cochrane-simplificationSimplification

cochrane-simplification

Cochrane is an English dataset for paragraph-level simplification of medical texts. Cochrane is a database of systematic reviews of clinical questions, many of which have summaries in plain English targeting readers without a university education. The dataset comprises about 4,500 of such pairs.

You can load the dataset via:

import datasets
  data = datasets.load_dataset('GEM/cochrane-simplification')

The data loader can be found here.

website

Link

paper

Link

authors

Ashwin Devaraj (The University of Texas at Austin), Iain J. Marshall (King's College London), Byron C. Wallace (Northeastern University), Junyi Jessy Li (The University of Texas at Austin)

Quick-Use

Contact Name

If known, provide the name of at least one person the reader can contact for questions about the dataset.

Ashwin Devaraj

Multilingual?

Is the dataset multilingual?

no

Covered Languages

What languages/dialects are covered in the dataset?

English

License

What is the license of the dataset?

cc-by-4.0: Creative Commons Attribution 4.0 International

Communicative Goal

Provide a short description of the communicative goal of a model trained for this task on this dataset.

A model trained on this dataset can be used to simplify medical texts to make them more accessible to readers without medical expertise.

Additional Annotations?

Does the dataset have additional annotations for each instance?

none

Contains PII?

Does the source language data likely contain Personal Identifying Information about the data creators or subjects?

yes/very likely

Dataset Overview

  • Where to find the Data and its Documentation

  • Languages and Intended Use

  • Credit

  • Dataset Structure

Where to find the Data and its Documentation

Webpage

What is the webpage for the dataset (if it exists)?

Link

Download

What is the link to where the original dataset is hosted?

Link

Paper

What is the link to the paper describing the dataset (open access preferred)?

Link

BibTex

Provide the BibTex-formatted reference for the dataset. Please use the correct published version (ACL anthology, etc.) instead of google scholar created Bibtex.

@inproceedings{devaraj-etal-2021-paragraph,
      title = "Paragraph-level Simplification of Medical Texts",
      author = "Devaraj, Ashwin  and
        Marshall, Iain  and
        Wallace, Byron  and
        Li, Junyi Jessy",
      booktitle = "Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
      month = jun,
      year = "2021",
      address = "Online",
      publisher = "Association for Computational Linguistics",
      url = "https://aclanthology.org/2021.naacl-main.395",
      doi = "10.18653/v1/2021.naacl-main.395",
      pages = "4972--4984",
  }
Contact Name

If known, provide the name of at least one person the reader can contact for questions about the dataset.

Ashwin Devaraj

Contact Email

If known, provide the email of at least one person the reader can contact for questions about the dataset.

ashwin.devaraj@utexas.edu

Has a Leaderboard?

Does the dataset have an active leaderboard?

no

Languages and Intended Use

Multilingual?

Is the dataset multilingual?

no

Covered Languages

What languages/dialects are covered in the dataset?

English

License

What is the license of the dataset?

cc-by-4.0: Creative Commons Attribution 4.0 International

Intended Use

What is the intended use of the dataset?

The intended use of this dataset is to train models that simplify medical text at the paragraph level so that it may be more accessible to the lay reader.

Primary Task

What primary task does the dataset support?

Simplification

Communicative Goal

Provide a short description of the communicative goal of a model trained for this task on this dataset.

A model trained on this dataset can be used to simplify medical texts to make them more accessible to readers without medical expertise.

Credit

Curation Organization Type(s)

In what kind of organization did the dataset curation happen?

academic

Curation Organization(s)

Name the organization(s).

The University of Texas at Austin, King's College London, Northeastern University

Dataset Creators

Who created the original dataset? List the people involved in collecting the dataset and their affiliation(s).

Ashwin Devaraj (The University of Texas at Austin), Iain J. Marshall (King's College London), Byron C. Wallace (Northeastern University), Junyi Jessy Li (The University of Texas at Austin)

Funding

Who funded the data creation?

National Institutes of Health (NIH) grant R01-LM012086, National Science Foundation (NSF) grant IIS-1850153, Texas Advanced Computing Center (TACC) computational resources

Who added the Dataset to GEM?

Who contributed to the data card and adding the dataset to GEM? List the people+affiliations involved in creating this data card and who helped integrate this dataset into GEM.

Ashwin Devaraj (The University of Texas at Austin)

Dataset Structure

Data Fields

List and describe the fields present in the dataset.

  • gem_id: string, a unique identifier for the example
  • doi: string, DOI identifier for the Cochrane review from which the example was generated
  • source: string, an excerpt from an abstract of a Cochrane review
  • target: string, an excerpt from the plain-language summary of a Cochrane review that roughly aligns with the source text
Example Instance

Provide a JSON formatted example of a typical instance in the dataset.

{
      "gem_id": "gem-cochrane-simplification-train-766",
      "doi": "10.1002/14651858.CD002173.pub2",
      "source": "Of 3500 titles retrieved from the literature, 24 papers reporting on 23 studies could be included in the review. The studies were published between 1970 and 1997 and together included 1026 participants. Most were cross-over studies. Few studies provided sufficient information to judge the concealment of allocation. Four studies provided results for the percentage of symptom-free days. Pooling the results did not reveal a statistically significant difference between sodium cromoglycate and placebo. For the other pooled outcomes, most of the symptom-related outcomes and bronchodilator use showed statistically significant results, but treatment effects were small. Considering the confidence intervals of the outcome measures, a clinically relevant effect of sodium cromoglycate cannot be excluded. The funnel plot showed an under-representation of small studies with negative results, suggesting publication bias. There is insufficient evidence to be sure about the efficacy of sodium cromoglycate over placebo. Publication bias is likely to have overestimated the beneficial effects of sodium cromoglycate as maintenance therapy in childhood asthma.",
      "target": "In this review we aimed to determine whether there is evidence for the effectiveness of inhaled sodium cromoglycate as maintenance treatment in children with chronic asthma. Most of the studies were carried out in small groups of patients. Furthermore, we suspect that not all studies undertaken have been published. The results show that there is insufficient evidence to be sure about the beneficial effect of sodium cromoglycate compared to placebo. However, for several outcome measures the results favoured sodium cromoglycate."
  }
Data Splits

Describe and name the splits in the dataset if there are more than one.

