Groundedness in Retrieval-augmented Long-form Generation: An Empirical Study
This addresses the issue of hallucinations in long-form generation for users relying on LLMs for accurate information, but it is incremental as it provides empirical insights rather than a new solution.
The study tackled the problem of groundedness in retrieval-augmented long-form question answering by evaluating whether generated sentences are based on retrieved documents or pre-training data, finding that a significant fraction of sentences are ungrounded even when they contain correct answers.
We present an empirical study of groundedness in long-form question answering (LFQA) by retrieval-augmented large language models (LLMs). In particular, we evaluate whether every generated sentence is grounded in the retrieved documents or the model's pre-training data. Across 3 datasets and 4 model families, our findings reveal that a significant fraction of generated sentences are consistently ungrounded, even when those sentences contain correct ground-truth answers. Additionally, we examine the impacts of factors such as model size, decoding strategy, and instruction tuning on groundedness. Our results show that while larger models tend to ground their outputs more effectively, a significant portion of correct answers remains compromised by hallucinations. This study provides novel insights into the groundedness challenges in LFQA and underscores the necessity for more robust mechanisms in LLMs to mitigate the generation of ungrounded content.