superlim-2.py

import os
import datasets
import pandas as pd

_CITATION = """\
"""

# You can copy an official description
_DESCRIPTION = """\
"""

_HOMEPAGE = ""

_LICENSE = ""

_SUPERLIM_CITATION = """\
Yvonne Adesam, Aleksandrs Berdicevskis, Felix Morger (2020): SwedishGLUE – Towards a Swedish Test Set for Evaluating Natural Language Understanding Models BibTeX
[1] Original Absabank:
Jacobo Rouces, Lars Borin, Nina Tahmasebi (2020): Creating an Annotated Corpus for Aspect-Based Sentiment Analysis in Swedish, in Proceedings of the 5th conference in Digital Humanities in the Nordic Countries, Riga, Latvia, October 21-23, 2020. BibTeX
[2] DaLAJ:
Volodina, Elena, Yousuf Ali Mohammed, and Julia Klezl (2021). DaLAJ - a dataset for linguistic acceptability judgments for Swedish. In Proceedings of the 10th Workshop on Natural Language Processing for Computer Assisted Language Learning (NLP4CALL 2021). Linköping Electronic Conference Proceedings 177:3, s. 28-37. https://ep.liu.se/ecp/177/003/ecp2021177003.pdf
[3] Analogy:
Tosin Adewumi, Foteini Liwicki, Markus Liwicki. (2020). Corpora compared: The case of the Swedish Gigaword & Wikipedia corpora. In: Proceedings of the 8th SLTC, Gothenburg. arXiv preprint arXiv:2011.03281
[4] Swedish Test Set for SemEval 2020 Task 1:
Unsupervised Lexical Semantic Change Detection: Dominik Schlechtweg, Barbara McGillivray, Simon Hengchen, Haim Dubossarsky, Nina Tahmasebi (2020): SemEval-2020 Task 1: Unsupervised Lexical Semantic Change Detection, in Proceedings of the Fourteenth Workshop on Semantic Evaluation (SemEval2020), Barcelona, Spain (Online), December 12, 2020. BibTeX
[5] Winogender:
Saga Hansson, Konstantinos Mavromatakis, Yvonne Adesam, Gerlof Bouma and Dana Dannélls (2021). The Swedish Winogender Dataset. In The 23rd Nordic Conference on Computational Linguistics (NoDaLiDa 2021), Reykjavik.
[6] SuperSim:
Hengchen, Simon and Tahmasebi, Nina (2021). SuperSim: a test set for word similarity and relatedness in Swedish. In The 23rd Nordic Conference on Computational Linguistics (NoDaLiDa 2021), Reykjavik. arXiv preprint arXiv:2014.05228
"""

_SUPERLIM_DESCRIPTION = """\
SuperLim, A standardized suite for evaluation and analysis of Swedish natural language understanding systems.
"""
_ABSABank_imm_DESCRIPTION = """\
Absabank-Imm (where ABSA stands for "Aspect-Based Sentiment Analysis" and Imm for "Immigration") is a subset of the Swedish ABSAbank, created to be a part of the SuperLim collection. In Absabank-Imm, texts and paragraphs are manually labelled according to the sentiment (on 1--5 scale) that the author expresses towards immigration in Sweden (this task is known as aspect-based sentiment analysis or stance analysis). To create Absabank-Imm, the original Absabank has been substantially reformatted, but no changes to the annotation were made. The dataset contains 4872 short texts.
"""
_DaLAJ_DESCRIPTION = """\
Determine whether a sentence is correct Swedish or not.
"""
_DaLAJ_CITATION = """\
[1] Original Absabank:
Jacobo Rouces, Lars Borin, Nina Tahmasebi (2020): Creating an Annotated Corpus for Aspect-Based Sentiment Analysis in Swedish, in Proceedings of the 5th conference in Digital Humanities in the Nordic Countries, Riga, Latvia, October 21-23, 2020. BibTeX
[2] DaLAJ:
Volodina, Elena, Yousuf Ali Mohammed, and Julia Klezl (2021). DaLAJ - a dataset for linguistic acceptability judgments for Swedish. In Proceedings of the 10th Workshop on Natural Language Processing for Computer Assisted Language Learning (NLP4CALL 2021). Linköping Electronic Conference Proceedings 177:3, s. 28-37. https://ep.liu.se/ecp/177/003/ecp2021177003.pdf
"""

