Back to AI News
arxiv

Cited but Unused: Parametric Provenance in Retrieval-Augmented Generation

This arXiv paper measures whether citations in retrieval-augmented generation actually reflect retrieved context or instead come from a model's parametric memory. Using counterfactual value substitution, it finds that 1.2% to 15.7% of citable claims across model families derive from parametric memory, with rates rising with model size within a family. Standard natural language inference verification performs no better than chance, even when restricted to the cited passage. Comparing ten attribution methods, the paper finds that the contrast between a model's output distribution with and without context best separates provenance, but the resulting detector remains too inaccurate to act on.

AI Brief

It provides a concrete, cautionary evaluation of citation reliability in RAG systems, showing that standard verification can miss provenance misattribution and that current detectors are not yet actionable.

arXiv:2611.17161v1 Announce Type: new Abstract: Retrieval-augmented generation systems attach citations to attribute claims to their sources. These citations frequently misstate provenance, and standard verification cannot detect this. We measure provenance by counterfactual value substitution: a fact in the retrieved passages is replaced with a false variant and the model re-queried. We find that across five sizes of one model family and a second comparator family, 1.2% to 15.7% of citable claims derive from parametric memory rather than the retrieved context. At the most conservative substitution distance these figures are 4.6% and 8.5% on the two models we tested at it, with the rate rising with model size within a family. We further find that natural language inference verification, the standard approach, gives results indistinguishable from chance across models, even when restricted to the cited passage. We determine that the cause is structural: a model answering from memory selects a citation that agrees with its output. Comparing ten attribution methods on identical claims, the contrast between the model's distribution with and without context best separates provenance. Acting on that signal does not improve provenance honesty at these base rates. The resulting detector is hence still not accurate enough to act on: it measures what the model knew rather than what it used, and separating those remains the open problem.
Be the first to rate this article.

Comments & Questions (0)

Captcha
Please be respectful — let's keep the conversation friendly.

No comments yet

Be the first to comment!