tatoeba_mt.py

# coding=utf-8
# Copyright 2020 The HuggingFace Datasets Authors and the current dataset script contributor.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""tatoeba-mt"""


import os
import csv
import datasets


_CITATION = """\
@inproceedings{tiedemann-2020-tatoeba,
    title = "The {T}atoeba {T}ranslation {C}hallenge {--} {R}ealistic Data Sets for Low Resource and Multilingual {MT}",
    author = {Tiedemann, J{\"o}rg},
    booktitle = "Proceedings of the Fifth Conference on Machine Translation",
    month = nov,
    year = "2020",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2020.wmt-1.139",
    pages = "1174--1182",
}
"""

_DESCRIPTION = """\
The Tatoeba Translation Challenge is a multilingual data set of
machine translation benchmarks derived from user-contributed
translations collected by [Tatoeba.org](https://tatoeba.org/) and
provided as parallel corpus from [OPUS](https://opus.nlpl.eu/). This
dataset includes test and development data sorted by language pair. It
includes test sets for hundreds of language pairs and is continuously
updated. Please, check the version number tag to refer to the release
that your are using.
"""

_URL = {
    "evaluation": "https://object.pouta.csc.fi/Tatoeba-Challenge-devtest/test.tar",
    "development": "https://object.pouta.csc.fi/Tatoeba-Challenge-devtest/dev.tar",
}