_SweAna_DESCRIPTION = """\
The Swedish analogy test set follows the format of the original Google version. However, it is bigger and balanced across the 2 major categories, 
having a total of 20,638 samples, made up of 10,381 semantic and 10,257 syntactic samples. It is also roughly balanced across the syntactic subsections. 
There are 5 semantic subsections and 6 syntactic subsections. The dataset was constructed, partly using the samples in the English version, 
with the help of tools dedicated to Swedish translation and it was proof-read for corrections by two native speakers (with a percentage agreement of 98.93\%)."""
_SweAna_CITATION = """\
[1] Original Absabank:
Jacobo Rouces, Lars Borin, Nina Tahmasebi (2020): Creating an Annotated Corpus for Aspect-Based Sentiment Analysis in Swedish, in Proceedings of the 5th conference in Digital Humanities in the Nordic Countries, Riga, Latvia, October 21-23, 2020. BibTeX
"""

_SweDiag_DESCRIPTION = """\
Färdig preliminär översättning av SuperGLUE diagnostik. Datan innehåller alla ursprungliga annoterade satspar från SuperGLUE tillsammans 
med deras svenska översättningar."""
_SweDiag_CITATION = """\
"""
_SweDN_DESCRIPTION = """\
AbstractThe SWE-DN corpus is based on 1,963,576 news articles from the Swedish newspaper Dagens Nyheter (DN) during the years 2000--2020. The articles are filtered to resemble the CNN/DailyMail dataset both regarding textual structure"""
_SweDiag_CITATION = """\
"""
_SweFaq_DESCRIPTION = """\
Vanliga frågor från svenska myndigheters webbsidor med svar i randomiserad ordning"""
_SweFaq_CITATION = """\
"""
_SweNLI_DESCRIPTION = """\
A textual inference/entailment problem set, derived from FraCas. The original English Fracas [1] was converted to html and edited by Bill MacCartney [2], 
and then automatically translated to Swedish by Peter Ljunglöf and Magdalena Siverbo [3]. The current tabular form of the set was created by Aleksandrs Berdicevskis 
by merging the Swedish and English versions and removing some of the problems. Finally, Lars Borin went through all the translations, correcting and Swedifying them manually. 
As a result, many translations are rather liberal and diverge noticeably from the English original."""
_SweFracas_CITATION = """\
  """
_SwePar_DESCRIPTION = """\
SweParaphrase is a subset of the automatically translated Swedish Semantic Textual Similarity dataset (Isbister and Sahlgren, 2020). 
It consists of 165 manually corrected Swedish sentence pairs paired with the original English sentences and their similarity scores 
ranging between 0 (no meaning overlap) and 5 (meaning equivalence). These scores were taken from the English data, they were assigned 
by Crowdsourcing through Mechanical Turk. Each sentence pair belongs to one genre (e.g. news, forums or captions). 
The task is to determine how similar two sentences are."""
_SwePar_CITATION = """\
"""
_SweSat_DESCRIPTION = """\
The dataset provides a gold standard for Swedish word synonymy/definition. The test items are collected from the Swedish Scholastic 
Aptitude Test (högskoleprovet), currently spanning the years 2006--2021 and 822 vocabulary test items. The task for the tested system 
is to determine which synonym or definition of five alternatives is correct for each test item.
"""
_SweSat_CITATION = """\
"""

_SweSim_DESCRIPTION = """\
SuperSim is a large-scale similarity and relatedness test set for Swedish built with expert human judgments. The test set is composed of 1360 word-pairs independently judged for both relatedness and similarity by five annotators."""

_SweWinogender_DESCRIPTION = """\
The SweWinogender test set is diagnostic dataset to measure gender bias in coreference resolution. It is modelled after the English Winogender benchmark, 
and is released with reference statistics on the distribution of men and women between occupations and the association between gender and occupation in modern corpus material."""

_SweWinograd_DESCRIPTION = """\
SweWinograd is a pronoun resolution test set, containing constructed items in the style of Winograd schema’s. The interpretation of the target pronouns is determined by (common sense) 
reasoning and knowledge, and not by syntactic constraints, lexical distributional information or discourse structuring patterns. 
The dataset contains 90 multiple choice with multiple correct answers test items."""

_SweWic_DESCRIPTION = """\
The Swedish Word-in-Context dataset provides a benchmark for evaluating distributional models of word meaning, in particular context-sensitive/dynamic models. Constructed following the principles of the (English)
 Word-in-Context dataset, SweWiC consists of 1000 sentence pairs, where each sentence in a pair contains an occurence of a potentially ambiguous focus word specific to that pair. The question posed to the tested 
 system is whether these two occurrences represent instances of the same word sense. There are 500 same-sense pairs and 500 different-sense pairs."""

