Hallucination Span Detection with Input-Side Evidence Alignment

Miyu Yamada, Yuki Arase

Hallucinations remain a major obstacle to the reliable use of large language models (LLMs) in conditional text generation. Existing methods primarily assess the factuality of an en…