🧑‍🔬 Michael Schlichtkrull

Michael Schlichtkrull works in natural language processing on automated fact-checking: how systems can find claims worth checking, gather evidence for them, and reach and explain a verdict. Schlichtkrull's papers here treat verification as a chain of connected steps rather than a single classification problem. They draw on both running text and structured sources such as Wikipedia tables and lists, and on shared benchmarks for comparing systems.

The most cited of these papers is A Survey on Automated Fact-Checking (2022), published in TACL. It starts from the point that human checkers need hours or days per statement and cannot match the speed at which claims circulate online. The survey sets out one framework running from claim detection through evidence retrieval to verification, with verification divided into predicting a verdict and producing a justification. It then uses that framework to organise existing datasets and modelling approaches, including the stages of selecting claims and explaining outcomes that earlier overviews had left aside. The second is the FEVEROUS shared task (2021), run at the FEVER 2021 workshop. It asks systems to check human-written claims against a mixture of sentences and table cells and to assign one of three labels: supported, refuted, or not enough information. Under its scoring, a system earns credit for a claim only if it predicts the correct label and also retrieves every item of a gold evidence set. These two works, from 2021 and 2022, are also the most recent ones listed here. Together they move from building a benchmark over mixed evidence to mapping the field as a whole.

Schlichtkrull is currently affiliated with Queen Mary University of London and the University of Cambridge, with earlier papers from the University of Amsterdam and the University of Edinburgh. Co-authors on the papers here include Andreas Vlachos and Zhijiang Guo, as well as James Thorne, Christos Christodoulopoulos, Arpit Mittal and Oana Cocarascu.

Michael Schlichtkrull has 2 papers in our library, in 2021–2022; often with Andreas Vlachos, Zhijiang Guo, Arpit Mittal.