ted_talks_iwslt.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.
"""TED TALKS IWSLT: Web Inventory of Transcribed and Translated Ted Talks in 109 languages."""


import io
import xml.etree.ElementTree as ET
import zipfile
from collections import defaultdict

import datasets

logger = datasets.logging.get_logger(__name__)


# TODO: Add BibTeX citation
# Find for instance the citation on arxiv or on the dataset repo/website
_CITATION = """\
@inproceedings{cettolo-etal-2012-wit3,
    title = "{WIT}3: Web Inventory of Transcribed and Translated Talks",
    author = "Cettolo, Mauro  and
      Girardi, Christian  and
      Federico, Marcello",
    booktitle = "Proceedings of the 16th Annual conference of the European Association for Machine Translation",
    month = may # " 28{--}30",
    year = "2012",
    address = "Trento, Italy",
    publisher = "European Association for Machine Translation",
    url = "https://www.aclweb.org/anthology/2012.eamt-1.60",
    pages = "261--268",
}
"""

# TODO: Add description of the dataset here
# You can copy an official description
_DESCRIPTION = """\
The core of WIT3 is the TED Talks corpus, that basically redistributes the original content published by the TED Conference website (http://www.ted.com). Since 2007,
the TED Conference, based in California, has been posting all video recordings of its talks together with subtitles in English
and their translations in more than 80 languages. Aside from its cultural and social relevance, this content, which is published under the Creative Commons BYNC-ND license, also represents a precious
language resource for the machine translation research community, thanks to its size, variety of topics, and covered languages.
This effort repurposes the original content in a way which is more convenient for machine translation researchers.
"""

# TODO: Add a link to an official homepage for the dataset here
_HOMEPAGE = "https://wit3.fbk.eu/"

# TODO: Add the licence for the dataset here if you can find it
_LICENSE = "CC-BY-NC-4.0"

# TODO: Add link to the official dataset URLs here
# The HuggingFace dataset library don't host the datasets but only point to the original files
# This can be an arbitrary nested dict/list of URLs (see below in `_split_generators` method)
_URL = "data/XML_releases.tgz"


_LANGUAGES = (
    "mr",
    "eu",
    "hr",
    "rup",
    "szl",
    "lo",
    "ms",
    "ht",
    "hy",
    "mg",
    "arq",
    "uk",
    "ku",
    "ig",
    "sr",
    "ug",
    "ne",
    "pt-br",
    "sq",
    "af",
    "km",
    "en",
    "tt",
    "ja",
    "inh",
    "mn",
    "eo",
    "ka",
    "nb",
    "fil",
    "uz",
    "fi",
    "tl",
    "el",
    "tg",
    "bn",
    "si",
    "gu",
    "sk",
    "kn",
    "ar",
    "hup",
    "zh-tw",
    "sl",
    "be",
    "bo",
    "fr",
    "ps",
    "tr",
    "ltg",
    "la",
    "ko",
    "lv",
    "nl",
    "fa",
    "ru",
    "et",
    "vi",
    "pa",
    "my",
    "sw",
    "az",
    "sv",
    "ga",
    "sh",
    "it",
    "da",
    "lt",
    "kk",
    "mk",
    "tlh",
    "he",
    "ceb",
    "bg",
    "fr-ca",
    "ha",
    "ml",
    "mt",
    "as",
    "pt",
    "zh-cn",
    "cnh",
    "ro",
    "hi",
    "es",
    "id",
    "bs",
    "so",
    "cs",
    "te",
    "ky",
    "hu",
    "th",
    "pl",
    "nn",
    "ca",
    "is",
    "ta",
    "de",
    "srp",
    "ast",
    "bi",
    "lb",
    "art-x-bork",
    "am",
    "oc",
    "zh",
    "ur",
    "gl",
)

# Please note that only few pairs are shown here. You can use config to generate data for all language pairs
_LANGUAGE_PAIRS = [
    ("eu", "ca"),
    ("nl", "en"),
    ("nl", "hi"),
    ("de", "ja"),
    ("fr-ca", "hi"),
]

