afriqa-gold-passages.py

"""AfriQA GOLD Passages dataset."""


import json
import os
from textwrap import dedent

import datasets


_HOMEPAGE = "https://github.com/masakhane-io/afriqa"

_DESCRIPTION = """\
AfriQA: Cross-lingual Open-Retrieval Question Answering for African Languages
AfriQA is the first cross-lingual question-answering (QA) dataset with a focus on African languages. 
The dataset includes over 12,000 XOR QA examples across 10 African languages, making it an invaluable resource for developing more equitable QA technology.
"""

_CITATION = """\
"""

_URL = "https://github.com/masakhane-io/afriqa/raw/main/data/gold_passages/"

_LANG_2_PIVOT = {
    "bem": "en",
    "fon": "fr",
    "hau": "en",
    "ibo": "en",
    "kin": "en",
    "swa": "en",
    "twi": "en",
    "yor": "en",
    "zul": "en",
}

_LANG_2_SPLITS = {
    "bem": ["train", "dev", "test"],
    "fon": ["train", "dev", "test"],
    "hau": ["train", "dev", "test"],
    "ibo": ["train", "dev", "test"],
    "kin": ["train", "dev", "test"],
    "swa": ["test"],
    "twi": ["train", "dev", "test"],
    "yor": ["train", "test"],
    "zul": ["train", "dev", "test"],

}

class AfriQAConfig(datasets.BuilderConfig):
    """BuilderConfig for AfriQA"""

    def __init__(self, **kwargs):
        """BuilderConfig for AfriQA.
        Args:
          **kwargs: keyword arguments forwarded to super.
        """
        super(AfriQAConfig, self).__init__(**kwargs)


class AfriQA(datasets.GeneratorBasedBuilder):
    """AfriQA dataset."""

    VERSION = datasets.Version("1.0.0")

    BUILDER_CONFIGS = [
        AfriQAConfig(name="bem", version=datasets.Version("1.0.0"), description="AfriQA Gold Passages Bemba dataset"),
        AfriQAConfig(name="fon", version=datasets.Version("1.0.0"), description="AfriQA Gold Passages Fon dataset"),
        AfriQAConfig(name="hau", version=datasets.Version("1.0.0"), description="AfriQA Gold Passages Hausa dataset"),
        AfriQAConfig(name="ibo", version=datasets.Version("1.0.0"), description="AfriQA Gold Passages Igbo dataset"),
        AfriQAConfig(name="kin", version=datasets.Version("1.0.0"), description="AfriQA Gold Passages Kinyarwanda dataset"),
        AfriQAConfig(name="swa", version=datasets.Version("1.0.0"), description="AfriQA Gold Passages Swahili dataset"),
        AfriQAConfig(name="twi", version=datasets.Version("1.0.0"), description="AfriQA Gold Passages Twi dataset"),
        AfriQAConfig(name="wol", version=datasets.Version("1.0.0"), description="AfriQA Gold Passages Wolof dataset"),
        AfriQAConfig(name="yor", version=datasets.Version("1.0.0"), description="AfriQA Gold Passages Yoruba dataset"),
        AfriQAConfig(name="zul", version=datasets.Version("1.0.0"), description="AfriQA Gold Passages Zulu dataset"),
    ]

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "question_lang": datasets.Value("string"),
                    "question_translated": datasets.Value("string"),
                    "context": datasets.Value("string"),
                    "title": datasets.Value("string"),
                    "answer_pivot": datasets.Value("string"),
                    "answer_start": datasets.Value("string"),
                    "answer_lang": datasets.Value("string"),
                }
            ),
            homepage=_HOMEPAGE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        """Returns SplitGenerators."""

        urls_to_download = {}

        for split in _LANG_2_SPLITS[self.config.name]:
            urls_to_download[split] = f"{_URL}{self.config.name}/gold_span_passages.afriqa.{self.config.name}.{_LANG_2_PIVOT[self.config.name]}.{split}.json"

        downloaded_files = dl_manager.download_and_extract(urls_to_download)

        splits_list = []

        for split in _LANG_2_SPLITS[self.config.name]:
            
            if split == "train":
                splits_list.append(datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": downloaded_files["train"]}))
            elif split == "dev":
                splits_list.append(datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": downloaded_files["dev"]}))
            elif split == "test":
                splits_list.append(datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": downloaded_files["test"]}))
        

        return splits_list

    def _generate_examples(self, filepath):
        """Yields examples."""
        with open(filepath, encoding="utf-8-sig") as f:
            for _, row in enumerate(f):
                example = json.loads(row)
                _id = example["id"]

                if not example["context"] or not  example["answer_pivot"]["answer_start"]:
                    continue

                yield _id, {
                    "question_lang": example["question_lang"],
                    "question_translated": example["question_translated"],
                    "context": example["context"],
                    "title": example["title"],
                    "answer_pivot": example["answer_pivot"]["text"][0],
                    "answer_start": example["answer_pivot"]["answer_start"][0],
                    "answer_lang": example["answer_lang"],
                }