🤖 AI Summary
This study addresses the factual verifiability of Wikipedia content—specifically, whether statements are supported by traceable, publicly accessible citations. To this end, we introduce PeopleProfiles, a fine-grained annotated dataset that systematically reveals a severe evidence disconnect between lead sections and main text (>80% of lead statements lack corresponding in-text citations). We propose a multi-level support annotation framework integrating human-constructed structured evidence chains, cross-paragraph provenance modeling, verification of citation source accessibility, and standardized retrieval benchmarking. Key findings include: (i) ~20% of lead statements receive no in-text support; (ii) 27% of in-text annotated statements lack verifiable, publicly accessible references; and (iii) state-of-the-art retrieval methods fail to recover complex grounding evidence. This work establishes a new benchmark and methodological foundation for Wikipedia credibility assessment and automated fact-checking.
📝 Abstract
Wikipedia is a critical resource for modern NLP, serving as a rich repository of up-to-date and citation-backed information on a wide variety of subjects. The reliability of Wikipedia -- its groundedness in its cited sources -- is vital to this purpose. This work provides a quantitative analysis of the extent to which Wikipedia *is* so grounded and of how readily grounding evidence may be retrieved. To this end, we introduce PeopleProfiles -- a large-scale, multi-level dataset of claim support annotations on Wikipedia articles of notable people. We show that roughly 20% of claims in Wikipedia *lead* sections are unsupported by the article body; roughly 27% of annotated claims in the article *body* are unsupported by their (publicly accessible) cited sources; and>80% of lead claims cannot be traced to these sources via annotated body evidence. Further, we show that recovery of complex grounding evidence for claims that *are* supported remains a challenge for standard retrieval methods.