disc_eval_mt.py

# Copyright 2023 The Inseq Team. All rights reserved.
#
# 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.
"""DiscEvalMT: Contrastive test sets for the evaluation of discourse in machine translation (v2)"""

from typing import Dict

import datasets
from datasets.utils.download_manager import DownloadManager

_CITATION = """\
@inproceedings{bawden-etal-2018-evaluating,
    title = "Evaluating Discourse Phenomena in Neural Machine Translation",
    author = "Bawden, Rachel and Sennrich, Rico and Birch, Alexandra and Haddow, Barry",
    booktitle = {{Proceedings of the 2018 Conference of the North {A}merican Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long Papers)}},
    month = jun,
    year = "2018",
    address = "New Orleans, Louisiana",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/N18-1118",
    doi = "10.18653/v1/N18-1118",
    pages = "1304--1313"
}
"""

_DESCRIPTION = """\
The test sets comprise hand-crafted examples that are inspired by similar examples in the parallel corpus OpenSubtitles2016 (in terms of vocabulary usage, style and syntactic formulation)
for the evaluation of discourse in English-to-French machine translation.
"""

_URL = "https://huggingface.co/datasets/inseq/disc_eval_mt/raw/main/examples"

_HOMEPAGE = "https://github.com/rbawden/discourse-mt-test-sets"

_LICENSE = "CC-BY-SA-4.0"

_CONFIGS = ["anaphora", "lexical-choice"]


class DiscEvalMTConfig(datasets.BuilderConfig):
    def __init__(self, source_language: str, target_language: str, **kwargs):
        """BuilderConfig for DiscEvalMT.
        Args:
            source_language: `str`, source language for translation.
            target_language: `str`, translation language.
            **kwargs: keyword arguments forwarded to super.
        """
        super().__init__(**kwargs)
        self.source_language = source_language
        self.target_language = target_language


class DiscEvalMT(datasets.GeneratorBasedBuilder):
    VERSION = datasets.Version("1.0.0")

    BUILDER_CONFIGS = [
        DiscEvalMTConfig(
            name=cfg,
            source_language="en",
            target_language="fr",
        )
        for cfg in _CONFIGS
    ]

    DEFAULT_CONFIG_NAME = "anaphora"

    def _info(self):
        features = datasets.Features(
            {
                "id": datasets.Value("int32"),
                "context_en": datasets.Value("string"),
                "en": datasets.Value("string"),
                "context_fr": datasets.Value("string"),
                "fr": datasets.Value("string"),
                "contrast_fr": datasets.Value("string"),
                "context_en_with_tags": datasets.Value("string"),
                "en_with_tags": datasets.Value("string"),
                "context_fr_with_tags": datasets.Value("string"),
                "fr_with_tags": datasets.Value("string"),
                "contrast_fr_with_tags": datasets.Value("string"),
                "type": datasets.Value("string"),
            }
        )
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            homepage=_HOMEPAGE,
            license=_LICENSE,
            citation=_CITATION,
        )

    @staticmethod
    def clean_string(txt: str):
        return txt.replace("<p>", "").replace("</p>", "").replace("<hon>", "").replace("<hoff>", "")

    def _split_generators(self, dl_manager: DownloadManager):
        """Returns SplitGenerators."""
        filepaths = {}
        for lang in ["en", "fr"]:
            for ftype in ["context", "current"]:
                fname = f"{self.config.name}.{ftype}.{lang}"
                filepaths[f"{ftype}_{lang}"] = dl_manager.download_and_extract(f"{_URL}/{fname}")
        filepaths["contrast_fr"] = dl_manager.download_and_extract(f"{_URL}/{self.config.name}.contrast.fr")
        filepaths["type"] = dl_manager.download_and_extract(f"{_URL}/{self.config.name}.type")
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "filepaths": filepaths,
                    "cfg_name": self.config.name,
                },
            )
        ]

    def _generate_examples(self, filepaths: Dict[str, str], cfg_name: str):
        """Yields examples as (key, example) tuples."""

        with open(filepaths["current_en"]) as f:
            current_en = f.read().splitlines()
        with open(filepaths["current_fr"]) as f:
            current_fr = f.read().splitlines()
        with open(filepaths["context_en"]) as f:
            context_en = f.read().splitlines()
        with open(filepaths["context_fr"]) as f:
            context_fr = f.read().splitlines()
        with open(filepaths["contrast_fr"]) as f:
            contrast_fr = f.read().splitlines()
        with open(filepaths["type"]) as f:
            alltyp = f.read().splitlines()
        for i, (curr_en, curr_fr, ctx_en, ctx_fr, con_fr, typ) in enumerate(
            zip(current_en, current_fr, context_en, context_fr, contrast_fr, alltyp)
        ):
            yield i, {
                "id": i,
                "context_en": self.clean_string(ctx_en),
                "en": self.clean_string(curr_en),
                "context_fr": self.clean_string(ctx_fr),
                "fr": self.clean_string(curr_fr),
                "contrast_fr": self.clean_string(con_fr),
                "context_en_with_tags": ctx_en,
                "en_with_tags": curr_en,
                "context_fr_with_tags": ctx_fr,
                "fr_with_tags": curr_fr,
                "contrast_fr_with_tags": con_fr,
                "type": typ,
            }