AI Science Tracker
EN

When Ground Truth Is Not Ground Truth: Automatic Labeling for Hallucination Detection

Not stated in the abstract reviewed · 3 Nov 2026 · computer science

Discovery

When Ground Truth Is Not Ground Truth: Automatic Labeling for Hallucination Detection

Source check: Source checked. Not independent scientific confirmation.

Finding: Human validation compared automatic supervision labels with human assessments across hallucination-detection methods. Misclassification differed substantially by judge: QGen-9B had the lowest reported rate, while BLEURT-based labels performed worse. This is an evaluation finding, not a universal ranking across tasks.

Technology: Automatic supervision judges including BLEURT and QGen-9B

Publication: Preprint · Test context: retrospective or benchmark. Catalog categories, not an evidence grade.

AI contribution
Automatic supervision judges including BLEURT and QGen-9B
Test and result
Across two QA datasets, two generators and five languages, human validation reported misclassification rates of 12.86% for QGen-9B and about 47% for BLEURT and BLEURT-n.
Limitation / still unknown
The comparison covers the study’s datasets, languages, judges and human-label protocol; rates may not transfer to other settings.
Evidence sources and annotation context

Evidence context (descriptions follow the cited sources; test type is a catalog classification):

Evidence sources: https://arxiv.org/abs/2611.04697

Source checked · editorial details

The source page was checked against this entry. Independent confirmation is not assessed by this label.

Source type: paper. Paper: linked separately.

Review date: 22 Nov 2026. Review scope: Primary result-bearing material for the stated method, result and evaluation, with bibliographic version information..

No follow-up recorded; this is not a claim that no follow-up exists. Follow-up review scope and date are not recorded. No source-supported relationships recorded.

Editorial assessment: Impact 3/5 · Validation 2/5 · Novelty 3/5. Impact and novelty are editorial judgments; validation follows the annotated test type where one exists. Not an evidence grade. Scoring criteria.

Open in the full catalog for filters, charts and related discoveries. Browse computer science discoveries.