_RELEASE = "v2021-08-07"
_RELEASES = [ "v2020-07-28", "v2021-03-30", "v2021-08-07", ]
_LanguagePairs = [ "afr-deu", "afr-eng", "afr-epo", "afr-nld", "afr-rus", "afr-spa", "ain-fin", "ara-ber", "ara-ber_Latn", "ara-deu", "ara-ell", "ara-eng", "ara-epo", "ara-fra", "ara-heb", "ara-ita", "ara-jpn", "ara-jpn_Hira", "ara-pol", "ara-rus", "ara-spa", "ara-tur", "arq-eng", "avk-fra", "avk-spa", "awa-eng", "aze-eng", "aze-spa", "aze-tur", "aze_Latn-tur", "bel-deu", "bel-eng", "bel-epo", "bel-fra", "bel-ita", "bel-lat", "bel-nld", "bel-pol", "bel-rus", "bel-spa", "bel-ukr", "bel-zho", "ben-eng", "ber-deu", "ber-eng", "ber-epo", "ber-fra", "ber-spa", "ber_Latn-deu", "ber_Latn-eng", "ber_Latn-epo", "ber_Latn-fra", "bre-eng", "bre-fra", "bua-rus", "bua_Cyrl-rus", "bul-bul", "bul-cmn_Hans", "bul-deu", "bul-eng", "bul-epo", "bul-fra", "bul-ita", "bul-jpn", "bul-jpn_Hira", "bul-rus", "bul-spa", "bul-tur", "bul-ukr", "bul-zho", "cat-deu", "cat-eng", "cat-epo", "cat-fra", "cat-ita", "cat-nld", "cat-por", "cat-spa", "cat-ukr", "cbk-eng", "ceb-deu", "ceb-eng", "ces-deu", "ces-eng", "ces-epo", "ces-fra", "ces-hun", "ces-ita", "ces-lat", "ces-pol", "ces-rus", "ces-slv", "ces-spa", "ces-ukr", "cha-eng", "chm-rus", "chv-eng", "chv-rus", "chv-tur", "cmn_Hans-wuu", "cor-deu", "cor-eng", "cor-epo", "cor-fra", "cor-ita", "cor-rus", "cor-spa", "crh-tur", "cym-eng", "dan-dan", "dan-deu", "dan-eng", "dan-epo", "dan-fin", "dan-fra", "dan-ita", "dan-jpn", "dan-jpn_Hira", "dan-nld", "dan-nob", "dan-nor", "dan-por", "dan-rus", "dan-spa", "dan-swe", "dan-tur", "deu-cmn_Hans", "deu-cmn_Hant", "deu-deu", "deu-dsb", "deu-ell", "deu-eng", "deu-epo", "deu-est", "deu-eus", "deu-fas", "deu-fin", "deu-fra", "deu-frr", "deu-gos", "deu-hbs", "deu-heb", "deu-hrv", "deu-hrx", "deu-hsb", "deu-hun", "deu-ido", "deu-ile", "deu-ina", "deu-ind", "deu-isl", "deu-ita", "deu-jbo", "deu-jpn", "deu-jpn_Hani", "deu-jpn_Hira", "deu-jpn_Kana", "deu-kab", "deu-kor", "deu-kor_Hang", "deu-kur", "deu-kur_Latn", "deu-lad", "deu-lat", "deu-lfn", "deu-lfn_Latn", "deu-lit", "deu-ltz", "deu-msa", "deu-nds", "deu-nld", "deu-nob", "deu-nor", "deu-pol", "deu-por", "deu-ron", "deu-run", "deu-rus", "deu-slv", "deu-spa", "deu-srp_Latn", "deu-swe", "deu-swg", "deu-tat", "deu-tgl", "deu-tlh", "deu-toki", "deu-tur", "deu-ukr", "deu-vie", "deu-vol", "deu-yid", "deu-zho", "dsb-hsb", "dsb-slv", "dtp-eng", "dtp-jpn", "dtp-jpn_Hira", "dtp-msa", "dtp-zsm_Latn", "egl-ita", "ell-ell", "ell-eng", "ell-epo", "ell-fra", "ell-ita", "ell-nld", "ell-por", "ell-rus", "ell-spa", "ell-swe", "ell-tur", "eng-bos_Latn", "eng-cmn_Hans", "eng-cmn_Hant", "eng-eng", "eng-epo", "eng-est", "eng-eus", "eng-fao", "eng-fas", "eng-fin", "eng-fra", "eng-fry", "eng-gla", "eng-gle", "eng-glg", "eng-gos", "eng-got", "eng-grc", "eng-gsw", "eng-hbs", "eng-heb", "eng-hin", "eng-hoc", "eng-hoc_Latn", "eng-hrv", "eng-hrx", "eng-hun", "eng-hye", "eng-ido", "eng-ido_Latn", "eng-ile", "eng-ilo", "eng-ina", "eng-ind", "eng-isl", "eng-ita", "eng-jav", "eng-jbo", "eng-jbo_Latn", "eng-jpn", "eng-jpn_Hani", "eng-jpn_Hira", "eng-jpn_Kana", "eng-kab", "eng-kat", "eng-kaz", "eng-kaz_Cyrl", "eng-kha", "eng-khm", "eng-kor", "eng-kor_Hang", "eng-kur", "eng-kur_Latn", "eng-kzj", "eng-lad", "eng-lad_Latn", "eng-lat", "eng-lav", "eng-lfn", "eng-lfn_Cyrl", "eng-lfn_Latn", "eng-lit", "eng-ltz", "eng-mal", "eng-mar", "eng-mkd", "eng-mlt", "eng-mon