PhD in Computer Science
Postdoc @ University of Copenhagen
Supporting Faithful Reporting on Scientific Research with AI Writing Assistant
Science reporting is not an easy task due to the discrepancy between scientific jargon and lay terms, as well as a discrepancy between the language of scientific papers and associated news articles. As such, not all scientific communication accurately conveys the original information, which is exemplified by skewed reporting of less technical topics and unfaithful reporting of scientific findings. To compound this problem, the average amount of time journalists can spend on individual articles has decreased due to funding cuts, lack of space, and increased commercialization. At the same time, the public relies on the media to learn about new scientific findings, and media portrayal of science affects people’s trust in science while at the same time influencing their future actions [7,26,27].
My project proposes to develop natural language processing (NLP) tools to support journalists in faithfully reporting on scientific findings, namely, tools for extracting key findings from scientific articles,, translating scientific jargon into lay language, and generating summaries of scientific articles in multiple languages while avoiding distortions of scientific findings.
In two recent studies which I led [20,21], we investigated automatically detecting exaggeration in health science press releases as well as general information change between science reporting and scientific papers, and found that large pre-trained language models can be successfully exploited for these tasks. This project will leverage my previous research and will be much more ambitious, focusing on: 1) detecting distortions between news articles and scientific articles in different languages and across multiple areas of science; 2) using a model which can detect such distortions to automatically generate more faithful news articles; 3) analyzing texts in the difficult domains of medicine, biology, psychology, and computer science research, which I have worked with previously and which garner some of the most media attention. This will result in trained models which can be used as writing assistants for journalists, helping to improve the quality of scientific reporting and information available to the public. In addition, the project will involve international collaboration with the University of Michigan, including a research stay in order to leverage their expertise and resources, as well as develop my competencies as a researcher.