IE-SemParse.py

# coding=utf-8


# Lint as: python3
"""IndicXNLI: The Cross-Lingual NLI Corpus for Indic Languages."""


import os
import json

import pandas as pd

import datasets

from datasets import DownloadManager


_CITATION = """\
@misc{aggarwal2023evaluating,
      title={Evaluating Inter-Bilingual Semantic Parsing for Indian Languages}, 
      author={Divyanshu Aggarwal and Vivek Gupta and Anoop Kunchukuttan},
      year={2023},
      eprint={2304.13005},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}
"""

_DESCRIPTION = """\
    IE-SemParse is an Inter-bilingual Seq2seq Semantic parsing dataset for 11 distinct Indian languages
"""

_LANGUAGES = (
    'hi',
    'bn',
    'mr',
    'as',
    'ta',
    'te',
    'or',
    'ml',
    'pa',
    'gu',
    'kn'
)


_DATASETS = (
    'itop',
    'indic-atis',
    'indic-TOP'
)


_URL = "https://huggingface.co/datasets/Divyanshu/IE-SemParse/resolve/main/"


class IE_SemParseConfig(datasets.BuilderConfig):
    """BuilderConfig for IE-SemParse."""

    def __init__(self, dataset: str, language: str, **kwargs):
        """BuilderConfig for IE-SemParse.

        Args:
        language: One of hi, bn, mr, as, ta, te, or, ml, pa, gu, kn
          **kwargs: keyword arguments forwarded to super.
        """
        super(IE_SemParseConfig, self).__init__(**kwargs)

        self.dataset = dataset
        self.language = language
        self.languages = _LANGUAGES
        self.datasets = _DATASETS

        self._URLS = [os.path.join(
            _URL, "unfiltered_data", dataset, f"{language}.json")]


class IE_SemParse(datasets.GeneratorBasedBuilder):
    """IE-SemParse: Inter-Bilingual Semantic Parsing Dataset for Indic Languages. Version 1.0."""

    VERSION = datasets.Version("1.0.0", "")
    BUILDER_CONFIG_CLASS = IE_SemParseConfig
    BUILDER_CONFIGS = [
        IE_SemParseConfig(
            name=f"{dataset}_{language}",
            language=language,
            dataset=dataset,
            version=datasets.Version("1.0.0", ""),
            description=f"Plain text import of IE-SemParse for the {language} language for {dataset} dataset",
        )
        for language, dataset in zip(_LANGUAGES, _DATASETS)
    ]

    def _info(self):
        dl_manager = datasets.DownloadManager()

        urls_to_download = self.config._URLS

        filepath = dl_manager.download_and_extract(urls_to_download)[0]

        with open(filepath, "r") as f:
            data = json.load(f)

        features = datasets.Features(
            {k: datasets.Value("string") for k in data['train'][0].keys()}
        )

        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            # No default supervised_keys (as we have to pass both premise
            # and hypothesis as input).
            supervised_keys=None,
            homepage="https://github.com/divyanshuaggarwal/IE-SemParse",
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        urls_to_download = self.config._URLS

        downloaded_file = dl_manager.download_and_extract(urls_to_download)[0]

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "split_key": "train",
                    "filepath": downloaded_file,
                    "data_format": "IE-SemParse"
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "split_key": "test",
                    "filepath": downloaded_file,
                    "data_format": "IE-SemParse"
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={
                    "split_key": "val",
                    "filepath": downloaded_file,
                    "data_format": "IE-SemParse"
                },
            ),
        ]

    def _generate_examples(self, data_format, split_key, filepath):
        """This function returns the examples in the raw (text) form."""

        with open(filepath, "r") as f:
            data = json.load(f)
            data = data[split_key]

        for idx, row in enumerate(data):
            yield idx, {
                k: v for k, v in row.items()
            }