The Pitfalls of Defining Hallucination
This addresses conceptual clarity issues for researchers in NLG evaluation, but is incremental as it synthesizes rather than introduces new methods.
The paper examines the unclear definitions of hallucination and omission in data-to-text natural language generation, proposing a logic-based synthesis of existing classifications and highlighting remaining limitations for large language models.
Despite impressive advances in Natural Language Generation (NLG) and Large Language Models (LLMs), researchers are still unclear about important aspects of NLG evaluation. To substantiate this claim, I examine current classifications of hallucination and omission in Data-text NLG, and I propose a logic-based synthesis of these classfications. I conclude by highlighting some remaining limitations of all current thinking about hallucination and by discussing implications for LLMs.