGUAGES,
+ languages=_LANGUAGES,
+ description="Multilingual News Article Similarity",
+ )
+ ] + [
+ STS22Task8Config(
+ name=lang,
+ languages=[lang],
+ description=f"{_LANGUAGES[lang]} examples from a collection of multilingual news articles",
+ )
+ for lang in _LANGUAGES
+ ]
+ BUILDER_CONFIG_CLASS = STS22Task8Config
+ DEFAULT_CONFIG_NAME = _ALL_LANGUAGES
+
+ def _info(self):
+ return datasets.DatasetInfo(
+ description=_DESCRIPTION,
+ features=datasets.Features(
+ {
+ "id": datasets.Value("string"),
+ "sentence1": datasets.Value("string"),
+ "sentence2": datasets.Value("string"),
+ "score": datasets.Value("float32"),
+ },
+ ),
+ supervised_keys=None,
+ license=_LICENSE,
+ homepage=_HOMEPAGE_URL,
+ citation=_CITATION,
+ )
+
+ def _split_generators(self, dl_manager):
+ train_urls = [_DOWNLOAD_URL.format(split="train", lang=lang) for lang in self.config.languages if lang in _LANGUAGES_TRAIN]
+ test_urls = [_DOWNLOAD_URL.format(split="test", lang=lang) for lang in self.config.languages if lang in _LANGUAGES_TEST]
+
+ train_paths = dl_manager.download_and_extract(train_urls)
+ test_paths = dl_manager.download_and_extract(test_urls)
+
+ return [
+ datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"file_paths": train_paths}),
+ datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"file_paths": test_paths}),
+ ]
+
+ def _generate_examples(self, file_paths):
+ row_count = 0
+ for file_path in file_paths:
+ with open(file_path, "r", encoding="utf-8") as f:
+ for line in f:
+ yield row_count, json.loads(line)
+ row_count += 1