Back to AI News
arxiv

One Framework to Diagnose Them All: A Stratified Benchmark for Bengali LLM Hallucination

Researchers introduce Alikbachan, a Bengali hallucination-detection benchmark stratified along three crossed axes: a human-validated cultural-distance band, six task types, and a five-family, 32-mode hallucination taxonomy. It combines a harvested 2,516-item test set (Phase 1) with a quota-constructed, gold-labeled 5,000-item set (Phase 2). Four zero-shot LLM judges vary by over 30 macro-F1 points across strata, and discourse-level omission defeats judges and most competing systems, while judge verdict probabilities rank items well (ROC-AUC 0.74-0.95) but are overconfident (ECE 0.04-0.29). An open competition drew 222 teams and 3,566 submissions; of 25 teams entering both phases, 24 regressed on Phase 2, and Phase-1 rank barely predicted Phase-2 rank (Spearman rho=0.35).

AI Brief

It gives multilingual evaluation teams a consolidated, stratified hallucination benchmark and concrete evidence that single-phase leaderboards can misrank systems.

arXiv:2611.26336v1 Announce Type: cross Abstract: Hallucination research in low-resource languages fragments easily: each newly named failure mode prompts its own dataset, spending scarce annotators on incomparable resources. We argue for consolidation and test it at community scale. Alikbachan is a Bengali hallucination-detection benchmark stratified along three crossed axes: a human-validated cultural-distance band, six task types, and a five-family, 32-mode hallucination taxonomy. It comprises a harvested 2,516-item test set (Phase 1) and a quota-constructed, gold-labeled 5,000-item set (Phase 2). The stratification is diagnostic. Four zero-shot LLM judges vary by over 30 macro-F1 points across strata, and the item-level predictions of 99 Phase-1 and 32 Phase-2 competition systems reproduce the profile: discourse-level omission defeats judges and most teams alike, whereas only a minority over-flag negative controls. The judges' verdict probabilities rank items well (ROC-AUC 0.74-0.95) but are overconfident (ECE 0.04-0.29). An open competition on Alikbachan drew 222 teams and 3,566 submissions, and a single-prompt 32B judge reaches the 57th percentile. Of the 25 teams in both phases, 24 regress on Phase 2 (mean $-17.6$ points) while every zero-shot judge improves, and Phase-1 rank barely predicts Phase-2 rank (Spearman $\rho=0.35$), so a Phase-1 leaderboard alone would have crowned a different winner.
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!