How Reliable Are Automatic Evaluation Methods for Instruction-Tuned LLMs?

Ehsan Doostmohammadi, Oskar Holmström, Marco Kuhlmann


Abstract
Work on instruction-tuned Large Language Models (LLMs) has used automatic methods based on text overlap and LLM judgments as cost-effective alternatives to human evaluation. In this paper, we perform a meta-evaluation of such methods and assess their reliability across a broad range of tasks. In evaluating how well automatic methods align with human evaluations, correlation metrics are the most commonly employed method despite their inherent limitations when dealing with ties and different scales. To address these shortcomings, we use Pairwise Accuracy as an alternative to standard correlation measures. We observe that while automatic evaluation methods can approximate human ratings under specific conditions, their validity is highly context-dependent. Specifically, the simple ROUGE-L metric correlates very well with human ratings for short-answer English tasks but is unreliable in free-form generation tasks and cross-lingual scenarios. The effectiveness of the more advanced method of using GPT-4 as a judge diminishes significantly if reference answers are not included in the prompt, which is the scenario where this method has the potential to provide the most value compared to other metrics. Our findings enhance the understanding of how automatic methods should be applied and interpreted when developing and evaluating instruction-tuned LLMs.
Anthology ID:
2024.findings-emnlp.367
Volume:
Findings of the Association for Computational Linguistics: EMNLP 2024
Month:
November
Year:
2024
Address:
Miami, Florida, USA
Editors:
Yaser Al-Onaizan, Mohit Bansal, Yun-Nung Chen
Venue:
Findings
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
6321–6336
Language:
URL:
https://aclanthology.org/2024.findings-emnlp.367
DOI:
Bibkey:
Cite (ACL):
Ehsan Doostmohammadi, Oskar Holmström, and Marco Kuhlmann. 2024. How Reliable Are Automatic Evaluation Methods for Instruction-Tuned LLMs?. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 6321–6336, Miami, Florida, USA. Association for Computational Linguistics.
Cite (Informal):
How Reliable Are Automatic Evaluation Methods for Instruction-Tuned LLMs? (Doostmohammadi et al., Findings 2024)
Copy Citation:
PDF:
https://aclanthology.org/2024.findings-emnlp.367.pdf