", "eng-mri", "eng-msa", "eng-mya", "eng-nds", "eng-nld", "eng-nno", "eng-nob", "eng-nor", "eng-nov", "eng-nst", "eng-oci", "eng-orv", "eng-ota", "eng-ota_Arab", "eng-ota_Latn", "eng-pam", "eng-pes", "eng-pms", "eng-pol", "eng-por", "eng-prg", "eng-que", "eng-rom", "eng-ron", "eng-run", "eng-rus", "eng-slv", "eng-spa", "eng-sqi", "eng-srp_Cyrl", "eng-srp_Latn", "eng-swa", "eng-swe", "eng-tam", "eng-tat", "eng-tel", "eng-tgl", "eng-tha", "eng-tlh", "eng-toki", "eng-tuk", "eng-tuk_Latn", "eng-tur", "eng-tzl", "eng-tzl_Latn", "eng-uig", "eng-uig_Arab", "eng-ukr", "eng-urd", "eng-uzb", "eng-uzb_Latn", "eng-vie", "eng-vol", "eng-war", "eng-xal", "eng-yid", "eng-yue_Hans", "eng-yue_Hant", "eng-zho", "eng-zsm_Latn", "eng-zza", "epo-cmn_Hans", "epo-cmn_Hant", "epo-epo", "epo-fas", "epo-fin", "epo-fra", "epo-glg", "epo-hbs", "epo-heb", "epo-hrv", "epo-hun", "epo-ido", "epo-ile", "epo-ile_Latn", "epo-ina", "epo-isl", "epo-ita", "epo-jbo", "epo-jpn", "epo-jpn_Hani", "epo-jpn_Hira", "epo-lad", "epo-lad_Latn", "epo-lat", "epo-lfn", "epo-lfn_Latn", "epo-lit", "epo-nds", "epo-nld", "epo-nob", "epo-nor", "epo-oci", "epo-pol", "epo-por", "epo-ron", "epo-rus", "epo-slv", "epo-spa", "epo-srp_Cyrl", "epo-srp_Latn", "epo-swe", "epo-tgl", "epo-tlh", "epo-toki", "epo-tur", "epo-ukr", "epo-vie", "epo-vol", "epo-yid", "epo-zho", "est-rus", "eus-jpn", "eus-rus", "eus-spa", "fas-fra", "fin-fin", "fin-fkv", "fin-fra", "fin-heb", "fin-hun", "fin-ita", "fin-jpn", "fin-jpn_Hani", "fin-jpn_Hira", "fin-jpn_Kana", "fin-kor", "fin-kor_Hang", "fin-kur", "fin-lat", "fin-nld", "fin-nno", "fin-nob", "fin-nor", "fin-pol", "fin-por", "fin-rus", "fin-spa", "fin-swe", "fin-tur", "fin-zho", "fra-cmn_Hans", "fra-cmn_Hant", "fra-fra", "fra-gcf", "fra-hbs", "fra-heb", "fra-hrv", "fra-hun", "fra-ido", "fra-ile", "fra-ina", "fra-ind", "fra-ita", "fra-jbo", "fra-jpn", "fra-jpn_Hani", "fra-jpn_Hira", "fra-kab", "fra-kor", "fra-kor_Hang", "fra-lat", "fra-lfn", "fra-lfn_Latn", "fra-msa", "fra-nds", "fra-nld", "fra-nob", "fra-nor", "fra-oci", "fra-pcd", "fra-pol", "fra-por", "fra-ron", "fra-run", "fra-rus", "fra-slv", "fra-spa", "fra-swe", "fra-tat", "fra-tgl", "fra-tlh", "fra-tlh_Latn", "fra-toki", "fra-toki_Latn", "fra-tur", "fra-uig", "fra-uig_Arab", "fra-ukr", "fra-vie", "fra-wuu", "fra-yid", "fra-zho", "fry-nld", "gcf-gcf", "gla-spa", "glg-por", "glg-spa", "gos-nld", "grn-por", "grn-spa", "hbs-ita", "hbs-jpn", "hbs-nor", "hbs-pol", "hbs-rus", "hbs-spa", "hbs-ukr", "hbs-zho", "heb-cmn_Hans", "heb-cmn_Hant", "heb-heb", "heb-hun", "heb-ina", "heb-ita", "heb-jpn", "heb-jpn_Hira", "heb-lad", "heb-lat", "heb-lfn", "heb-lfn_Latn", "heb-nld", "heb-pol", "heb-por", "heb-rus", "heb-spa", "heb-tur", "heb-ukr", "heb-yid", "heb-zho", "hin-urd", "hin-zho", "hrv-jpn_Hira", "hrv-pol", "hrv-spa", "hrv-ukr", "hsb-slv", "hun-cmn_Hans", "hun-hun", "hun-ita", "hun-jpn", "hun-jpn_Hani", "hun-jpn_Hira", "hun-kor", "hun-kor_Hang", "hun-lat", "hun-nld", "hun-pol", "hun-por", "hun-rus", "hun-spa", "hun-swe", "hun-tur", "hun-ukr", "hun-zho", "hye-rus", "ido-ina", "ido-ita", "ido-lfn", "ido-spa", "ido-yid", "ido_Latn-lfn_Latn", "ina-ita", "ina-lad", "ina-lat", "ina-lfn", "ina-nld", "ina-por", "ina-rus", "ina-spa", "ina-tlh", "ina-tur", "ina-yid", "ina_Latn-lad_Latn", "ina_Latn-lfn_Latn", "ina_Latn-tlh_Latn", "ind-zsm_Latn", "isl-ita", "isl-jpn", "isl-jpn_Hira", "isl-spa", "ita-cmn_Hans", "ita-cmn_Hant", "ita-ind", "ita-ita", "ita-jpn", "ita-jpn_Hani", "ita-jpn_Hira", "ita-lat", "ita-lit", "ita-msa", "ita-nds", "ita-nld", "ita-nor", "ita-pms", "ita-pol", "ita-por", "ita-ron", "ita-rus", "ita-spa", "ita-swe", "ita-toki", "ita-tur", "ita-ukr", "ita-vie", "ita-yid", "ita-zho", "jbo-jpn", "jbo-rus", "jbo-spa", "jbo-swe", "jbo-zho", "jbo_Latn-cmn_Hans", "jbo_Latn-cmn_Hant", "jbo_Latn-jpn_Hira", "jpn-jpn", "jpn-kor", "jpn-lit", "jpn-mar", "jpn-msa", "jpn-nds", "jpn-nld", "jpn-nor", "jpn-pol", "jpn-por", "jpn-rus", "jpn-spa", "jpn-swe", "jpn-tlh", "jpn-toki", "jpn-tur", "jpn-ukr", "jpn-vie", "jpn-zho", "jpn_Hani-cmn_Hans", "jpn_Hani-nld", "jpn_Hani-pol", "jpn_Hani-por", "jpn_Hani-rus", "jpn_Hani-spa", "jpn_Hira-cmn_Hans", "jpn_Hira-cmn_Hant", "jpn_Hira-ind", "jpn_Hira-jpn_Hira", "jpn_Hira-kor_Hang", "jpn_Hira-lit", "jpn_Hira-mar", "jpn_Hira-nds", "jpn_Hira-nld", "jpn_Hira-nob", "jpn_Hira-pol", "jpn_Hira-por", "jpn_Hira-rus", "jpn_Hira-spa", "jpn_Hira-swe", "jpn_Hira-tlh_Latn", "jpn_Hira-tur", "jpn_Hira-ukr", "jpn_Hira-vie", "jpn_Kana-rus", "jpn_Kana-spa", "kab-kab", "kab-rus", "kab-spa", "kat-rus", "kaz-rus", "kaz_Cyrl-rus", "khm-spa", "kor-rus", "kor-spa", "kor-zho", "kor_Hang-cmn_Hans", "kor_Hang-rus", "kor_Hang-spa", "kzj-msa", "kzj_Latn-zsm_Latn", "lad-lat", "lad-lfn", "lad-spa", "lad-yid", "lad_Latn-lfn_Latn", "lad_Latn-spa", "lad_Latn-yid", "lat-lat", "lat-lfn", "lat-nld", "lat-nor", "lat-pol", "lat-por", "lat-rus", "lat-spa", "lat-tlh", "lat-ukr", "lat-yid", "lat_Latn-lfn_Latn", "lat_Latn-por", "lav-rus", "lfn-por", "lfn-rus", "lfn-spa", "lfn-yid", "lfn_Cyrl-por", "lfn_Latn-por", "lfn_Latn-yid", "lit-pol", "lit-rus", "lit-spa", "lit-tur", "ltz-nld", "mkd-spa", "msa-msa", "msa-spa", "msa-zho", "nds-nld", "nds-por", "nds-rus", "nds-spa", "nld-cmn_Hant", "nld-nld", "nld-nor", "nld-pol", "nld-por", "nld-ron", "nld-rus", "nld-spa", "nld-toki", "nld-tur", "nld-ukr", "nld-zho", "nno-nob", "nob-nno", "nob-rus", "nob-spa", "nob-swe", "nor-nor", "nor-pol", "nor-por", "nor-rus", "nor-spa", "nor-swe", "nor-ukr", "nor-zho", "orv-ukr", "ota-tur", "pol-cmn_Hans", "pol-cmn_Hant", "pol-por", "pol-rus", "pol-spa", "pol-swe", "pol-tur", "pol-ukr", "pol-zho", "por-cmn_Hans", "por-cmn_Hant", "por-por", "por-ron", "por-rus", "por-spa", "por-swe", "por-tgl", "por-toki", "por-tur", "por-ukr", "por-zho", "ron-rus", "ron-spa", "ron-tur", "run-rus", "run-spa", "rus-cmn_Hans", "rus-cmn_Hant", "rus-rus", "rus-sah", "rus-slv", "rus-spa", "rus-swe", "rus-tat", "rus-tlh", "rus-toki", "rus-toki_Latn", "rus-tur", "rus-uig", "rus-uig_Arab", "rus-ukr", "rus-vie", "rus-xal", "rus-yue_Hans", "rus-zho", "slv-cmn_Hans", "slv-ukr", "slv-zho", "spa-cmn_Hans", "spa-cmn_Hant", "spa-spa", "spa-swe", "spa-tat", "spa-tgl", "spa-tlh", "spa-toki", "spa-tur", "spa-ukr", "spa-vie", "spa-yid", "spa-zho", "srp_Cyrl-rus", "srp_Cyrl-ukr", "srp_Latn-ita", "srp_Latn-nob", "srp_Latn-rus", "srp_Latn-ukr", "swe-cmn_Hans", "swe-cmn_Hant", "swe-swe", "swe-tur", "swe-zho", "tat-tur", "tat-vie", "tlh-yid", "tlh-zho", "tlh_Latn-cmn_Hans", "tlh_Latn-cmn_Hant", "tlh_Latn-yid", "tur-cmn_Hans", "tur-cmn_Hant", "tur-tur", "tur-uig", "tur-ukr", "tur-uzb", "tur-zho", "uig-zho", "uig_Arab-cmn_Hans", "uig_Arab-cmn_Hant", "ukr-cmn_Hans", "ukr-cmn_Hant", "ukr-ukr", "ukr-zho", "vie-cmn_Hans", "vie-vie", "vie-zho", "wuu-cmn_Hans", "yid-yid", "zho-zho", "zsm_Latn-ind", ]


