fever.py

# Modified by Nora Belrose of EleutherAI (2023)
# coding=utf-8
# Copyright 2020 The TensorFlow Datasets Authors and the HuggingFace Datasets Authors.
#
# 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.

# Lint as: python3
"""FEVER dataset."""

import json
import os
import textwrap

import datasets


class FeverConfig(datasets.BuilderConfig):
    """BuilderConfig for FEVER."""

    def __init__(self, homepage: str = None, citation: str = None, base_url: str = None, urls: dict = None, **kwargs):
        """BuilderConfig for FEVER.

        Args:
            homepage (`str`): Homepage.
            citation (`str`): Citation reference.
            base_url (`str`): Data base URL that precedes all data URLs.
            urls (`dict`): Data URLs (each URL will pe preceded by `base_url`).
            **kwargs: keyword arguments forwarded to super.
        """
        super().__init__(**kwargs)
        self.homepage = homepage
        self.citation = citation
        self.base_url = base_url
        self.urls = {key: f"{base_url}/{url}" for key, url in urls.items()}


class Fever(datasets.GeneratorBasedBuilder):
    """Fact Extraction and VERification Dataset."""

    BUILDER_CONFIGS = [
        FeverConfig(
            name="v1.0",
            version=datasets.Version("1.0.0"),
            description=textwrap.dedent(
                "FEVER  v1.0\n"
                "FEVER (Fact Extraction and VERification) consists of 185,445 claims generated by altering sentences "
                "extracted from Wikipedia and subsequently verified without knowledge of the sentence they were "
                "derived from. The claims are classified as Supported, Refuted or NotEnoughInfo. For the first two "
                "classes, the annotators also recorded the sentence(s) forming the necessary evidence for their "
                "judgment."
            ),
            homepage="https://fever.ai/dataset/fever.html",
            citation=textwrap.dedent(
                """\
                @inproceedings{Thorne18Fever,
                    author = {Thorne, James and Vlachos, Andreas and Christodoulopoulos, Christos and Mittal, Arpit},
                    title = {{FEVER}: a Large-scale Dataset for Fact Extraction and {VERification}},
                    booktitle = {NAACL-HLT},
                    year = {2018}
                }"""
            ),
            base_url="https://fever.ai/download/fever",
            urls={
                datasets.Split.TRAIN: "train.jsonl",
                "dev": "shared_task_dev.jsonl",
                "paper_dev": "paper_dev.jsonl",
                "paper_test": "paper_test.jsonl",
            },
        ),
        FeverConfig(
            name="v2.0",
            version=datasets.Version("2.0.0"),
            description=textwrap.dedent(
                "FEVER  v2.0:\n"
                "The FEVER 2.0 Dataset consists of 1174 claims created by the submissions of participants in the "
                "Breaker phase of the 2019 shared task. Participants (Breakers) were tasked with generating "
                "adversarial examples that induce classification errors for the existing systems. Breakers submitted "
                "a dataset of up to 1000 instances with equal number of instances for each of the three classes "
                "(Supported, Refuted NotEnoughInfo). Only novel claims (i.e. not contained in the original FEVER "
                "dataset) were considered as valid entries to the shared task. The submissions were then manually "
                "evaluated for Correctness (grammatical, appropriately labeled and meet the FEVER annotation "
                "guidelines requirements)."
            ),
            homepage="https://fever.ai/dataset/adversarial.html",
            citation=textwrap.dedent(
                """\
                @inproceedings{Thorne19FEVER2,
                    author = {Thorne, James and Vlachos, Andreas and Cocarascu, Oana and Christodoulopoulos, Christos and Mittal, Arpit},
                    title = {The {FEVER2.0} Shared Task},
                    booktitle = {Proceedings of the Second Workshop on {Fact Extraction and VERification (FEVER)}},
                    year = {2018}
                }"""
            ),
            base_url="https://fever.ai/download/fever2.0",
            urls={
                datasets.Split.VALIDATION: "fever2-fixers-dev.jsonl",
            },
        ),
        FeverConfig(
            name="wiki_pages",
            version=datasets.Version("1.0.0"),
            description=textwrap.dedent(
                "Wikipedia pages for FEVER v1.0:\n"
                "FEVER (Fact Extraction and VERification) consists of 185,445 claims generated by altering sentences "
                "extracted from Wikipedia and subsequently verified without knowledge of the sentence they were "
                "derived from. The claims are classified as Supported, Refuted or NotEnoughInfo. For the first two "
                "classes, the annotators also recorded the sentence(s) forming the necessary evidence for their "
                "judgment."
            ),
            homepage="https://fever.ai/dataset/fever.html",
            citation=textwrap.dedent(
                """\
                @inproceedings{Thorne18Fever,
                    author = {Thorne, James and Vlachos, Andreas and Christodoulopoulos, Christos and Mittal, Arpit},
                    title = {{FEVER}: a Large-scale Dataset for Fact Extraction and {VERification}},
                    booktitle = {NAACL-HLT},
                    year = {2018}
                }"""
            ),
            base_url="https://fever.ai/download/fever",
            urls={
                "wikipedia_pages": "wiki-pages.zip",
            },
        ),
    ]