  • train: 3568 examples
  • validation: 411 examples
  • test: 480 examples

Dataset in GEM

  • Rationale for Inclusion in GEM

  • GEM-Specific Curation

  • Getting Started with the Task

Rationale for Inclusion in GEM

Why is the Dataset in GEM?

What does this dataset contribute toward better generation evaluation and why is it part of GEM?

This dataset is the first paragraph-level simplification dataset published (as prior work had primarily focused on simplifying individual sentences). Furthermore, this dataset is in the medical domain, which is an especially useful domain for text simplification.

Similar Datasets

Do other datasets for the high level task exist?

no

Ability that the Dataset measures

What aspect of model ability can be measured with this dataset?

This dataset measures the ability for a model to simplify paragraphs of medical text through the omission non-salient information and simplification of medical jargon.

GEM-Specific Curation

Modificatied for GEM?

Has the GEM version of the dataset been modified in any way (data, processing, splits) from the original curated data?

no

Additional Splits?

Does GEM provide additional splits to the dataset?

no

Getting Started with the Task

Previous Results

  • Previous Results

Previous Results

Measured Model Abilities

What aspect of model ability can be measured with this dataset?

This dataset measures the ability for a model to simplify paragraphs of medical text through the omission non-salient information and simplification of medical jargon.

Metrics

What metrics are typically used for this task?

Other: Other Metrics, BLEU

Other Metrics

Definitions of other metrics

SARI measures the quality of text simplification

Previous results available?

Are previous results available?

yes

Relevant Previous Results

What are the most relevant previous results for this task/dataset?

The paper which introduced this dataset trained BART models (pretrained on XSum) with unlikelihood training to produce simplification models achieving maximum SARI and BLEU scores of 40 and 43 respectively.

Dataset Curation

  • Original Curation

  • Language Data

  • Structured Annotations

  • Consent

  • Private Identifying Information (PII)

  • Maintenance

Original Curation

Sourced from Different Sources

Is the dataset aggregated from different data sources?

no

Language Data

Data Validation

Was the text validated by a different worker or a data curator?

not validated

Was Data Filtered?

Were text instances selected or filtered?

not filtered

Structured Annotations

Additional Annotations?

Does the dataset have additional annotations for each instance?

none

Annotation Service?

Was an annotation service used?

no

Consent

Any Consent Policy?

Was there a consent policy involved when gathering the data?

no

Private Identifying Information (PII)

Contains PII?

Does the source language data likely contain Personal Identifying Information about the data creators or subjects?

yes/very likely

Any PII Identification?

Did the curators use any automatic/manual method to identify PII in the dataset?

no identification

Maintenance

Any Maintenance Plan?

Does the original dataset have a maintenance plan?

no

Broader Social Context

  • Previous Work on the Social Impact of the Dataset

  • Impact on Under-Served Communities

  • Discussion of Biases

Previous Work on the Social Impact of the Dataset

Usage of Models based on the Data

Are you aware of cases where models trained on the task featured in this dataset ore related tasks have been used in automated systems?

no

Impact on Under-Served Communities

Addresses needs of underserved Communities?

Does this dataset address the needs of communities that are traditionally underserved in language technology, and particularly language generation technology? Communities may be underserved for exemple because their language, language variety, or social or geographical context is underepresented in NLP and NLG resources (datasets and models).

yes

Details on how Dataset Addresses the Needs

Describe how this dataset addresses the needs of underserved communities.

This dataset can be used to simplify medical texts that may otherwise be inaccessible to those without medical training.

Discussion of Biases

Any Documented Social Biases?

Are there documented social biases in the dataset? Biases in this context are variations in the ways members of different social categories are represented that can have harmful downstream consequences for members of the more disadvantaged group.

unsure

Are the Language Producers Representative of the Language?

Does the distribution of language producers in the dataset accurately represent the full distribution of speakers of the language world-wide? If not, how does it differ?

The dataset was generated from abstracts and plain-language summaries of medical literature reviews that were written by medical professionals and thus does was not generated by people representative of the entire English-speaking population.

Considerations for Using the Data

  • PII Risks and Liability

  • Licenses

  • Known Technical Limitations

PII Risks and Liability

Licenses

Known Technical Limitations

Technical Limitations

Describe any known technical limitations, such as spurrious correlations, train/test overlap, annotation biases, or mis-annotations, and cite the works that first identified these limitations when possible.

The main limitation of this dataset is that the information alignment between the abstract and plain-language summary is often rough, so the plain-language summary may contain information that isn't found in the abstract. Furthermore, the plain-language targets often contain formulaic statements like "this evidence is current to [month][year]" not found in the abstracts. Another limitation is that some plain-language summaries do not simplify the technical abstracts very much and still contain medical jargon.

Unsuited Applications

When using a model trained on this dataset in a setting where users or the public may interact with its predictions, what are some pitfalls to look out for? In particular, describe some applications of the general task featured in this dataset that its curation or properties make it less suitable for.

The main pitfall to look out for is errors in factuality. Simplification work so far has not placed a strong emphasis on the logical fidelity of model generations with the input text, and the paper introducing this dataset does not explore modeling techniques to combat this. These kinds of errors are especially pernicious in the medical domain, and the models introduced in the paper do occasionally alter entities like disease and medication names.