# Year subscripts for the specific folder
_YEAR = {"2014": "-20140120", "2015": "-20150530", "2016": "-20160408"}

_YEAR_FOLDER = {
    "2014": "XML_releases/xml-20140120",
    "2015": "XML_releases/xml-20150616",
    "2016": "XML_releases/xml",
}


class TedTalksIWSLTConfig(datasets.BuilderConfig):
    """ "Builder Config for the TedTalks IWSLT dataset"""

    def __init__(self, language_pair=(None, None), year=None, **kwargs):
        """BuilderConfig for TedTalks IWSLT dataset.
        Args:
            for the `datasets.features.text.TextEncoder` used for the features feature.
            language_pair: pair of languages that will be used for translation. Should
            contain 2-letter coded strings. First will be used at source and second
            as target in supervised mode. For example: ("pl", "en").
          **kwargs: keyword arguments forwarded to super.
        """
        # Validate language pair.
        name = "%s_%s_%s" % (language_pair[0], language_pair[1], year)
        source, target = language_pair
        assert source in _LANGUAGES, f"Invalid source code in language pair: {source}"
        assert target in _LANGUAGES, f"Invalid target code in language pair: {target}"
        assert (
            source != target
        ), f"Source::{source} and Target::{target} language pairs cannot be the same!"
        assert year in _YEAR.keys()

        description = (
            f"Translation Ted Talks dataset (WIT3) between {source} and {target}"
        )
        super(TedTalksIWSLTConfig, self).__init__(
            name=name,
            description=description,
            **kwargs,
        )

        self.language_pair = language_pair
        self.year = year


# TODO: Name of the dataset usually match the script name with CamelCase instead of snake_case
class TedTalksIWSLT(datasets.GeneratorBasedBuilder):
    """TED TALKS IWSLT: Web Inventory of Transcribed and Translated Ted Talks in 109 languages."""

    VERSION = datasets.Version("1.1.0")

    BUILDER_CONFIG_CLASS = TedTalksIWSLTConfig

    BUILDER_CONFIGS = [
        TedTalksIWSLTConfig(
            language_pair=language_pair, year=year, version=datasets.Version("1.1.0")
        )
        for language_pair in _LANGUAGE_PAIRS
        for year in _YEAR.keys()
    ]

    def _info(self):
        features = datasets.Features(
            {
                "translation": datasets.features.Translation(
                    languages=self.config.language_pair
                ),
            },
        )

        return datasets.DatasetInfo(
            # This is the description that will appear on the datasets page.
            description=_DESCRIPTION,
            # This defines the different columns of the dataset and their types
            features=features,  # Here we define them above because they are different between the two configurations
            # If there's a common (input, target) tuple from the features,
            # specify them here. They'll be used if as_supervised=True in
            # builder.as_dataset.
            supervised_keys=None,
            # Homepage of the dataset for documentation
            homepage=_HOMEPAGE,
            # License for the dataset if available
            license=_LICENSE,
            # Citation for the dataset
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        """Returns SplitGenerators."""
        data_dir = dl_manager.download(_URL)

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "files": dl_manager.iter_archive(data_dir),
                },
            ),
        ]

    def _generate_examples(self, files):
        """Yields examples."""

        def parse_zip_file(path, file):
            def et_to_dict(tree):
                """This is used to convert the xml to a list of dicts"""

                dct = {tree.tag: {} if tree.attrib else None}
                children = list(tree)
                if children:
                    dd = defaultdict(list)
                    for dc in map(et_to_dict, children):
                        for k, v in dc.items():
                            dd[k].append(v)
                    dct = {tree.tag: dd}
                if tree.attrib:
                    dct[tree.tag].update((k, v) for k, v in tree.attrib.items())
                if tree.text:
                    text = tree.text.strip()
                    if children or tree.attrib:
                        if text:
                            dct[tree.tag]["text"] = text
                    else:
                        dct[tree.tag] = text
                return dct

