Datasets:

tasksource
/

bioscope-spans

license: cc-by-2.0
language:
  - en
tags:
  - tasksource
  - extraction
configs:
  - config_name: abstracts
    data_files:
      - split: train
        path: abstracts/train.parquet
      - split: validation
        path: abstracts/validation.parquet
      - split: test
        path: abstracts/test.parquet
  - config_name: papers
    data_files:
      - split: train
        path: papers/train.parquet

BioScope negation and speculation spans

Expert annotations of negation and speculation cues and their scopes in biomedical abstracts and full papers (the clinical part, whose text is masked, is left out). Every sentence asks for the negation and speculation cues and the scope of each cue. BioScope has no native split: abstracts gets a deterministic document-disjoint 80/10/10 split (crc32(DocID) % 100); papers has only 9 documents and stays train-only.

Data-only mirror of the original distribution (SHA256 b20b525506f7), documented by bioscope.py. Reproduce with scripts/repackage_dataset/span_structure.py; provenance.json records hashes, settings and counts. License: cc-by-2.0 (evidence). Public biological abstracts and papers only; article texts retain their original terms. Clinical masked text excluded.

Schema: context, question, answers (text, answer_start: every span, not alternatives) and JSON-string metadata (field, sentence group). Used by the extraction config of tasksource-jev-typed-decisions.

Citation

Please cite the original dataset:

@article{vincze2008bioscope,
    title = {The {BioScope} corpus: biomedical texts annotated for uncertainty, negation and their scopes},
    author = {Vincze, Veronika and Szarvas, Gy{\"o}rgy and Farkas, Rich{\'a}rd and M{\'o}ra, Gy{\"o}rgy and Csirik, J{\'a}nos},
    journal = {BMC Bioinformatics},
    volume = {9},
    number = {Suppl 11},
    pages = {S9},
    year = {2008},
    url = {https://doi.org/10.1186/1471-2105-9-S11-S9},
}

Citing tasksource

If you use tasksource, its standardized task annotations, or datasets derived from them in published research, please cite the tasksource paper (Sileo, LREC-COLING 2024).

@inproceedings{sileo-2024-tasksource,
    title = "tasksource: A Large Collection of {NLP} tasks with a Structured Dataset Preprocessing Framework",
    author = "Sileo, Damien",
    booktitle = "Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)",
    month = may,
    year = "2024",
    address = "Torino, Italia",
    publisher = "ELRA and ICCL",
    url = "https://aclanthology.org/2024.lrec-main.1361",
    doi = "10.63317/2h4pkr685mkq",
    pages = "15655--15684",
}