from pathlib import Path from typing import List import datasets from seacrowd.utils import schemas from seacrowd.utils.configs import SEACrowdConfig from seacrowd.utils.constants import (DEFAULT_SEACROWD_VIEW_NAME, DEFAULT_SOURCE_VIEW_NAME, Licenses, Tasks) _DATASETNAME = "wikimatrix" _SOURCE_VIEW_NAME = DEFAULT_SOURCE_VIEW_NAME _UNIFIED_VIEW_NAME = DEFAULT_SEACROWD_VIEW_NAME # ilo min sun are actually not available _LANGUAGES = ["ilo", "min", "jav", "sun", "ceb", "ind", "tgl", "vie"] # We follow ISO639-3 language code (https://iso639-3.sil.org/code_tables/639/data) _LOCAL = False _CITATION = """\ @inproceedings{schwenk-etal-2021-wikimatrix, title = "{W}iki{M}atrix: Mining 135{M} Parallel Sentences in 1620 Language Pairs from {W}ikipedia", author = "Schwenk, Holger and Chaudhary, Vishrav and Sun, Shuo and Gong, Hongyu and Guzm{\'a}n, Francisco", editor = "Merlo, Paola and Tiedemann, Jorg and Tsarfaty, Reut", booktitle = "Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume", month = apr, year = "2021", address = "Online", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2021.eacl-main.115", doi = "10.18653/v1/2021.eacl-main.115", pages = "1351--1361", abstract = "We present an approach based on multilingual sentence embeddings to automatically extract parallel sentences from the content of Wikipedia articles in 96 languages, including several dialects or low-resource languages. We do not limit the extraction process to alignments with English, but we systematically consider all possible language pairs. In total, we are able to extract 135M parallel sentences for 16720 different language pairs, out of which only 34M are aligned with English. This corpus is freely available. To get an indication on the quality of the extracted bitexts, we train neural MT baseline systems on the mined data only for 1886 languages pairs, and evaluate them on the TED corpus, achieving strong BLEU scores for many language pairs. The WikiMatrix bitexts seem to be particularly interesting to train MT systems between distant languages without the need to pivot through English.", } """ _DESCRIPTION = """\ WikiMatrix is automatically extracted parallel sentences from the content of Wikipedia articles in 96 languages, including several dialects or low-resource languages. 8 languages among them are spoken in Southeast Asia region. In total, there are 135M parallel sentences from 1620 different language pairs. """ _HOMEPAGE = "https://github.com/facebookresearch/LASER/tree/main/tasks/WikiMatrix" _LICENSE = Licenses.CC_BY_SA_4_0.value _URLs = "https://dl.fbaipublicfiles.com/laser/WikiMatrix/v1/WikiMatrix.{lang1}-{lang2}.tsv.gz" _SUPPORTED_TASKS = [Tasks.MACHINE_TRANSLATION] _SOURCE_VERSION = "1.0.0" _SEACROWD_VERSION = "2024.06.20" config = { "jv": ["en", "es", "fr", "id", "it", "pt"], "ceb": ["bg", "ar", "ca", "cs", "de", "en", "es", "fi", "fr", "hu", "it", "ja", "nl", "no", "pl", "pt", "ro", "ru", "sv", "uk"], "id": [ "jv", "is", "it", "ja", "ko", "lt", "mk", "ml", "mr", "ne", "nl", "no", "pl", "pt", "ro", "ru", "sh", "si", "sk", "sl", "sq", "sr", "sv", "sw", "ta", "te", "tl", "tr", "tt", "uk", "vi", "zh", "ar", "az", "ba", "bg", "bn", "bs", "ca", "cs", "da", "de", "el", "en", "eo", "es", "et", "eu", "fa", "fi", "fr", "gl", "he", "hi", "hr", "hu", ], "tl": ["ar", "bg", "bs", "ca", "cs", "da", "de", "el", "en", "eo", "es", "et", "fi", "fr", "gl", "he", "hr", "hu", "id", "it", "ja", "lt", "mk", "nl", "no", "pl", "pt", "ro", "ru", "sh", "sk", "sl", "sq", "sr", "sv", "tr", "uk", "vi", "zh"], "vi": [ "ar", "az", "bg", "bn", "bs", "ca", "cs", "da", "de", "el", "en", "eo", "es", "et", "eu", "fa", "fi", "fr", "gl", "he", "hi", "hr", "hu", "id", "is", "it", "ja", "ko", "lt", "mk", "ml", "mr", "nl", "no", "pl", "pt", "ro", "ru", "sh", "si", "sk", "sl", "sq", "sr", "sv", "sw", "ta", "te", "tl", "tr", "uk", "zh", ], } _SUBSETS = set() for lang, pairs in config.items(): for pair in pairs: _SUBSETS.add("{}-{}".format(lang, pair) if lang < pair else "{}-{}".format(pair, lang)) _SUBSETS = list(_SUBSETS) class WikiMatrixDataset(datasets.GeneratorBasedBuilder): """WikiMatrix is automatically extracted parallel sentences from the content of Wikipedia articles in 96 languages, including several dialects or low-resource languages.""" BUILDER_CONFIGS = [ SEACrowdConfig( name=f"wikimatrix_{subset.replace('-', '_')}_source", version=datasets.Version(_SOURCE_VERSION), description="WikiMatrix source schema", schema="source", subset_id=f"wikimatrix_{subset.replace('-', '_')}", ) for subset in _SUBSETS ] + [ SEACrowdConfig( name=f"wikimatrix_{subset.replace('-', '_')}_seacrowd_t2t", version=datasets.Version(_SEACROWD_VERSION), description="WikiMatrix Nusantara schema", schema="seacrowd_t2t", subset_id=f"wikimatrix_{subset.replace('-', '_')}", ) for subset in _SUBSETS ] DEFAULT_CONFIG_NAME = "wikimatrix_en_id_source" def _info(self): if self.config.schema == "source": features = datasets.Features( { "id": datasets.Value("string"), "text_1": datasets.Value("string"), "text_2": datasets.Value("string"), "text_1_name": datasets.Value("string"), "text_2_name": datasets.Value("string"), } ) elif self.config.schema == "seacrowd_t2t": features = schemas.text2text_features return datasets.DatasetInfo( description=_DESCRIPTION, features=features, homepage=_HOMEPAGE, license=_LICENSE, citation=_CITATION, ) def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]: lang1, lang2 = self.config.name.split("_")[1], self.config.name.split("_")[2] filepath = Path(dl_manager.download_and_extract(_URLs.format(lang1=lang1, lang2=lang2))) return [ datasets.SplitGenerator( name=datasets.Split.TEST, gen_kwargs={"filepath": filepath}, ), ] def _generate_examples(self, filepath: Path): with open(filepath, "r") as f: data = f.readlines() lang1, lang2 = self.config.name.split("_")[1], self.config.name.split("_")[2] if self.config.schema == "source": for _id, line in enumerate(data): line = line.strip().split("\t") ex = { "id": str(_id), "text_1": line[1], "text_2": line[2], "text_1_name": lang1, "text_2_name": lang2, } yield _id, ex elif self.config.schema == "seacrowd_t2t": for _id, line in enumerate(data): line = line.strip().split("\t") ex = { "id": str(_id), "text_1": line[1], "text_2": line[2], "text_1_name": lang1, "text_2_name": lang2, } yield _id, ex else: raise ValueError(f"Invalid config: {self.config.name}")