retrieval-summarization-task-pulls-definitions-surface-term-match-general CLM-0136When retrieval for a summarization task pulls definitions by surface term match from a general encyclopedia, ambiguous terms fetch the wrong sense and the model incorporates that unsupported background into its output, lowering claim-level factuality below the no-retrieval baseline even when the prompt tells it to ignore unrelated retrieved content.
Capability: Stating false facts confidently
Observed on
Radiology-report lay summarization with a small general model (Qwen3.5-0.8B) and BioBART-v2-large, retrieval via Wikipedia API first-sentence definitions for terms extracted from t.
Applies when
- Effect on the capability:
- hurts
- Needs:
- documents or an index to retrieve from
- Fails when:
- Retrieval matches on surface terms and fetches the wrong sense of an ambiguous term. Telling the model to ignore unrelated retrieved content does not prevent it incorporating it.
Sources
- One paper, 5,000 held-out radiology reports, nine metrics. FENICE fell versus baseline with RAG for both models in few-shot and fine-tuned settings; authors attribute this to manual inspection of wrong-sense Wikipedia matches. Sense ambiguity is diagnosed qualitatively, not isolated by an experiment holding retrieval quality fixed.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- For response-level hallucination detection in retrieval-augmented settings, having the model first convert the reference documents into a relational SQL database and then check the response against SELECT query results raises precision and macro F1 over zero-shot direct prediction, over atomic-claim decomposition, and over flat JSON key-value extraction, without any task-specific fine-tuning.Stating false facts confidently · reviewed by AI
- For agentic retrieval-augmented multi-hop QA with 7B-class policy models, scoring each intermediate step for both logical validity and grounding in the retrieved passages — rather than scoring steps by the final answer — yields higher EM/F1 than outcome-reward and final-answer-based process-reward training, because it removes preference pairs where flawed retrieval happened to reach the right answer.Carrying a chain of inference through to its conclusion · reviewed by AI
- In ESG-report RAG question answering with retrieval held fixed at a single top-1 chunk, open-weight 2B-30B generators vary widely in how well they stay anchored to the retrieved passage (faithfulness 0.607-0.822) yet all land in a narrow low band of factual correctness (0.387-0.449), and the most faithful model is the least factually correct, so groundedness to context and correctness of extracted ESG facts do not track each other.Competence in a specialist domain · reviewed by AI
- In retrieval-augmented answers, splitting an answer into claims, checking each against the retrieved source, and then acting on the flagged claims reduces the share of answers judged to contain unsupported content — but the strategies trade grounding against preservation: deleting unsupported claims reduces unsupported content most while retaining the least original text, and rewriting retains the most while reducing least.Stating false facts confidently · reviewed by AI
- On long-context tasks over 8K-128K-token documents in the six official UN languages, models answer word-level span questions well (about 73% average accuracy across 12 models) but degrade sharply as the required unit coarsens to sentence selection (about 21%), paragraph filling and document-level summarization and translation, so long-context competence measured by needle-style word retrieval overstates comprehension of larger discourse units.Losing information in long inputs · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: medium. Falsifier as drafted: Same pipeline with the same surface-form retrieval showing claim-level factuality (FENICE/SummaC) at or above the no-retrieval baseline, or showing that the factuality drop persists when the retrieved definitions are all sense-correct. Drafted stance toward retrieval-moves-the-trust-boundary-rather-than-removing-it: supports -- Retrieved definitions were treated as reliable context and injected unsupported content, so grounding shifted the failure into the retrieval channel rather than removing it.