Improved Detection of Statistical Entities


In this work, we perform an in-depth analysis of the statements made in the political discourse that are natural candidates for fact-checking. In particular, we focus on statements about statistical entities: these entities of public interest which we can measure periodically and are subject to change, such as unemployment. Such statements can potentially be easily verified by checking statistical data released by institutes such as EuroStat or INSEE. A first task toward the automated fact-checking of statistical statements is detecting them automatically. We created an annotated dataset and we evaluated a small state-of-the-art language model on the task of detecting mentions of statistical entities.