Does the Truthfulness Signal Survive Code-Mixing? Probing Hidden States for Hallucination Detection in Hinglish
It is found that the hallucination signal survives code-mixing well: transfer AUROC ranges from 0.88 to 0.99, with gaps of mostly under 0.05 AUROC relative to in-distribution performance, and that Hindi-trained probes transfer to Hinglish more reliably than English-trained probes.
Tanveer Singh
· 0 citations