2000 character limit reached
Compare without Despair: Reliable Preference Evaluation with Generation Separability (2407.01878v3)
Published 2 Jul 2024 in cs.CL
Abstract: Human evaluation of generated language through pairwise preference judgments is pervasive. However, under common scenarios, such as when generations from a model pair are very similar, or when stochastic decoding results in large variations in generations, it results in inconsistent preference ratings. We address these challenges by introducing a meta-evaluation measure, separability, which estimates how suitable a test instance is for pairwise preference evaluation. For a candidate test instance, separability samples multiple generations from a pair of models, and measures how distinguishable the two sets of generations are. Our experiments show that instances with high separability values yield more consistent preference ratings from both human- and auto-raters. Further, the distribution of separability allows insights into which test benchmarks are more valuable for comparing models. Finally, we incorporate separability into ELO ratings, accounting for how suitable each test instance might be for reliably ranking LLMs. Overall, separability has implications for consistent, efficient and robust preference evaluation of LLMs with both human- and auto-raters.
- Agreement is overrated: A plea for correlation to assess human evaluation reliability. In Proceedings of the 12th International Conference on Natural Language Generation, pages 344–354, Tokyo, Japan. Association for Computational Linguistics.
- Label-efficient model selection for text generation. arXiv preprint arXiv:2402.07891.
- Findings of the 2019 conference on machine translation (WMT19). In Proceedings of the Fourth Conference on Machine Translation (Volume 2: Shared Task Papers, Day 1), pages 1–61, Florence, Italy. Association for Computational Linguistics.
- It’s MBR all the way down: Modern generation techniques through the lens of minimum Bayes risk. In Proceedings of the Big Picture Workshop, pages 108–122, Singapore. Association for Computational Linguistics.
- Abductive commonsense reasoning. In International Conference on Learning Representations.
- Which prompts make the difference? data prioritization for efficient human llm evaluation. arXiv preprint arXiv:2310.14424.
- Elo uncovered: Robustness and best practices in language model evaluation. arXiv preprint arXiv:2311.17295.
- Adversarial filters of dataset biases. In Proc. of ICML.
- A survey on evaluation of large language models. ACM Trans. Intell. Syst. Technol., 15(3).
- Chatbot arena: An open platform for evaluating llms by human preference. arXiv preprint arXiv:2403.04132.
- Length-controlled alpacaeval: A simple way to debias automatic evaluators. arXiv preprint arXiv:2404.04475.
- Kawin Ethayarajh and Dan Jurafsky. 2022. The authenticity gap in human evaluation. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 6056–6070, Abu Dhabi, United Arab Emirates. Association for Computational Linguistics.
- Repairing the cracked foundation: A survey of obstacles in evaluation practices for generated text. arXiv preprint arXiv:2202.06935.
- What comes next? evaluating uncertainty in neural text generators against human production variability. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 14349–14371, Singapore. Association for Computational Linguistics.
- SAMSum corpus: A human-annotated dialogue dataset for abstractive summarization. In Proceedings of the 2nd Workshop on New Frontiers in Summarization, pages 70–79, Hong Kong, China. Association for Computational Linguistics.
- News summarization and evaluation in the era of gpt-3. arXiv preprint arXiv:2209.12356.
- Mistral 7b. arXiv preprint arXiv:2310.06825.
- BiSECT: Learning to split and rephrase sentences with bitexts. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 6193–6209, Online and Punta Cana, Dominican Republic. Association for Computational Linguistics.
- Wildbench: Benchmarking llms with challenging tasks from real users in the wild. arXiv preprint arXiv:2406.04770.
- Chin-Yew Lin. 2004. ROUGE: A package for automatic evaluation of summaries. In Text Summarization Branches Out, pages 74–81, Barcelona, Spain. Association for Computational Linguistics.
- G-eval: NLG evaluation using gpt-4 with better human alignment. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 2511–2522, Singapore. Association for Computational Linguistics.
- LMSys. 2023. Vicuna: A cloud-native computing service for machine learning workflows.
- The flan collection: Designing data and methods for effective instruction tuning. In International Conference on Machine Learning, pages 22631–22648. PMLR.
- Abstractive text summarization using sequence-to-sequence RNNs and beyond. In Proceedings of the 20th SIGNLL Conference on Computational Natural Language Learning, pages 280–290, Berlin, Germany. Association for Computational Linguistics.
- Llm evaluators recognize and favor their own generations. arXiv preprint arXiv:2404.13076.
- Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pages 311–318, Philadelphia, Pennsylvania, USA. Association for Computational Linguistics.
- On releasing annotator-level labels and information in datasets. In Proceedings of the Joint 15th Linguistic Annotation Workshop (LAW) and 3rd Designing Meaning Representations (DMR) Workshop, pages 133–138, Punta Cana, Dominican Republic. Association for Computational Linguistics.
- Evaluation examples are not equally informative: How should that change NLP leaderboards? In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 4486–4503, Online. Association for Computational Linguistics.
- Roy Schwartz and Gabriel Stanovsky. 2022. On the limitations of dataset balancing: The lost battle against spurious correlations. In Findings of the Association for Computational Linguistics: NAACL 2022, pages 2182–2194, Seattle, United States. Association for Computational Linguistics.
- How to compare summarizers without target length? pitfalls, solutions and re-examination of the neural summarization literature. In Proceedings of the Workshop on Methods for Optimizing and Evaluating Neural Language Generation, pages 21–29, Minneapolis, Minnesota. Association for Computational Linguistics.
- Follow the wisdom of the crowd: Effective text generation via minimum bayes risk decoding. arXiv preprint arXiv:2211.07634.
- Predictions from language models for multiple-choice tasks are not robust under variation of scoring methods. arXiv preprint arXiv:2403.00998.
- Comparing test sets with item response theory. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers), pages 1141–1158, Online. Association for Computational Linguistics.
- Large language models are not fair evaluators. arXiv preprint arXiv:2305.17926.
- Minghao Wu and Alham Fikri Aji. 2023. Style over substance: Evaluation biases for large language models. arXiv preprint arXiv:2307.03025.
- Evaluating large language models at evaluating instruction following. In The Twelfth International Conference on Learning Representations.
- Bertscore: Evaluating text generation with bert. In International Conference on Learning Representations.
- Benchmarking large language models for news summarization. arXiv preprint arXiv:2301.13848.
- Judging llm-as-a-judge with mt-bench and chatbot arena. Advances in Neural Information Processing Systems, 36.
Paper Prompts
Sign up for free to create and run prompts on this paper using GPT-5.
Top Community Prompts
Collections
Sign up for free to add this paper to one or more collections.