Hallucination Augmented Recitations for Language Models

TL;DR

This paper introduces Hallucination Augmented Recitations (HAR), a novel method for creating counterfactual datasets using hallucination in LLMs to enhance attribution and factual grounding in open book QA tasks.

Contribution

The paper proposes HAR, a new approach to generate counterfactual datasets with hallucination, leading to improved attribution and QA performance over factual datasets.

Findings

Up to 8.0% increase in F1 score on open book QA.

Counterfactual datasets outperform factual datasets even with smaller size.

Improvements are consistent across various datasets and model sizes.

Abstract

Attribution is a key concept in large language models (LLMs) as it enables control over information sources and enhances the factuality of LLMs. While existing approaches utilize open book question answering to improve attribution, factual datasets may reward language models to recall facts that they already know from their pretraining data, not attribution. In contrast, counterfactual open book QA datasets would further improve attribution because the answer could only be grounded in the given text. We propose Hallucination Augmented Recitations (HAR) for creating counterfactual datasets by utilizing hallucination in LLMs to improve attribution. For open book QA as a case study, we demonstrate that models finetuned with our counterfactual datasets improve text grounding, leading to better open book QA performance, with up to an 8.0% increase in F1 score. Our counterfactual dataset leads to significantly better performance than using humanannotated factual datasets, even with 4x smaller datasets and 4x smaller models. We observe that improvements are consistent across various model sizes and datasets, including multi-hop, biomedical, and adversarial QA datasets.