G-Eval: NLG Evaluation using GPT-4 with Better Human Alignment
This addresses the challenge of low correlation with human judgments in NLG evaluation, particularly for tasks requiring creativity and diversity, though it is incremental as it builds on existing LLM-based evaluators.
The authors tackled the problem of automatically evaluating natural language generation (NLG) systems by proposing G-Eval, a framework using GPT-4 with chain-of-thoughts and form-filling, which achieved a Spearman correlation of 0.514 with human judgments on summarization tasks, outperforming previous methods.
The quality of texts generated by natural language generation (NLG) systems is hard to measure automatically. Conventional reference-based metrics, such as BLEU and ROUGE, have been shown to have relatively low correlation with human judgments, especially for tasks that require creativity and diversity. Recent studies suggest using large language models (LLMs) as reference-free metrics for NLG evaluation, which have the benefit of being applicable to new tasks that lack human references. However, these LLM-based evaluators still have lower human correspondence than medium-size neural evaluators. In this work, we present G-Eval, a framework of using large language models with chain-of-thoughts (CoT) and a form-filling paradigm, to assess the quality of NLG outputs. We experiment with two generation tasks, text summarization and dialogue generation. We show that G-Eval with GPT-4 as the backbone model achieves a Spearman correlation of 0.514 with human on summarization task, outperforming all previous methods by a large margin. We also propose preliminary analysis on the behavior of LLM-based evaluators, and highlight the potential issue of LLM-based evaluators having a bias towards the LLM-generated texts. The code is at https://github.com/nlpyang/geval