_argumentation_sentences_DESCRIPTION = """\
Argumentation sentences is a translated corpus for the task of identifying stance in relation to a topic. It consists of sentences labeled with pro, con or non in relation to one of six topics. 
The original dataset can be found here https://github.com/trtm/AURC.  The test set is manually corrected translations, the training set is machine translated. """

_argumentation_sentences_DESCRIPTION_CITATION = """\
"""

_RELEASE_VERSION = "2.0.4"
_GH_REPOSITORY = "https://raw.githubusercontent.com/spraakbanken/SuperLim-2/"
_URL = f"{_GH_REPOSITORY}/{_RELEASE_VERSION}/"

_TASKS = {
    "absabank-imm": "absabank-imm",
    "argumentation_sent":"argumentation-sentences",
    "dalaj-ged": "dalaj-ged-superlim",
    "sweana": "sweanalogy",
    "swediagnostics": "swediagnostics",
    "swedn": "swedn",
    "swefaq": "swefaq",
    "swenli": "swenli",
    "swepar": "sweparaphrase",
    "swesat": "swesat-synonyms",
    "swesim_relatedness": "supersim-superlim-relatedness",
    "swesim_similarity": "supersim-superlim-similarity",
    "swewic": "swewic",
    "swewinogender": "swewinogender",
    "swewinograd": "swewinograd"

}

class SuperLimConfig(datasets.BuilderConfig):
    """BuilderConfig for SuperLim."""

    def __init__(self, features, data_url, citation, url, label_classes=("False", "True"), **kwargs):
        """BuilderConfig for SuperLim.

        Args:
        features: `list[string]`, list of the features that will appear in the
            feature dict. Should not include "label".
        data_url: `string`, url to download the zip file from.
        citation: `string`, citation for the data set.
        url: `string`, url for information about the data set.
        label_classes: `list[string]`, the list of classes for the label if the
            label is present as a string. Non-string labels will be cast to either
            'False' or 'True'.
        **kwargs: keyword arguments forwarded to super.
        """
        # Version history:
        # 1.0.2: Fixed non-nondeterminism in ReCoRD.
        # 1.0.1: Change from the pre-release trial version of SuperLim (v1.9) to
        #        the full release (v2.0).
        # 1.0.0: S3 (new shuffling, sharding and slicing mechanism).
        # 0.0.2: Initial version.
        super(SuperLimConfig, self).__init__(version=datasets.Version("2.0.0"), **kwargs)
        self.features = features
        self.label_classes = label_classes
        self.data_url = data_url
        self.citation = citation
        self.url = url

class SuperLim(datasets.GeneratorBasedBuilder):
    """The SuperLim benchmark."""

    VERSION = datasets.Version("2.0.3")

    BUILDER_CONFIGS = [
        datasets.BuilderConfig(name="absabank-imm", version=VERSION, description=_ABSABank_imm_DESCRIPTION),
        datasets.BuilderConfig(name="argumentation_sent", version=VERSION, description=_argumentation_sentences_DESCRIPTION),
        datasets.BuilderConfig(name="dalaj-ged", version=VERSION, description=_DaLAJ_DESCRIPTION),
        datasets.BuilderConfig(name="sweana", version=VERSION, description=_SweAna_DESCRIPTION),
        datasets.BuilderConfig(name="swediagnostics", version=VERSION, description=_SweDiag_DESCRIPTION),
        datasets.BuilderConfig(name="swedn", version=VERSION, description=_SweDN_DESCRIPTION),
        datasets.BuilderConfig(name="swefaq", version=VERSION, description=_SweFaq_DESCRIPTION),
        datasets.BuilderConfig(name="swenli", version=VERSION, description=_SweNLI_DESCRIPTION),
        datasets.BuilderConfig(name="swepar", version=VERSION, description=_SwePar_DESCRIPTION),
        datasets.BuilderConfig(name="swesat", version=VERSION, description=_SweSat_DESCRIPTION),
        datasets.BuilderConfig(name="swesim_relatedness", version=VERSION, description=_SweSim_DESCRIPTION),
        datasets.BuilderConfig(name="swesim_similarity", version=VERSION, description=_SweSim_DESCRIPTION),
        datasets.BuilderConfig(name="swewic", version=VERSION, description=_SweWic_DESCRIPTION),
        datasets.BuilderConfig(name="swewinogender", version=VERSION, description=_SweWinogender_DESCRIPTION),
        datasets.BuilderConfig(name="swewinograd", version=VERSION, description=_SweWinograd_DESCRIPTION)
    ]

