# Some code referenced from https://huggingface.co/datasets/xnli/blob/main/xnli.py """ The Cross-lingual Natural Language Inference (XNLI) corpus is a crowd-sourced collection of 5,000 test and 2,500 dev pairs for the MultiNLI corpus. The pairs are annotated with textual entailment and translated into 14 languages: French, Spanish, German, Greek, Bulgarian, Russian, Turkish, Arabic, Vietnamese, Thai, Chinese, Hindi, Swahili and Urdu. This results in 112.5k annotated pairs. Each premise can be associated with the corresponding hypothesis in the 15 languages, summing up to more than 1.5M combinations. """ import csv import os from pathlib import Path from typing import Dict, List, Tuple import datasets from seacrowd.utils import schemas from seacrowd.utils.configs import SEACrowdConfig from seacrowd.utils.constants import Licenses, Tasks _CITATION = """\ @InProceedings{conneau2018xnli, author = "Conneau, Alexis and Rinott, Ruty and Lample, Guillaume and Williams, Adina and Bowman, Samuel R. and Schwenk, Holger and Stoyanov, Veselin", title = "XNLI: Evaluating Cross-lingual Sentence Representations", booktitle = "Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing", year = "2018", publisher = "Association for Computational Linguistics", location = "Brussels, Belgium", } """ _DATASETNAME = "xnli" _DESCRIPTION = """\ XNLI is a subset of a few thousand examples from MNLI which has been translated into a 14 different languages (some low-ish resource). As with MNLI, the goal is to predict textual entailment (does sentence A imply/contradict/neither sentence B) and is a classification task (given two sentences, predict one of three labels). """ _HOMEPAGE = "https://github.com/facebookresearch/XNLI" # We follow ISO639-3 language code (https://iso639-3.sil.org/code_tables/639/data) _LANGUAGES = ["tha", "vie"] _LANGUAGE_MAPPER = {"tha": "th", "vie": "vi"} _LICENSE = Licenses.CC_BY_NC_4_0.value _LOCAL = False _URLS = { _DATASETNAME: { "train": "https://dl.fbaipublicfiles.com/XNLI/XNLI-MT-1.0.zip", "test": "https://dl.fbaipublicfiles.com/XNLI/XNLI-1.0.zip", } } _SUPPORTED_TASKS = [Tasks.TEXTUAL_ENTAILMENT] _SOURCE_VERSION = "1.1.0" _SEACROWD_VERSION = "2024.06.20" class XNLIDataset(datasets.GeneratorBasedBuilder): """ XNLI is an evaluation corpus for language transfer and cross-lingual sentence classification in 15 languages. In SeaCrowd, we currently only have Thailand and Vietnam Language. """ SOURCE_VERSION = datasets.Version(_SOURCE_VERSION) SEACROWD_VERSION = datasets.Version(_SEACROWD_VERSION) subsets = ["xnli.tha", "xnli.vie"] BUILDER_CONFIGS = [ SEACrowdConfig( name=f"{sub}_source", version=datasets.Version(_SOURCE_VERSION), description=f"{sub} source schema", schema="source", subset_id=f"{sub}", ) for sub in subsets ] + [ SEACrowdConfig( name=f"{sub}_seacrowd_pairs", version=datasets.Version(_SEACROWD_VERSION), description=f"{sub} SEACrowd schema", schema="seacrowd_pairs", subset_id=f"{sub}", ) for sub in subsets ] DEFAULT_CONFIG_NAME = "xnli.vie_source" labels = ["contradiction", "entailment", "neutral"] def _info(self) -> datasets.DatasetInfo: if self.config.schema == "source": features = datasets.Features( { "premise": datasets.Value("string"), "hypothesis": datasets.Value("string"), "label": datasets.ClassLabel(names=self.labels), } ) elif self.config.schema == "seacrowd_pairs": features = schemas.pairs_features(self.labels) return datasets.DatasetInfo( description=_DESCRIPTION, features=features, homepage=_HOMEPAGE, license=_LICENSE, citation=_CITATION, ) def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]: """Returns SplitGenerators.""" urls = _URLS[_DATASETNAME] data_dir = dl_manager.download_and_extract(urls) xnli_train = os.path.join(data_dir["train"], "XNLI-MT-1.0", "multinli") train_data_path = os.path.join(xnli_train, "multinli.train.{}.tsv") xnli_test = os.path.join(data_dir["test"], "XNLI-1.0") val_data_path = os.path.join(xnli_test, "xnli.dev.tsv") test_data_path = os.path.join(xnli_test, "xnli.test.tsv") lang = self.config.name.split("_")[0].split(".")[-1] return [ datasets.SplitGenerator( name=datasets.Split.TRAIN, gen_kwargs={ "filepath": train_data_path, "split": "train", "language": _LANGUAGE_MAPPER[lang], }, ), datasets.SplitGenerator( name=datasets.Split.VALIDATION, gen_kwargs={ "filepath": val_data_path, "split": "dev", "language": _LANGUAGE_MAPPER[lang], }, ), datasets.SplitGenerator( name=datasets.Split.TEST, gen_kwargs={ "filepath": test_data_path, "split": "test", "language": _LANGUAGE_MAPPER[lang], }, ), ] def _generate_examples(self, filepath: Path, split: str, language: str) -> Tuple[int, Dict]: """Yields examples as (key, example) tuples.""" if self.config.schema == "source": if split == "train": file = open(filepath.format(language), encoding="utf-8") reader = csv.DictReader(file, delimiter="\t", quoting=csv.QUOTE_NONE) for row_idx, row in enumerate(reader): key = str(row_idx) yield key, { "premise": row["premise"], "hypothesis": row["hypo"], "label": row["label"].replace("contradictory", "contradiction"), } else: with open(filepath, encoding="utf-8") as f: reader = csv.DictReader(f, delimiter="\t", quoting=csv.QUOTE_NONE) for row in reader: if row["language"] == language: yield row["pairID"], { "premise": row["sentence1"], "hypothesis": row["sentence2"], "label": row["gold_label"], } elif self.config.schema == "seacrowd_pairs": if split == "train": file = open(filepath.format(language), encoding="utf-8") reader = csv.DictReader(file, delimiter="\t", quoting=csv.QUOTE_NONE) for row_idx, row in enumerate(reader): yield str(row_idx), { "id": str(row_idx), "text_1": row["premise"], "text_2": row["hypo"], "label": row["label"].replace("contradictory", "contradiction"), } else: with open(filepath, encoding="utf-8") as f: reader = csv.DictReader(f, delimiter="\t", quoting=csv.QUOTE_NONE) skip = set() for row in reader: if row["language"] == language: if row["pairID"] in skip: continue skip.add(row["pairID"]) yield row["pairID"], { "id": row["pairID"], "text_1": row["sentence1"], "text_2": row["sentence2"], "label": row["gold_label"], } else: raise ValueError(f"Invalid config: {self.config.name}")