Datasets:
Tasks:
Text Classification
Sub-tasks:
semantic-similarity-classification
Multilinguality:
multilingual
Size Categories:
10K<n<100K
Language Creators:
found
Annotations Creators:
expert-generated
Source Datasets:
original
Tags:
License:
| from dataclasses import dataclass | |
| import datasets | |
| from datasets.info import DatasetInfo | |
| from datasets.utils.download_manager import DownloadManager | |
| import os | |
| _DESCRIPTION = """A system's task on any of the XL-WiC datasets is to identify the intended meaning of a word in a context of a given language. XL-WiC is framed as a binary classification task. Each instance in XL-WiC has a target word w, either a verb or a noun, for which two contexts are provided. Each of these contexts triggers a specific meaning of w. The task is to identify if the occurrences of w in the two contexts correspond to the same meaning or not. | |
| XL-WiC provides dev and test sets in the following 12 languages: | |
| Bulgarian (BG) | |
| Danish (DA) | |
| German (DE) | |
| Estonian (ET) | |
| Farsi (FA) | |
| French (FR) | |
| Croatian (HR) | |
| Italian (IT) | |
| Japanese (JA) | |
| Korean (KO) | |
| Dutch (NL) | |
| Chinese (ZH) | |
| and training sets in the following 3 languages: | |
| German (DE) | |
| French (FR) | |
| Italian (IT) | |
| """ | |
| _CITATION = """@inproceedings{raganato-etal-2020-xl-wic, | |
| title={XL-WiC: A Multilingual Benchmark for Evaluating Semantic Contextualization}, | |
| author={Raganato, Alessandro and Pasini, Tommaso and Camacho-Collados, Jose and Pilehvar, Mohammad Taher}, | |
| booktitle={Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)}, | |
| pages={7193--7206}, | |
| year={2020} | |
| } | |
| """ | |
| _DOWNLOAD_URL = "https://pilehvar.github.io/xlwic/data/xlwic_datasets.zip" | |
| _VERSION = "1.0.0" | |
| _WN_LANGS = ["EN", "BG", "ZH", "HR", "DA", "NL", "ET", "FA", "JA", "KO"] | |
| _WIKT_LANGS = ["IT", "FR", "DE"] | |
| _CODE_TO_LANG_ID = { | |
| "EN": "english", | |
| "BG": "bulgarian", | |
| "ZH": "chinese", | |
| "HR": "croatian", | |
| "DA": "danish", | |
| "NL": "dutch", | |
| "ET": "estonian", | |
| "FA": "farsi", | |
| "JA": "japanese", | |
| "KO": "korean", | |
| "IT": "italian", | |
| "FR": "french", | |
| "DE": "german", | |
| } | |
| _AVAILABLE_PAIRS = ( | |
| list(zip(["EN"] * (len(_WN_LANGS) - 1), _WN_LANGS[1:])) | |
| + list(zip(["EN"] * len(_WIKT_LANGS), _WIKT_LANGS)) | |
| + [("IT", "IT"), ("FR", "FR"), ("DE", "DE")] | |
| ) | |
| class XLWiCConfig(datasets.BuilderConfig): | |
| version:str=None | |
| training_lang:str = None | |
| target_lang:str = None | |
| name:str = None | |
| class XLWIC(datasets.GeneratorBasedBuilder): | |
| BUILDER_CONFIGS = [ | |
| XLWiCConfig( | |
| name=f"xlwic_{source.lower()}_{target.lower()}", | |
| training_lang=source, | |
| target_lang=target, | |
| version=datasets.Version(_VERSION, ""), | |
| ) | |
| for source, target in _AVAILABLE_PAIRS | |
| ] | |
| def _info(self) -> DatasetInfo: | |
| return datasets.DatasetInfo( | |
| description=_DESCRIPTION, | |
| features=datasets.Features( | |
| { | |
| "id": datasets.Value("string"), | |
| "context_1": datasets.Value("string"), | |
| "context_2": datasets.Value("string"), | |
| "target_word": datasets.Value("string"), | |
| "pos": datasets.Value("string"), | |
| "target_word_location_1": | |
| { | |
| "char_start": datasets.Value("int32"), | |
| "char_end": datasets.Value("int32"), | |
| }, | |
| "target_word_location_2": | |
| { | |
| "char_start": datasets.Value("int32"), | |
| "char_end": datasets.Value("int32"), | |
| }, | |
| "language": datasets.Value("string"), | |
| "label": datasets.Value("int32"), | |
| } | |
| ), | |
| supervised_keys=None, | |