            with zipfile.ZipFile(io.BytesIO(file)) as zf:
                try:
                    tree = ET.parse(zf.open(path.split("/")[-1][:-3] + "xml"))
                    root = tree.getroot()
                    talks = et_to_dict(root).get("xml").get("file")
                    ids = [talk.get("head")[0].get("talkid") for talk in talks]
                except Exception as pe:
                    logger.warning(f"ERROR: {pe}")
                    logger.warning(
                        "This likely means that you have a malformed XML file!"
                    )
                    ids = []
            return talks, ids

        language_pair = self.config.language_pair
        year = self.config.year

        source_file_path = (
            _YEAR_FOLDER[year] + "/ted_" + language_pair[0] + _YEAR[year] + ".zip"
        )
        target_file_path = (
            _YEAR_FOLDER[year] + "/ted_" + language_pair[1] + _YEAR[year] + ".zip"
        )

        source_talks, source_ids = None, None
        target_talks, target_ids = None, None
        for path, file in files:
            if source_ids is not None and target_ids is not None:
                break

            if source_ids is None and path.endswith(source_file_path):
                source_talks, source_ids = parse_zip_file(path, file.read())
            elif target_ids is None and path.endswith(target_file_path):
                target_talks, target_ids = parse_zip_file(path, file.read())

        if source_ids is None or target_ids is None:
            source_ids = list()
            target_ids = list()

        comm_talkids = [talkid for talkid in target_ids if talkid in source_ids]

        translation = list()

        for talkid in comm_talkids:
            source = list(
                filter(
                    lambda talk: talk.get("head")[0].get("talkid") == talkid,
                    source_talks,
                )
            )
            target = list(
                filter(
                    lambda talk: talk.get("head")[0].get("talkid") == talkid,
                    target_talks,
                )
            )

            if len(source) == 0 or len(target) == 0:
                pass
            else:
                source = source[0]
                target = target[0]

            if source.get("head")[0].get("description") and target.get("head")[0].get(
                "description"
            ):
                if (
                    source.get("head")[0].get("description")[0]
                    and target.get("head")[0].get("description")[0]
                ):
                    temp_dict = dict()
                    temp_dict["id"] = source.get("head")[0].get("talkid")[0] + "_1"
                    temp_dict[language_pair[0]] = (
                        source.get("head")[0]
                        .get("description")[0]
                        .replace("TED Talk Subtitles and Transcript: ", "")
                    )
                    temp_dict[language_pair[1]] = (
                        target.get("head")[0]
                        .get("description")[0]
                        .replace("TED Talk Subtitles and Transcript: ", "")
                    )
                    translation.append(temp_dict)

            if source.get("head")[0].get("title") and target.get("head")[0].get(
                "title"
            ):
                if (
                    source.get("head")[0].get("title")[0]
                    and target.get("head")[0].get("title")[0]
                ):
                    temp_dict = dict()
                    temp_dict["id"] = source.get("head")[0].get("talkid")[0] + "_2"
                    temp_dict[language_pair[0]] = source.get("head")[0].get("title")[0]
                    temp_dict[language_pair[1]] = target.get("head")[0].get("title")[0]
                    translation.append(temp_dict)

            if source.get("head")[0].get("seekvideo") and target.get("head")[0].get(
                "seekvideo"
            ):
                source_transc = (
                    source.get("head")[0].get("transcription")[0].get("seekvideo")
                )
                target_transc = (
                    target.get("head")[0].get("transcription")[0].get("seekvideo")
                )

                transc = zip(source_transc, target_transc)
                transcriptions = [
                    {
                        "id": s.get("id"),
                        language_pair[0]: s.get("text"),
                        language_pair[1]: t.get("text"),
                    }
                    for s, t in transc
                ]
                translation.extend(transcriptions)
        for talk_segment in translation:
            result = {
                "translation": {
                    language_pair[0]: talk_segment[language_pair[0]],
                    language_pair[1]: talk_segment[language_pair[1]],
                }
            }
            yield talk_segment["id"], result