Dataset Card for sci-definition

Dataset Summary

This is a dataset of scientific and medical definitions. They are drawn from two sources:

(MedQuAD)[https://github.com/abachaa/MedQuAD] (Wikipedia Science Glossaries)[https://en.wikipedia.org/wiki/Category:Glossaries_of_science]

For both sources, all terms and definitions are formatted as "What is (are) X?" or "Do you have more information about X?" with the answer being the definition of X.

Supported Tasks and Leaderboards

The task this dataset was created for was generating definitions of scientific terms. It has additionally be used to vary the complexity of generating definitions. Both versions of this task are described in the original paper.

Languages

English

Dataset Structure

Data Instances

  {
  "question": "What is (are) complement component 2 deficiency?",
  "category": "medquad",
  "first_sentence": "Complement component 2 deficiency is a disorder that causes the immune system to malfunction, resulting in a form of immunodeficiency.",
  "support_doc_sparse_s2orc":"The occurrence of membranoproliferative glomerulonephritis in a 13 year old boy with inherited complete deficiency of the second component of complement (C2) is described here....<P>...."
  }

Data Fields

question: The question containing the term

category: either medquad or wikipedia

first_sentence: The answer (or definition)

support_doc_sparse_s2orc: the support document, made up of 10 journal abstracts related to the term. Journal abstracts are drawn from S2ORC (https://github.com/allenai/s2orc) using ElasticSearch.

Data Splits

Splits are in a test, train, and dev. As a note, the original results reported in the paper on generating definitions was on the dev set and training was done on a split of the train set. The original test set was reserved for evaluating complexity control.

Dataset Creation

Curation Rationale

We collected from these two sources because they contain expert-written and factually correct information about scientific and medical terms. For the MedQuAD data, we filter for questions of the form: What is (are) X? or Do you have more information on X? We then format the Wikipedia science glossaries in the same form ("What is (are) X?).

Source Data

Initial Data Collection and Normalization

The data was collected by filtering the original MedQuAD dataset and drawing on Wikpiedia science glossaries. This occurred between September 2020-March 2021.

Who are the source language producers?

NIH website (MedQuAD), and Wikpedia editors

Annotations

Annotation process

N/A

Who are the annotators?

N/A

Personal and Sensitive Information

N/A

Considerations for Using the Data

Social Impact of Dataset

The goal of this dataset is to enable a wider audience of readers to understand and engage with scientific writing. A risk, though, is that such attempts might instead widen the gap to accessing scientific information.

Discussion of Biases

The texts in the datasets we train our models on are in General or Academic American English. Many people, especially those who have been historically underrepresented in STEM disciplines and medicine, may not be comfortable with this dialect of English. This risks further alienating the readers we hope to serve.

Other Known Limitations

[Needs More Information]

Additional Information

Dataset Curators