| homepage="https://pilehvar.github.io/xlwic/", | |
| citation=_CITATION, | |
| ) | |
| def _split_generators(self, dl_manager: DownloadManager): | |
| downloaded_file = dl_manager.download_and_extract(_DOWNLOAD_URL) | |
| dataset_root_folder = os.path.join(downloaded_file, "xlwic_datasets") | |
| return [ | |
| datasets.SplitGenerator( | |
| name=datasets.Split.TRAIN, | |
| # These kwargs will be passed to _generate_examples | |
| gen_kwargs={ | |
| "dataset_root": dataset_root_folder, | |
| "lang": self.config.training_lang, | |
| "split": "train", | |
| }, | |
| ), | |
| datasets.SplitGenerator( | |
| name=datasets.Split.VALIDATION, | |
| # These kwargs will be passed to _generate_examples | |
| gen_kwargs={ | |
| "dataset_root": dataset_root_folder, | |
| "lang": self.config.target_lang, | |
| "split": "valid", | |
| }, | |
| ), | |
| datasets.SplitGenerator( | |
| name=datasets.Split.TEST, | |
| # These kwargs will be passed to _generate_examples | |
| gen_kwargs={ | |
| "dataset_root": dataset_root_folder, | |
| "lang": self.config.target_lang, | |
| "split": "test", | |
| }, | |
| ), | |
| ] | |
| def _yield_from_lines(self, lines, lang): | |
| for i, ( | |
| tw, | |
| pos, | |
| char_start_1, | |
| char_end_1, | |
| char_start_2, | |
| char_end_2, | |
| context_1, | |
| context_2, | |
| label, | |
| ) in enumerate(lines): | |
| _id = f"{lang}_{i}" | |
| yield _id, { | |
| "id": _id, | |
| "target_word": tw, | |
| "context_1": context_1, | |
| "context_2": context_2, | |
| "label": int(label), | |
| "target_word_location_1": { | |
| "char_start": int(char_start_1), | |
| "char_end": int(char_end_1), | |
| }, | |
| "target_word_location_2": { | |
| "char_start": int(char_start_2), | |
| "char_end": int(char_end_2) | |
| }, | |
| "pos": pos, | |
| "language": lang, | |
| } | |
| def _from_selfcontained_file(self, dataset_root, lang, split): | |
| ext_lang = _CODE_TO_LANG_ID[lang] | |
| if lang in _WIKT_LANGS: | |
| path = os.path.join( | |
| dataset_root, | |
| "xlwic_wikt", | |
| f"{ext_lang}_{lang.lower()}", | |
| f"{lang.lower()}_{split}.txt", | |
| ) | |
| elif lang != "EN" and lang in _WN_LANGS: | |
| path = os.path.join( | |
| dataset_root, | |
| "xlwic_wn", | |
| f"{ext_lang}_{lang.lower()}", | |
| f"{lang.lower()}_{split}.txt", | |
| ) | |
| elif lang == "EN" and lang in _WN_LANGS: | |
| path = os.path.join( | |
| dataset_root, "wic_english", f"{split}_{lang.lower()}.txt" | |
| ) | |
| with open(path) as lines: | |
| all_lines = [line.strip().split("\t") for line in lines] | |
| yield from self._yield_from_lines(all_lines, lang) | |
| def _from_test_files(self, dataset_root, lang, split): | |
| ext_lang = _CODE_TO_LANG_ID[lang] | |
| if lang in _WIKT_LANGS: | |
| path_data = os.path.join( | |
| dataset_root, | |
| "xlwic_wikt", | |
| f"{ext_lang}_{lang.lower()}", | |
| f"{lang.lower()}_{split}_data.txt", | |
| ) | |
| elif lang != "EN" and lang in _WN_LANGS: | |
| path_data = os.path.join( | |
| dataset_root, | |
| "xlwic_wn", | |
| f"{ext_lang}_{lang.lower()}", | |
| f"{lang.lower()}_{split}_data.txt", | |
| ) | |
| path_gold = path_data.replace('_data.txt', '_gold.txt') | |
| with open(path_data) as lines: | |
| all_lines = [line.strip().split("\t") for line in lines] | |
| with open(path_gold) as lines: | |
| all_labels = [line.strip() for line in lines] | |
| for line, label in zip(all_lines, all_labels): | |
| line.append(label) | |
| yield from self._yield_from_lines(all_lines, lang) | |
| def _generate_examples(self, dataset_root, lang, split, **kwargs): | |
| if split in {"train", "valid"}: | |
| yield from self._from_selfcontained_file(dataset_root, lang, split) | |
| else: | |
| yield from self._from_test_files(dataset_root, lang, split) | |