class tatoebaMTConfig(datasets.BuilderConfig):
    """BuilderConfig for tatoeba-mt"""

    def __init__(self, language_pair, release=_RELEASE, **kwargs):
        super().__init__(**kwargs)
        """

        Args:
            language_pair: language pair, you want to load
            **kwargs: keyword arguments forwarded to super.
        """
        self.language_pair = language_pair
        self.release = release


class tatoebaMT(datasets.GeneratorBasedBuilder):
    """tatoeba-mt is multilingual and covers a large number of language pairs."""

    VERSION = datasets.Version("1.0.0")

    BUILDER_CONFIG_CLASS = tatoebaMTConfig
    BUILDER_CONFIGS = [
        tatoebaMTConfig(name=pair, description=_DESCRIPTION, language_pair=pair )
        for pair in _LanguagePairs
    ]

    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features({
                "sourceLang": datasets.Value("string"),
                "targetlang": datasets.Value("string"),
                "sourceString": datasets.Value("string"),
                "targetString": datasets.Value("string")
            }),
            homepage="https://github.com/Helsinki-NLP/Tatoeba-Challenge/",
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):

        lang_pair = self.config.language_pair
        release = self.config.release
        output = []

        # testfile = os.path.join("test", release, f"tatoeba-test.{lang_pair}.tsv")
        testfile = os.path.join("test", f"tatoeba-test.{lang_pair}.tsv")
        if os.path.exists(testfile):
            test = datasets.SplitGenerator(
                name=datasets.Split.TEST,
                # These kwargs will be passed to _generate_examples
                gen_kwargs={
                    "filepath": testfile,
                },
            )
            output.append(test)

        # devfile = os.path.join("dev", release, f"tatoeba-dev.{lang_pair}.tsv")
        devfile = os.path.join("dev", f"tatoeba-dev.{lang_pair}.tsv")
        if os.path.exists(devfile):
            valid = datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                # These kwargs will be passed to _generate_examples
                gen_kwargs={
                    "filepath": devfile,
                },
            )
            output.append(valid)

        return output

    def _generate_examples(self, filepath):
        """Yields examples."""
        with open(filepath, encoding="utf-8") as f:
            reader = csv.reader(f, delimiter="\t", quotechar='"')
            for id_, row in enumerate(reader):
                if id_ == 0:
                    continue
                yield id_, {
                    "sourceLang": row[0],
                    "targetlang": row[1],
                    "sourceString": row[2],
                    "targetString": row[3]
                }