    def _info(self):
        # TODO: This method specifies the datasets.DatasetInfo object which contains informations and typings for the dataset
        if self.config.name == 'absabank-imm': # This is the name of the configuration selected in BUILDER_CONFIGS above
            features = datasets.Features({
                "id": datasets.Value("string"),
                "text": datasets.Value("string"),
                "label": datasets.Value(dtype='float32')
            })
        elif self.config.name == 'argumentation_sent':
            features = datasets.Features({
                "sentence_id": datasets.Value("string"),
                "topic": datasets.Value("string"),
                "label": datasets.ClassLabel(num_classes=3, names=['pro', 'con', 'non']),
                "sentence": datasets.Value("string")
            })
        elif self.config.name == "dalaj-ged":
            features = datasets.Features({
                "sentence": datasets.Value("string"),
                "label": datasets.ClassLabel(num_classes=2, names=['correct', 'incorrect']),
                "meta": datasets.Features({
                   'error_span': datasets.Features({
                        'start': datasets.Value(dtype='int64'),
                        'stop': datasets.Value(dtype='int64')
                    }),
                    'confusion_pair': datasets.Features({
                        'incorrect_span': datasets.Value("string"),
                        'correction': datasets.Value('string')
                    }),
                    'error_label': datasets.Value("string"),
                    'education_level': datasets.Value("string"),
                    'l1': datasets.Value("string"),
                    'data_source': datasets.Value("string")
                })
            })
        elif self.config.name == "sweana":
            features = datasets.Features({
                "pair1_element1": datasets.Value("string"),
                "pair1_element2": datasets.Value("string"),
                "pair2_element1": datasets.Value("string"),
                "label": datasets.Value("string"),
                "category": datasets.Value("string"),
            })
        elif self.config.name == 'swediagnostics':
            features = datasets.Features({
                'premise': datasets.Value("string"),
                'hypothesis': datasets.Value("string"),
                'label': datasets.ClassLabel(num_classes=3, names=['entailment', 'contradiction', 'neutral']),
                'meta': datasets.Features({
                    'lexical_semantics': datasets.Value("string"), 
                    'predicate_argument_structure': datasets.Value("string"),
                    'logic': datasets.Value("string"),
                    'knowledge': datasets.Value("string"),
                    'domain': datasets.Value("string")
                })
            })
        elif self.config.name == 'swedn':
            features = datasets.Features({
                "id": datasets.Value("string"),
                "headline": datasets.Value("string"),
                "summary": datasets.Value("string"),
                "article": datasets.Value("string"),
                "article_category": datasets.Value("string")
            })
        elif self.config.name == "swefaq":
            features = datasets.Features({
                "category_id": datasets.Value("string"),
                "candidate_answers": datasets.Sequence(datasets.Value("string")),
                "question": datasets.Value("string"),
                "label": datasets.Value(dtype='int32'),
                "meta": datasets.Features({
                    "category": datasets.Value("string"),
                    "source": datasets.Value("string"),
                    "link": datasets.Value("string"),
                })
            })
        elif self.config.name == 'swenli':
            features = datasets.Features({
                "id": datasets.Value("string"),
                "premise": datasets.Value("string"),
                "hypothesis": datasets.Value("string"),
                "label": datasets.ClassLabel(num_classes=3, names=['entailment', 'contradiction', 'neutral'])
            })
        elif self.config.name == "swepar":
            features = datasets.Features({
                "genre": datasets.Value("string"),
                "file": datasets.Value("string"),
                "sentence_1": datasets.Value("string"),
                "sentence_2": datasets.Value("string"),
                "label": datasets.Value(dtype='float32'),
            })
        elif self.config.name == "swesat":
            features = datasets.Features({
                "id": datasets.Value("string"),
                "item": datasets.Value("string"),
                "candidate_answers": datasets.Sequence(
                    datasets.Value("string"),
                    length=5
                ),
                "label": datasets.ClassLabel(5),
                "meta": datasets.Features({
                    "comment": datasets.Value("string")
                })
            })
        elif self.config.name == "swesim_relatedness":
            features = datasets.Features({
                "word_1": datasets.Value("string"),
                "word_2": datasets.Value("string"),
                "label": datasets.Value(dtype='float32')
            })
        elif self.config.name == "swesim_similarity":
            features = datasets.Features({
                "word_1": datasets.Value("string"),
                "word_2": datasets.Value("string"),
                "label": datasets.Value(dtype='float32')
            })
        elif self.config.name == "swewic":
