Datasets:

biu-nlp
/

qa_discourse

Dataset card Files Files and versions Community

qa_discourse / qa_discourse.py

Aligning "no-QA sentences" denotation to QANom - records for sentences with no QAs would have both 'question' and 'answers' as empty lists

b77519b almost 2 years ago

raw history blame contribute delete

No virus

7.04 kB

	# 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: datasets.utils.download_manager.DownloadManager):
	"""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', error_bad_lines=False) for fn in filepaths]).reset_index(drop=True)
	df = df.applymap(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,
	}