# Lint as: python3 """CURRICULUM Benchmark""" import json import os import datasets logger = datasets.logging.get_logger(__name__) _CITATION = """\ @misc{https://doi.org/10.48550/arxiv.2204.06283, doi = {10.48550/ARXIV.2204.06283}, url = {https://arxiv.org/abs/2204.06283}, author = {Chen, Zeming and Gao, Qiyue}, keywords = {Computation and Language (cs.CL), Artificial Intelligence (cs.AI), FOS: Computer and information sciences, FOS: Computer and information sciences}, title = {Curriculum: A Broad-Coverage Benchmark for Linguistic Phenomena in Natural Language Understanding}, publisher = {arXiv}, year = {2022}, copyright = {Creative Commons Attribution 4.0 International} } """ _DESCRIPTION = """\ We introduce Curriculum as a new format of NLI benchmark for evaluation of broad-coverage linguistic phenomena. Curriculum contains a collection of datasets that covers 36 types of major linguistic phenomena and an evaluation procedure for diagnosing how well a language model captures reasoning skills for distinct types of linguistic phenomena. We show that this linguistic-phenomena-driven benchmark can serve as an effective tool for diagnosing model behavior and verifying model learning quality. """ _HOMEPAGE = "https://github.com/eric11eca/curriculum-ling" _LICENSE = "CC BY-SA 3.0" _URL = "https://github.com/eric11eca/curriculum-ling/blob/main/benchmark/tasks/" _DESCRIPTION_MAP = { "analytic": "analytical thinking.", "atomic": "reasoning on commonsense knowledge graph.", } _TAKS_NAMES = ["analytic", "defeasible", "boolean", "comparative", "conditional", "context_align", "control", "coreference", "cosmoqa", "counterfactual", "counting", "drop", "entailment_tree", "ester", "hellaswag", "hypernymy", "hyponymy", "kg_relations", "lexical", "logiqa", "monotonicity_infer", "negation", "ner", "physicalqa", "puns", "quantifier", "sentiment", "socialqa", "spatial", "sprl", "syntactic_alternation", "syntactic_variation", "temporal", "transitive", "verbcorner", "verbnet"] task_label_dict = { "lexical": ["entailed", "not-entailed"], "transitive": ["entailed", "not-entailed"], "hypernymy": ["entailed", "not-entailed"], "hyponymy": ["entailed", "not-entailed"], "ner": ["entailed", "not-entailed"], "verbnet": ["entailed", "not-entailed"], "verbcorner": ["entailed", "not-entailed"], "syntactic_alternation": ["entailed", "not-entailed"], "syntactic_variation": ["entailed", "not-entailed"], "boolean": ["entailment", "contradiction", "neutral"], "comparative": ["entailment", "contradiction", "neutral"], "conditional": ["entailment", "contradiction", "neutral"], "counting": ["entailment", "contradiction", "neutral"], "negation": ["entailment", "contradiction", "neutral"], "quantifier": ["entailment", "contradiction", "neutral"], "monotonicity_infer": ["entailed", "not-entailed"], "sentiment": ["entailed", "not-entailed"], "kg_relations": ["entailed", "not-entailed"], "puns": ["entailed", "not-entailed"], "coreference": ["entailed", "not-entailed"], "context_align": ["entailed", "not-entailed"], "sprl": ["entailed", "not-entailed"], "analytic": ["entailed", "not-entailed"], "entailment_tree": ["entailed", "not-entailed"], "socialqa": ["entailed", "not-entailed"], "physicalqa": ["entailed", "not-entailed"], "hellaswag": ["entailed", "not-entailed"], "cosmoqa": ["entailed", "not-entailed"], "logiqa": ["entailed", "not-entailed"], "ester": ["entailed", "not-entailed"], "drop": ["entailed", "not-entailed"], "control": ["entailment", "contradiction", "neutral"], "spatial": ["entailed", "not-entailed"], "temporal": ["entailed", "not-entailed"], "defeasible": ["entailed", "not-entailed"], "counterfactual": ["entailed", "not-entailed"] } def read_file(path, mode="r", **kwargs): with open(path, mode=mode, **kwargs) as f: return f.read() def write_file(data, path, mode="w", **kwargs): with open(path, mode=mode, **kwargs) as f: f.write(data) def read_json(path, mode="r", **kwargs): return json.loads(read_file(path, mode=mode, **kwargs)) def write_json(data, path): return write_file(json.dumps(data, indent=2), path) def read_jsonl(path, mode="r", **kwargs): # Manually open because .splitlines is different from iterating over lines ls = [] with open(path, mode, **kwargs) as f: for line in f: ls.append(json.loads(line)) return ls def write_jsonl(data, path): assert isinstance(data, list) lines = [to_jsonl(elem) for elem in data] write_file("\n".join(lines), path) def to_jsonl(data): return json.dumps(data).replace("\n", "") class CurriculumConfig(datasets.BuilderConfig): """BuilderConfig for Curriculum.""" def __init__(self, features, data_url, citation, url, label_classes=["entailed", "not-entailed"], **kwargs): """BuilderConfig for Curriculum. 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: # 1.0.0: Initial version. super(CurriculumConfig, self).__init__( version=datasets.Version("1.0.0"), **kwargs) self.features = features self.label_classes = label_classes self.data_url = data_url self.citation = citation self.url = url class CurriculumBenchmark(datasets.GeneratorBasedBuilder): """Curriculum Benchmark. Version 1.0.0""" BUILDER_CONFIGS = [ CurriculumConfig( name=task_name, description=_DESCRIPTION, label_classes=task_label_dict[task_name], features=["premise", "hypothesis", "idx", "gold_label"], data_url=f"https://github.com/eric11eca/curriculum-ling/raw/main/benchmark/tasks/{task_name}.zip", citation=_CITATION, url="https://github.com/eric11eca/curriculum-ling/", ) for task_name in _TAKS_NAMES ] def _info(self): features = {feature: datasets.Value( "string") for feature in self.config.features} return datasets.DatasetInfo( description=_DESCRIPTION, features=datasets.Features(features), supervised_keys=None, homepage=_HOMEPAGE, citation=_CITATION, ) @staticmethod def _get_filepath(dl_dir, split): return os.path.join(dl_dir, split + ".jsonl") def _split_generators(self, dl_manager): dl_dir = dl_manager.download_and_extract(self.config.data_url) or "" task_name = _get_task_name_from_data_url(self.config.data_url) dl_dir = os.path.join(dl_dir, task_name) return [ datasets.SplitGenerator( name=datasets.Split.TRAIN, gen_kwargs={ "data_file": os.path.join(dl_dir, "train.jsonl"), "split": datasets.Split.TRAIN, }, ), datasets.SplitGenerator( name=datasets.Split.VALIDATION, gen_kwargs={ "data_file": os.path.join(dl_dir, "val.jsonl"), "split": datasets.Split.VALIDATION, }, ) ] def _generate_examples(self, data_file, split): """This function returns the examples in the raw (text) form.""" logger.info("generating examples from = %s", data_file) dataset = read_jsonl(data_file) for id_, data in enumerate(dataset): yield id_, { "premise": data["premise"], "hypothesis": data["hypothesis"], "gold_label": data["gold_label"], "idx": id_ } def _get_task_name_from_data_url(data_url): return data_url.split("/")[-1].split(".")[0]