license: cc-by-sa-4.0
task_categories:
- text-generation
- text2text-generation
- summarization
language:
- en
Dataset Card for AbLit
Dataset Description
- Homepage: https://github.com/roemmele/AbLit
- Repository: https://github.com/roemmele/AbLit
- Paper: https://arxiv.org/pdf/2302.06579.pdf
- Point of Contact: [email protected]
Dataset Summary
The AbLit dataset contains abridged versions of 10 classic English literature books, aligned with their original versions on various passage levels. The abridgements were written and made publically available by Emma Laybourn here. This is the first known dataset for NLP research that focuses on the abridgement task.
See the paper for a detailed description of the dataset, as well as the results of several modeling experiments. The GitHub repo also provides more extensive ways to interact with the data beyond what is provided here.
Languages
English
Dataset Structure
Each passage in the original version of a book chapter is aligned with its corresponding passage in the abridged version. These aligned pairs are available for various passage sizes: sentences, paragraphs, and multi-paragraph "chunks". The passage size is specified when loading the dataset. There are train/dev/test splits for items of each size.
| Passage Size | Description | # Train | # Dev | # Test |
|---|---|---|---|---|
| chapters | Each passage is a single chapter | 808 | 10 | 50 |
| sentences | Each passage is a sentence delimited by the NLTK sentence tokenizer | 122,219 | 1,143 | 10,431 |
| paragraphs | Each passage is a paragraph delimited by a line break | 37,227 | 313 | 3,125 |
| chunks-10-sentences | Each passage consists of up to X=10 number of sentences, which may span more than one paragraph. To derive chunks with other lengths X, see GitHub repo above | 14,857 | 141 | 1,264 |
Example Usage
To load aligned paragraphs:
from datasets import load_dataset
data = load_dataset("roemmele/ablit", "paragraphs")
Data Fields
- original: passage text in the original version
- abridged: passage text in the abridged version
- book: title of book containing passage
- chapter: title of chapter containing passage
Dataset Creation
Curation Rationale
Abridgement is the task of making a text easier to understand while preserving its linguistic qualities. Abridgements are different from typical summaries: whereas summaries abstractively describe the original text, abridgements simplify the original primarily through a process of extraction. We present this dataset to promote further research on modeling the abridgement process.
Source Data
The author Emma Laybourn wrote abridged versions of classic English literature books available through Project Gutenberg. She has also provided her abridgements for free on her website. This is how she describes her work: “This is a collection of famous novels which have been shortened and slightly simplified for the general reader. These are not summaries; each is half to two-thirds of the original length. I’ve selected works that people often find daunting because of their density or complexity: the aim is to make them easier to read, while keeping the style intact.”
Initial Data Collection and Normalization
We obtained the original and abridged versions of the books from the respective websites.
Who are the source language producers?
Emma Laybourn
Annotations
Annotation process
We designed a procedure for automatically aligning passages between the original and abridged version of each chapter. We conducted a human evaluation to verify these alignments had high accuracy. The training split of the dataset has ~99% accuracy. The dev and test splits of the dataset were fully human-validated to ensure 100% accuracy. See the paper for further explanation.
Who are the annotators?
The alignment accuracy evaluation was conducted by the authors of the paper, who have expertise in linguistics and NLP.
Personal and Sensitive Information
None
Considerations for Using the Data
Social Impact of Dataset
We hope this dataset will promote more research on the authoring process for producing abridgements, including models for automatically generating abridgements. Because it is a labor-intensive writing task, there are relatively few abridged versions of books. Systems that automatically produce abridgements could vastly expand the number of abridged versions of books and thus increase their readership.
Discussion of Biases
We present this dataset to introduce abridgement as an NLP task, but these abridgements are scoped to one small set of texts associated with a specific domain and author. There are significant practical reasons for this limited scope. In particular, in constrast to the books in AbLit, most recently published books are not included in publicly accessible datasets due to copyright restrictions, and the same restrictions typically apply to any abridgements of these books. For this reason, AbLit consists of British English literature from the 18th and 19th centuries. Some of the linguistic properties of these original books do not generalize to other types of English texts that would be beneficial to abridge. Moreover, the narrow cultural perspective reflected in these books is certainly not representative of the diverse modern population. Readers may find some content offensive.
Dataset Curators
The curators are the authors of the paper.
Licensing Information
cc-by-sa-4.0
Citation Information
Roemmele, Melissa, Kyle Shaffer, Katrina Olsen, Yiyi Wang, and Steve DeNeefe. "AbLit: A Resource for Analyzing and Generating Abridged Versions of English Literature." Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume (2023).