            features = datasets.Features({
                "idx": datasets.Value(dtype='int32'),
                "first": datasets.Features({
                    "context": datasets.Value("string"),
                    "word": datasets.Features({
                        "location": datasets.Features({
                            "start": datasets.Value(dtype='int32'),
                            "stop": datasets.Value(dtype='int32')
                        }),
                        "text": datasets.Value("string")
                    })
                }),
                "second": datasets.Features({
                    "context": datasets.Value("string"),
                    "word": datasets.Features({
                        "location": datasets.Features({
                            "start": datasets.Value(dtype='int32'),
                            "stop": datasets.Value(dtype='int32')
                        }),
                        "text": datasets.Value("string")
                    })
                }),
                "label": datasets.ClassLabel(num_classes=2, names=['same_sense', 'different_sense']),
                "meta": datasets.Features({
                    "first_source": datasets.Value("string"),
                    "first_sense_id": datasets.Value("string"),
                    "second_source": datasets.Value("string"),
                    "second_sense_id": datasets.Value("string"),
                    "pos": datasets.Value("string")
                })
            })
        elif self.config.name == 'swewinogender':
            features = datasets.Features({
                "idx": datasets.Value(dtype='int32'),
                'premise': datasets.Value("string"),
                'hypothesis': datasets.Value("string"),
                'label': datasets.ClassLabel(num_classes=3, names=['entailment', 'contradiction', 'neutral']),
                'meta': datasets.Features({
                    'tuple_id': datasets.Value("string"),
                    'template_id': datasets.Value("string"),
                    'occupation_participant': datasets.Value("string"),
                    'other_participant': datasets.Value("string"),
                    'pronoun': datasets.Value("string")
                })
            })
        elif self.config.name == 'swewinograd':
            features = datasets.Features({
                "idx": datasets.Value(dtype='int32'),
                'text': datasets.Value("string"),
                'label': datasets.ClassLabel(num_classes=2, names=['not_coreferring', 'coreferring']),
                'pronoun': datasets.Features({
                    'text': datasets.Value("string"),
                    'location': datasets.Features({
                        "start": datasets.Value(dtype='int32'),
                        "stop": datasets.Value(dtype='int32')
                    })
                }),
                'candidate_antecedent': datasets.Features({
                    "text": datasets.Value("string"),
                    'location': datasets.Features({
                        "start": datasets.Value(dtype='int32'),
                        "stop": datasets.Value(dtype='int32')
                    })
                }),
                'meta': datasets.Features({
                    'snippet_id': datasets.Value("string")
                })
            })
        else:
            raise ValueError(f"Subset {self.config.name} does not exist.")
        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  define them above because they are different between the two configurations
            # If there's a common (input, target) tuple from the features, uncomment supervised_keys line below and
            # specify them. They'll be used if as_supervised=True in builder.as_dataset.
            # supervised_keys=("sentence", "label"),
            # 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):
        file_format = 'jsonl'
        splits = []
        DATA_FOLDER = 'supersim-superlim' if self.config.name.startswith('swesim') else _TASKS[self.config.name]
        data_dir_test = dl_manager.download_and_extract(os.path.join(_URL,DATA_FOLDER,f"{_TASKS[self.config.name]}_test.{file_format}"))
        split_test = datasets.SplitGenerator(
            name=datasets.Split.TEST,
            gen_kwargs={
                "filepath": data_dir_test,
                "split": "test"
            },
        )
        splits.append(split_test)
        if self.config.name in ("absabank-imm", "argumentation_sent", "dalaj-ged", "swefaq",
                                "swewic", "swenli", "swedn", "swepar", "swewinograd"):
            data_dir_dev = dl_manager.download_and_extract(os.path.join(_URL,DATA_FOLDER,f"{_TASKS[self.config.name]}_dev.{file_format}"))
            split_dev = datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={
                    "filepath": data_dir_dev,
                    "split": "dev",
                },
            )
            splits.append(split_dev)
        if self.config.name in ("absabank-imm", "argumentation_sent", "dalaj-ged", "swefaq",
                                "swewic", "swenli", "swedn", "swepar", "swesim_relatedness",
                                "swesim_similarity", "swesat", "sweana", "swewinograd"):
            data_dir_train = dl_manager.download_and_extract(os.path.join(_URL,DATA_FOLDER,f"{_TASKS[self.config.name]}_train.{file_format}"))
            split_train = datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "filepath": data_dir_train,
                    "split": "train",
                },
            )
            splits.append(split_train)
        return splits

