sranantongo.py

import datasets


_DESCRIPTION = """\
This datasets consists of monolingual (Sranantongo) and parallel (Sranantongo - Dutch) data.
"""

_CITATION = """\
@article{zwennicker2022towards,
  title={Towards a general purpose machine translation system for Sranantongo},
  author={Zwennicker, Just and Stap, David},
  journal={arXiv preprint arXiv:2212.06383},
  year={2022}
}
"""

_DATA_URL = "data/sranantongo.tar"

_LANGUAGE2FILES = {
    "srn":          {"train": "srn_mono_SIL.csv", "validation": None, "test": None}, 
    "srn-nl_jw":    {split:f"srn-nl_JW_{split}.csv" for split in ["train", "validation", "test"]},
    "srn-nl_other": {split:f"srn-nl_other_{split}.csv" for split in ["train", "validation", "test"]},
}


class SranantongoConfig(datasets.BuilderConfig):
    """BuilderConfig for Sranantongo dataset."""

    def __init__(self, name=str, **kwargs):
        self.name = name
        description = "Monolingual sentences in `Sranantongo`." if "mono" in self.name else f"Parallel sentences in `Sranantongo` and `Dutch`."
        super(SranantongoConfig, self).__init__(name=self.name, description=description, **kwargs)


class Sranantongo(datasets.GeneratorBasedBuilder):
    """Sranantongo data from https://arxiv.org/abs/2212.06383"""    
    BUILDER_CONFIGS = [
        SranantongoConfig(name=name, version=datasets.Version("1.0.0", ""))
        for name in _LANGUAGE2FILES.keys()
    ]


    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features(
                {
                    "srn": datasets.Value("string"),
                    **(
                        {"nl": datasets.Value("string")}
                        if "srn-nl" in self.config.name else {}
                    )
                }
            ),
            homepage="https://arxiv.org/abs/2212.06383",
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        files = dl_manager.iter_archive(dl_manager.download(_DATA_URL))

        # Always generate the train split
        generators = [
            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"files": files, "split": "train"})
        ]
        
        # If the dataset configuration is for parallel data, add validation and test splits
        if "srn-nl" in self.config.name:
            for split in [datasets.Split.VALIDATION, datasets.Split.TEST]:
                generators.append(
                    datasets.SplitGenerator(name=split, gen_kwargs={"files": files, "split": split})
                )

        return generators


    def _generate_examples(self, split, files):
        """Returns examples as raw text."""

        if "srn-nl" in self.config.name:
            return self._generate_examples_parallel(split=split, files=files)
        else:
            return self._generate_examples_mono(split=split, files=files)


    def _generate_examples_mono(self, split, files):
        for path, file in files:
            if path == _LANGUAGE2FILES[self.config.name][split]:
                data = file.read().decode("utf-8").split("\n")

        for idx, sentence in enumerate(data):
            yield idx, {"srn": sentence}



    def _generate_examples_parallel(self, split, files):
        for path, file in files:
            if path == _LANGUAGE2FILES[self.config.name][split]:
                data = file.read().decode("utf-8").split("\n")

        for idx, sentence in enumerate(data):
            nl, srn = sentence.split("|")
            yield idx, {"nl": nl, "srn": srn}