Symbolic Augmentation Closes a Canonical-Equivalence Blind Spot in Neural Fact-Checkers

Explain Like I'm Normal

Researchers found that even powerful models like ModernBERT fail to recognize that physically identical quantities are the same when written in different units, leading to massive fact-checking errors. By introducing a new taxonomy and symbolic augmentation, they've created a way to force models to understand units rather than just guessing numbers. This significantly reduces hallucinations in scientific and technical summarization.