    def _generate_examples(self, filepath, split):
        # TODO: This method handles input defined in _split_generators to yield (key, example) tuples from the dataset.
        # The `key` is for legacy reasons (tfds) and is not important in itself, but must be unique for each example.
        df = pd.read_json(filepath, lines=True)
        for key, row in df.iterrows():
            if self.config.name == "absabank-imm":
                yield key, {
                    "id": row['id'],
                    "text": row["text"],
                    "label": row["label"],
                }
            elif self.config.name == "argumentation_sent":
                yield key, {
                    "sentence_id": row["sentence_id"],
                    "topic": row["topic"],
                    "label": row["label"],
                    "sentence" : row["sentence"],
                }
            elif self.config.name == "dalaj-ged":
            # Yields examples as (key, example) tuples
                meta = row['meta']
                # Add None values when error span and confusion_pair values are missing.
                if not meta['error_span'] and not meta['confusion_pair']:
                    meta['error_span']['start'] = None
                    meta['error_span']['stop'] = None
                    meta['confusion_pair']['incorrect_span'] = None
                    meta['confusion_pair']['correction'] = None
                yield key, {
                    "sentence": row["sentence"],
                    "label": row["label"],
                    "meta": meta,
                }
            elif self.config.name == "sweana":
                yield key, {
                    "pair1_element1": row["pair1_element1"],
                    "pair1_element2": row["pair1_element2"],
                    "pair2_element1": row["pair2_element1"],
                    "label": row["label"],
                    "category": row["category"],
                }
            elif self.config.name == "swediagnostics":
                yield key, {
                    'premise': row['premise'],
                    'hypothesis': row['hypothesis'],
                    'label': row['label'],
                    'meta': row['meta'],
                }
            elif self.config.name == "swedn":
                yield key, {
                    'id': row['id'],
                    'headline': row['headline'],
                    'summary': row['summary'],
                    'article': row['article'],
                    'article_category': row['article_category']
                }
            elif self.config.name == "swefaq":
                yield key, {
                    "category_id": row['category_id'],
                    "question": row["question"],
                    "candidate_answers": row['candidate_answers'],
                    "label": row["label"],
                    "meta": row['meta'],
                }
            elif self.config.name == "swenli":
                yield key, {
                    'id': row['id'],
                    'premise': row['premise'],
                    'hypothesis': row['hypothesis'],
                    'label': row['label']
                }
            elif self.config.name == "swepar":
                yield key, {
                    "genre": row["genre"],
                    "file": row["file"],
                    "sentence_1": row["sentence_1"],
                    "sentence_2": row["sentence_2"],
                    "label": row["label"],
                }
            elif self.config.name == "swesat":
                yield key, {
                    "id": row["id"],
                    "item": row["item"],
                    "candidate_answers": row["candidate_answers"],
                    "label": row["label"],
                    "meta": row["meta"],
                }
            elif self.config.name == "swesim_relatedness":
                yield key, {
                    "word_1": row["word_1"],
                    "word_2": row["word_2"],
                    "label": row["label"],
                }
            elif self.config.name == "swesim_similarity":
                yield key, {
                    "word_1": row["word_1"],
                    "word_2": row["word_2"],
                    "label": row["label"],
                }
            elif self.config.name == "swewic":
                yield key, {
                    "idx": row["idx"],
                    "first": row["first"],
                    "second": row["second"],
                    "label": row["label"],
                    "meta": row["meta"],
                }
            elif self.config.name == "swewinogender":
                yield key, {
                    "idx": row["idx"],
                    "premise": row["premise"],
                    "hypothesis": row["hypothesis"],
                    "label": row["label"],
                    "meta": row["meta"],
                }
            elif self.config.name == "swewinograd":
                yield key, {
                    "idx": row["idx"],
                    "text": row["text"],
                    "label": row["label"],
                    "pronoun": row["pronoun"],
                    "candidate_antecedent": row["candidate_antecedent"],
                    "meta": row["meta"]
                }
            else:
                raise ValueError(f"Subset {self.config.name} does not exist")