On the Benchmarking of LLMs for Open-Domain Dialogue Evaluation
This work identifies a critical gap in evaluation practices for open-domain dialogue systems, which could mislead developers and researchers about model performance.
The paper critiques current benchmarks for evaluating open-domain dialogue systems, arguing they use outdated datasets and metrics that fail to capture modern chatbot capabilities. An annotation experiment on the SODA dataset shows that LLM evaluators like GPT-4 struggle to detect deficiencies in dialogues generated by current LLMs.
Large Language Models (LLMs) have showcased remarkable capabilities in various Natural Language Processing tasks. For automatic open-domain dialogue evaluation in particular, LLMs have been seamlessly integrated into evaluation frameworks, and together with human evaluation, compose the backbone of most evaluations. However, existing evaluation benchmarks often rely on outdated datasets and evaluate aspects like Fluency and Relevance, which fail to adequately capture the capabilities and limitations of state-of-the-art chatbot models. This paper critically examines current evaluation benchmarks, highlighting that the use of older response generators and quality aspects fail to accurately reflect modern chatbot capabilities. A small annotation experiment on a recent LLM-generated dataset (SODA) reveals that LLM evaluators such as GPT-4 struggle to detect actual deficiencies in dialogues generated by current LLM chatbots.