Summarising unreliable data

Abstract

Unreliable data is present in datasets, and is either ignored, acknowledged ad hoc, or undetected. This paper discusses data quality issues with a potential framework in mind to deal with them. Such a framework should be applied within data-to-text systems at the generation of text rather than being an afterthought. This paper also shows ways to express uncertainty through language and World Health Organisation (WHO) corpus studies, and an experiment which analyses how subjects approached summarising data with data quality issues. This work is still ongoing.