# Making Science Simple: Corpora for the Lay Summarisation of Scientific Literature

**Authors:** Tomas Goldsack, Zhihao Zhang, Chenghua Lin, Carolina Scarton  
**arXiv:** [2210.09932](https://arxiv.org/abs/2210.09932)  
**Date:** 2022-10-18  
**Summary page:** https://www.summarizepaper.com/en/paper/2210.09932v1/

## Key points

- Lay summarisation simplifies complex texts for non-experts
- Automatic approaches are crucial for broadening access to scientific literature
- Current datasets for lay summarisation are limited in size and scope
- Introduction of new lay summarisation datasets: PLOS (large-scale) and eLife (medium-scale)
- Characterization of lay summaries in the datasets, noting differences in readability and abstractiveness
- Benchmarking of datasets using mainstream summarisation approaches and manual evaluation with domain experts
- Public availability of code and datasets provided by researchers
- Discussion on previous attempts at automatically summarising scientific content for non-experts
- Highlighting limitations in existing datasets and models, emphasizing the need for comprehensive resources like PLOS and eLife

## Summary

In this study, the researchers focus on lay summarisation, which involves summarising and simplifying complex texts to make them more understandable for non-experts. They highlight the importance of automatic approaches for lay summarisation in broadening access to scientific literature and facilitating interdisciplinary knowledge sharing and public understanding of research findings. However, they note that current datasets for this task are limited in size and scope, hindering the development of effective data-driven approaches. To address these limitations, the researchers introduce two new lay summarisation datasets: PLOS (large-scale) and eLife (medium-scale). These datasets contain biomedical journal articles along with expert-written lay summaries. The researchers thoroughly characterize the lay summaries in these datasets, noting differences in readability and abstractiveness that can cater to different application needs. The researchers then benchmark their datasets using mainstream summarisation approaches and conduct a manual evaluation with domain experts. Through this evaluation, they demonstrate the utility of their datasets and shed light on key challenges in the task of lay summarisation. Additionally, they provide their code and datasets for public access. In related work, the researchers discuss previous attempts at automatically summarising scientific content for non-experts. They mention the LaySumm subtask of CL-SciSumm 2020 shared task series as well as other efforts using sources like The Cochrane Database of Systematic Reviews and science news websites. They highlight limitations in existing datasets and models for lay summarisation, emphasizing the need for more comprehensive resources like PLOS and eLife. Overall, this study contributes valuable insights into lay summarisation by introducing new datasets, evaluating existing approaches, and addressing key challenges in making scientific literature more accessible to a wider audience.