nusax_mt.py

from pathlib import Path
from typing import Dict, List, Tuple

import datasets
import pandas as pd

from seacrowd.utils import schemas
from seacrowd.utils.configs import SEACrowdConfig
from seacrowd.utils.constants import (DEFAULT_SEACROWD_VIEW_NAME,
                                       DEFAULT_SOURCE_VIEW_NAME, Tasks)

_DATASETNAME = "nusax_mt"
_SOURCE_VIEW_NAME = DEFAULT_SOURCE_VIEW_NAME
_UNIFIED_VIEW_NAME = DEFAULT_SEACROWD_VIEW_NAME

_LANGUAGES = ["ind", "ace", "ban", "bjn", "bbc", "bug", "jav", "mad", "min", "nij", "sun", "eng"]  # We follow ISO639-3 language code (https://iso639-3.sil.org/code_tables/639/data)
_LOCAL = False

_CITATION = """\
@misc{winata2022nusax,
      title={NusaX: Multilingual Parallel Sentiment Dataset for 10 Indonesian Local Languages},
      author={Winata, Genta Indra and Aji, Alham Fikri and Cahyawijaya,
      Samuel and Mahendra, Rahmad and Koto, Fajri and Romadhony,
      Ade and Kurniawan, Kemal and Moeljadi, David and Prasojo,
      Radityo Eko and Fung, Pascale and Baldwin, Timothy and Lau,
      Jey Han and Sennrich, Rico and Ruder, Sebastian},
      year={2022},
      eprint={2205.15960},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}
"""

_DESCRIPTION = """\
NusaX is a high-quality multilingual parallel corpus that covers 12 languages, Indonesian, English, and 10 Indonesian local languages, namely Acehnese, Balinese, Banjarese, Buginese, Madurese, Minangkabau, Javanese, Ngaju, Sundanese, and Toba Batak.

NusaX-MT is a parallel corpus for training and benchmarking machine translation models across 10 Indonesian local languages + Indonesian and English. The data is presented in csv format with 12 columns, one column for each language.
"""

_HOMEPAGE = "https://github.com/IndoNLP/nusax/tree/main/datasets/mt"

_LICENSE = "Creative Commons Attribution Share-Alike 4.0 International"

_SUPPORTED_TASKS = [Tasks.MACHINE_TRANSLATION]

_SOURCE_VERSION = "1.0.0"

_SEACROWD_VERSION = "2024.06.20"

_URLS = {
    "train": "https://raw.githubusercontent.com/IndoNLP/nusax/main/datasets/mt/train.csv",
    "validation": "https://raw.githubusercontent.com/IndoNLP/nusax/main/datasets/mt/valid.csv",
    "test": "https://raw.githubusercontent.com/IndoNLP/nusax/main/datasets/mt/test.csv",
}


def seacrowd_config_constructor(lang_source, lang_target, schema, version):
    """Construct SEACrowdConfig with nusax_mt_{lang_source}_{lang_target}_{schema} as the name format"""
    if schema != "source" and schema != "seacrowd_t2t":
        raise ValueError(f"Invalid schema: {schema}")

    if lang_source == "" and lang_target == "":
        return SEACrowdConfig(
            name="nusax_mt_{schema}".format(schema=schema),
            version=datasets.Version(version),
            description="nusax_mt with {schema} schema for all 132 language pairs".format(schema=schema),
            schema=schema,
            subset_id="nusax_mt",
        )
    else:
        return SEACrowdConfig(
            name="nusax_mt_{lang_source}_{lang_target}_{schema}".format(lang_source=lang_source, lang_target=lang_target, schema=schema),
            version=datasets.Version(version),
            description="nusax_mt with {schema} schema for {lang_source} source language and  {lang_target} target language".format(lang_source=lang_source, lang_target=lang_target, schema=schema),
            schema=schema,
            subset_id="nusax_mt",
        )


LANGUAGES_MAP = {
    "ace": "acehnese",
    "ban": "balinese",
    "bjn": "banjarese",
    "bug": "buginese",
    "eng": "english",
    "ind": "indonesian",
    "jav": "javanese",
    "mad": "madurese",
    "min": "minangkabau",
    "nij": "ngaju",
    "sun": "sundanese",
    "bbc": "toba_batak",
}


class NusaXMT(datasets.GeneratorBasedBuilder):
    """NusaX-MT is a parallel corpus for training and benchmarking machine translation models across 10 Indonesian local languages + Indonesian and English. The data is presented in csv format with 12 columns, one column for each language."""

    BUILDER_CONFIGS = (
        [seacrowd_config_constructor(lang1, lang2, "source", _SOURCE_VERSION) for lang1 in LANGUAGES_MAP for lang2 in LANGUAGES_MAP if lang1 != lang2]
        + [seacrowd_config_constructor(lang1, lang2, "seacrowd_t2t", _SEACROWD_VERSION) for lang1 in LANGUAGES_MAP for lang2 in LANGUAGES_MAP if lang1 != lang2]
        + [seacrowd_config_constructor("", "", "source", _SOURCE_VERSION), seacrowd_config_constructor("", "", "seacrowd_t2t", _SEACROWD_VERSION)]
    )

    DEFAULT_CONFIG_NAME = "nusax_senti_ind_eng_source"

    def _info(self) -> datasets.DatasetInfo:
        if self.config.schema == "source" or self.config.schema == "seacrowd_t2t":
            features = schemas.text2text_features
        else:
            raise ValueError(f"Invalid config schema: {self.config.schema}")

        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."""
        train_csv_path = Path(dl_manager.download_and_extract(_URLS["train"]))
        validation_csv_path = Path(dl_manager.download_and_extract(_URLS["validation"]))
        test_csv_path = Path(dl_manager.download_and_extract(_URLS["test"]))

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={"filepath": train_csv_path},
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={"filepath": validation_csv_path},
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={"filepath": test_csv_path},
            ),
        ]

    def _generate_examples(self, filepath: Path) -> Tuple[int, Dict]:
        if self.config.schema != "source" and self.config.schema != "seacrowd_t2t":
            raise ValueError(f"Invalid config schema: {self.config.schema}")

        df = pd.read_csv(filepath).reset_index()
        if self.config.name == "nusax_mt_source" or self.config.name == "nusax_mt_seacrowd_t2t":
            # load all 132 language pairs
            id_count = -1
            for lang_source in LANGUAGES_MAP:
                for lang_target in LANGUAGES_MAP:
                    if lang_source == lang_target:
                        continue

                    for _, row in df.iterrows():
                        id_count += 1
                        ex = {
                            "id": str(id_count),
                            "text_1": row[LANGUAGES_MAP[lang_source]],
                            "text_2": row[LANGUAGES_MAP[lang_target]],
                            "text_1_name": lang_source,
                            "text_2_name": lang_target,
                        }
                        yield id_count, ex

        else:
            df = pd.read_csv(filepath).reset_index()
            lang_source = self.config.name[9:12]
            lang_target = self.config.name[13:16]

            for index, row in df.iterrows():
                ex = {
                    "id": str(index),
                    "text_1": row[LANGUAGES_MAP[lang_source]],
                    "text_2": row[LANGUAGES_MAP[lang_target]],
                    "text_1_name": lang_source,
                    "text_2_name": lang_target,
                }
                yield str(index), ex