{"slug": "symbolic-augmentation-closes-a-canonical-equivalence-blind-spot-in-neural-fact", "title": "Symbolic Augmentation Closes a Canonical-Equivalence Blind Spot in Neural Fact-Checkers", "summary": "A new study from arXiv introduces Symbolic Augmentation, a training-time framework that generates label-preserving augmented data to fix a structural blind spot in neural fact-checkers: their accuracy on canonical-equivalent rewrites of physically equivalent quantities collapses to 36.5%. The method lifts robustness to 98.2% while slightly improving in-distribution macro-F1 from 0.899 to 0.902, and the augmented encoder matches a closed-frontier LLM at no inference cost.", "body_md": "arXiv:2607.16212v1 Announce Type: new\nAbstract: Large language models hallucinate numbers and units when summarizing scientific text, a failure mode that can silently invert a scientific claim. We recast the detection of such errors as typed verification: we introduce a five-class typed-quantity error taxonomy and a 1500-item benchmark, rewritten from PMC and arXiv sources and labeled by two independent LLM annotators with adjudication (Krippendorff's alpha = 0.882). A ModernBERT encoder fine-tuned on this benchmark reaches macro-F1 = 0.899, far above any off-the-shelf neural fact-checker, yet four probes expose a sharp structural blind spot: on canonical-equivalent rewrites of physically equivalent quantities (e.g., 95{\\deg}C and 368.15 K) its accuracy collapses to 36.5%. We propose Symbolic Augmentation, a training-time framework that runs the modules of a symbolic verifier in reverse to generate label-preserving augmented training data. The augmentation lifts canonical-equivalence robustness to 98.2% while slightly improving in-distribution accuracy (macro-F1: 0.899 to 0.902); the augmented encoder matches a closed-frontier LLM at no inference cost and transfers to an external benchmark (SciFact-Open binary macro-F1: 0.791 to 0.828). Two negative results sharpen the claim: symbolic features as auxiliary encoder inputs add nothing, and symbolic silver labels scale negatively under teacher noise. Together these results identify training-time augmentation as the right integration point between symbolic and learned components.", "url": "https://wpnews.pro/news/symbolic-augmentation-closes-a-canonical-equivalence-blind-spot-in-neural-fact", "canonical_source": "https://arxiv.org/abs/2607.16212", "published_at": "2026-07-21 04:00:00+00:00", "updated_at": "2026-07-21 04:07:32.352588+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-research", "natural-language-processing", "ai-safety"], "entities": ["arXiv", "ModernBERT", "SciFact-Open"], "alternates": {"html": "https://wpnews.pro/news/symbolic-augmentation-closes-a-canonical-equivalence-blind-spot-in-neural-fact", "markdown": "https://wpnews.pro/news/symbolic-augmentation-closes-a-canonical-equivalence-blind-spot-in-neural-fact.md", "text": "https://wpnews.pro/news/symbolic-augmentation-closes-a-canonical-equivalence-blind-spot-in-neural-fact.txt", "jsonld": "https://wpnews.pro/news/symbolic-augmentation-closes-a-canonical-equivalence-blind-spot-in-neural-fact.jsonld"}}