Integrating Causal Reasoning into Automated Fact-Checking

Fact-checking is more useful when it can explain the chain of events behind a verdict.

Integrating Causal Reasoning into Automated Fact-Checking

A claim can sound convincing while relying on a flawed cause-and-effect chain. Yet automated fact-checking systems often compare claims and evidence without explicitly reasoning over how events relate.

In this paper, we introduce a pipeline that identifies whether events cause, prevent, enable, or intend other events, then applies reasoning rules to detect logical alignment, contradiction, causal loops, and cherry-picking.

The approach provides an interpretable layer alongside verdict prediction, tested on two fact-checking benchmarks:

On a manually curated set of causal cases, the system reaches an F1-score of 0.50.

Rather than treating fact-checking as a black-box prediction task, causal reasoning helps make the logic behind a verdict visible and inspectable.