qa_discourse.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.
"""A Dataset loading script for the QA-Discourse dataset (Pyatkin et. al., ACL 2020)."""


import datasets
from pathlib import Path
from typing import List
import pandas as pd


_CITATION = """\
@inproceedings{pyatkin2020qadiscourse,
  title={QADiscourse-Discourse Relations as QA Pairs: Representation, Crowdsourcing and Baselines},
  author={Pyatkin, Valentina and Klein, Ayal and Tsarfaty, Reut and Dagan, Ido},
  booktitle={Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
  pages={2804--2819},
  year={2020}
}"""


_DESCRIPTION = """\
The dataset contains question-answer pairs to model discourse relations. 
While answers roughly correspond to spans of the sentence, these spans could have been freely adjusted by annotators to grammaticaly fit the question;
Therefore, answers are given just as text and not as identified spans of the original sentence.    
See the paper for details: QADiscourse - Discourse Relations as QA Pairs: Representation, Crowdsourcing and Baselines, Pyatkin et. al., 2020
"""

_HOMEPAGE = "https://github.com/ValentinaPy/QADiscourse"

_LICENSE = """Resources on this page are licensed CC-BY 4.0, a Creative Commons license requiring Attribution (https://creativecommons.org/licenses/by/4.0/)."""


_URLs = {
    "wikinews.train": "https://github.com/ValentinaPy/QADiscourse/raw/master/Dataset/wikinews_train.tsv",
    "wikinews.dev": "https://github.com/ValentinaPy/QADiscourse/raw/master/Dataset/wikinews_dev.tsv",
    "wikinews.test": "https://github.com/ValentinaPy/QADiscourse/raw/master/Dataset/wikinews_test.tsv",
    "wikipedia.train": "https://github.com/ValentinaPy/QADiscourse/raw/master/Dataset/wikipedia_train.tsv",
    "wikipedia.dev": "https://github.com/ValentinaPy/QADiscourse/raw/master/Dataset/wikipedia_dev.tsv",
    "wikipedia.test": "https://github.com/ValentinaPy/QADiscourse/raw/master/Dataset/wikipedia_test.tsv",      
}

COLUMNS = ['qasrl_id', 'sentence', 'worker_id', 'full_question', 'full_answer',
       'question_start', 'question_aux', 'question_body', 'answer',
       'untokenized sentence', 'target indices for untok sent']


# TODO: Name of the dataset usually match the script name with CamelCase instead of snake_case
class QaDiscourse(datasets.GeneratorBasedBuilder):
    """QA-Discourse: Discourse Relations as Question-Answer Pairs.  """

    VERSION = datasets.Version("1.0.2")

    BUILDER_CONFIGS = [
        datasets.BuilderConfig(
            name="plain_text", version=VERSION, description="This provides the QA-Discourse dataset"
        ),
    ]

    DEFAULT_CONFIG_NAME = (
        "plain_text"  # It's not mandatory to have a default configuration. Just use one if it make sense.
    )

    def _info(self):
        features = datasets.Features(
            {
                "sentence": datasets.Value("string"),
                "sent_id": datasets.Value("string"),
                "question": datasets.Sequence(datasets.Value("string")),
                "answers": datasets.Sequence(datasets.Value("string")),
            }
        )
        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."""            
        
        # Download and prepare all files - keep same structure as _URLs 
        corpora = {section:  Path(dl_manager.download_and_extract(_URLs[section])) 
                   for section in _URLs} 
            
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                # These kwargs will be passed to _generate_examples
                gen_kwargs={
                    "filepaths": [corpora["wikinews.train"], 
                                  corpora["wikipedia.train"]],
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                # These kwargs will be passed to _generate_examples
                gen_kwargs={
                    "filepaths": [corpora["wikinews.dev"], 
                                  corpora["wikipedia.dev"]],
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                # These kwargs will be passed to _generate_examples
                gen_kwargs={
                    "filepaths": [corpora["wikinews.test"], 
                                  corpora["wikipedia.test"]],
                },
            ),
        ]
    
    def _generate_examples(self, filepaths: List[str]):

        """ 
        Yields QA-Discourse examples from a tsv file.
        Sentences with no QAs will yield an ``empty QA'' record, where both 'question' and 'answers' are empty lists.
        """

        # merge annotations from sections 
        df = pd.concat([pd.read_csv(fn, sep='\t', on_bad_lines='skip') for fn in filepaths]).reset_index(drop=True) 
        df = df.map(str) # must turn all values to strings explicitly to avoid type errors
        for counter, row in df.iterrows():
            # Prepare question (3 "slots" and question mark)
            question = [row.question_start, row.question_aux, row.question_body.rstrip('?'), '?']
            answer = [row.answer]
            if row.question_start == "_":    # sentence has no QAs
                question = []
                answer = []
            
            yield counter, {
                "sentence": row.sentence,
                "sent_id": row.qasrl_id,
                "question": question,
                "answers": answer,
            }