EMNLP

2023

ElabQUD

Elaborative Simplification as Implicit Questions Under Discussion

Yating Wu†, William Sheffield†, Kyle Mahowald, Junyi Jessy Li

Conference on Empirical Methods in Natural Language Processing (EMNLP) · 2023

† Equal contribution.

paperCode and dataBibTeX

Overview

Editors sometimes add an explanation when simplifying an article. ElabQUD asks which implicit question that explanation answers, and which earlier words prompted it. The dataset covers 1,299 elaborations in simplified news articles.

We use these annotations to study what editors explain. We also generate elaborations in two stages: first a question, then its answer. This improves the generated explanations in our experiments.

Paper

The reason behind a skills gap

New jobs in New Haven require skills that many local young people lack. In the simplified article, the editor adds an explanation.

Implicit question

Why don’t people acquire the necessary skills?

Sentence added by the editor

Many do not have the money to get the training they need.
Source

Figure 1, page 1; discussion on page 2. Question and added sentence quoted from Figure 1. The article example is from Srikanth and Li (2021); context paraphrased.

Question–target annotations

There are 2,878 question–target annotations for 1,299 elaborations. An elaboration may have more than one annotation.

Elaboration
file_num identifies the article; elab_line gives the line containing the added explanation.
Annotation
question is the implicit question. target and target_line locate the earlier text that prompted it; worker_num identifies the annotator.
Articles
Full Newsela articles require separate access. The data README links the source-text access route.
Read the TSVPython

Requires pandas.

import pandas as pd

revision = "8a253a8f014c642d5156333313f9d1573a031767"
url = (
    "https://raw.githubusercontent.com/sheffwb/elabQUD/"
    f"{revision}/data/targets_and_questions.tsv"
)
annotations = pd.read_csv(url, sep="\t")
print(len(annotations))  # 2878 question–target pairs
print(annotations.groupby(["file_num", "elab_line"]).ngroups)  # 1299
Source

Annotation fields · Released TSV

ElabQUD data
Questions and targets for the Srikanth and Li (2021) data.
Code
Processing and experiment scripts; marked work in progress.

Citation

Download BibTeX
View BibTeX
@inproceedings{elabqud2023,
  title={{Elaborative Simplification as Implicit Questions Under Discussion}},
  author={Wu, Yating and Sheffield, William and Mahowald, Kyle and Li, Junyi Jessy},
  year={2023},
  booktitle={Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing},
  doi={10.18653/v1/2023.emnlp-main.336},
  pages={5525--5537},
  url={https://aclanthology.org/2023.emnlp-main.336/},
  publisher={Association for Computational Linguistics}
}