ANAH: Analytical Annotation of Hallucinations in Large Language Models

TL;DR

ANAH is a bilingual dataset with detailed annotations of hallucinations in LLM-generated answers, enabling better measurement, training, and evaluation of hallucination detection and correction methods.

Contribution

The paper introduces ANAH, a comprehensive dataset with fine-grained hallucination annotations for LLM answers, and demonstrates its effectiveness in training and evaluating hallucination annotators.

Findings

Generative annotators trained on ANAH outperform open-source LLMs.

ANAH enables training models that approach GPT-4's performance.

Fine-grained annotations help understand hallucination accumulation in LLMs.

Abstract

Reducing the `$\textit{hallucination}$' problem of Large Language Models (LLMs) is crucial for their wide applications. A comprehensive and fine-grained measurement of the hallucination is the first key step for the governance of this issue but is under-explored in the community. Thus, we present $\textbf{ANAH}$, a bilingual dataset that offers $\textbf{AN}$alytical $\textbf{A}$nnotation of $\textbf{H}$allucinations in LLMs within Generative Question Answering. Each answer sentence in our dataset undergoes rigorous annotation, involving the retrieval of a reference fragment, the judgment of the hallucination type, and the correction of hallucinated content. ANAH consists of ~12k sentence-level annotations for ~4.3k LLM responses covering over 700 topics, constructed by a human-in-the-loop pipeline. Thanks to the fine granularity of the hallucination annotations, we can quantitatively confirm that the hallucinations of LLMs progressively accumulate in the answer and use ANAH to train and evaluate hallucination annotators. We conduct extensive experiments on studying generative and discriminative annotators and show that, although current open-source LLMs have difficulties in fine-grained hallucination annotation, the generative annotator trained with ANAH can surpass all open-source LLMs and GPT-3.5, obtain performance competitive with GPT-4, and exhibits better generalization ability on unseen questions.