    def _info(self):
        if self.config.name == "wiki_pages":
            features = {
                "id": datasets.Value("string"),
                "text": datasets.Value("string"),
                "lines": datasets.Value("string"),
            }
        elif self.config.name == "v1.0" or self.config.name == "v2.0":
            features = {
                "id": datasets.Value("int32"),
                "label": datasets.ClassLabel(names=["REFUTES", "SUPPORTS"]),
                "claim": datasets.Value("string"),
                "evidence_annotation_id": datasets.Value("int32"),
                "evidence_id": datasets.Value("int32"),
                "evidence_wiki_url": datasets.Value("string"),
                "evidence_sentence_id": datasets.Value("int32"),
            }
        return datasets.DatasetInfo(
            description=self.config.description,
            features=datasets.Features(features),
            homepage=self.config.homepage,
            citation=self.config.citation,
        )

    def _split_generators(self, dl_manager):
        """Returns SplitGenerators."""
        dl_paths = dl_manager.download_and_extract(self.config.urls)
        return [
            datasets.SplitGenerator(
                name=split,
                gen_kwargs={
                    "filepath": dl_paths[split]
                    if self.config.name != "wiki_pages"
                    else dl_manager.iter_files(os.path.join(dl_paths[split], "wiki-pages")),
                },
            )
            for split in dl_paths.keys()
        ]

    def _generate_examples(self, filepath):
        """Yields examples."""
        if self.config.name == "v1.0" or self.config.name == "v2.0":
            with open(filepath, encoding="utf-8") as f:
                for row_id, row in enumerate(f):
                    data = json.loads(row)
                    id_ = data["id"]
                    label = data.get("label", "")

                    # Drop the examples with label "NOT ENOUGH INFO"
                    if label not in ("REFUTES", "SUPPORTS"):
                        continue

                    claim = data["claim"]
                    evidences = data.get("evidence", [])
                    if len(evidences) > 0:
                        for i in range(len(evidences)):
                            for j in range(len(evidences[i])):
                                annot_id = evidences[i][j][0] if evidences[i][j][0] else -1
                                evidence_id = evidences[i][j][1] if evidences[i][j][1] else -1
                                wiki_url = evidences[i][j][2] if evidences[i][j][2] else ""
                                sent_id = evidences[i][j][3] if evidences[i][j][3] else -1
                                yield str(row_id) + "_" + str(i) + "_" + str(j), {
                                    "id": id_,
                                    "label": label,
                                    "claim": claim,
                                    "evidence_annotation_id": annot_id,
                                    "evidence_id": evidence_id,
                                    "evidence_wiki_url": wiki_url,
                                    "evidence_sentence_id": sent_id,
                                }
                    else:
                        yield row_id, {
                            "id": id_,
                            "label": label,
                            "claim": claim,
                            "evidence_annotation_id": -1,
                            "evidence_id": -1,
                            "evidence_wiki_url": "",
                            "evidence_sentence_id": -1,
                        }
        elif self.config.name == "wiki_pages":
            for file_id, file in enumerate(filepath):
                with open(file, encoding="utf-8") as f:
                    for row_id, row in enumerate(f):
                        data = json.loads(row)
                        yield f"{file_id}_{row_id}", data