Research map / Large language models / Language, safety and interpretability
Automatic Construction of Evaluation Suites for Natural Language Generation Datasets
Simon Mille, Kaustubh Dhole, Saad Mahamood, Laura Perez-Beltrachini, Varun Gangal, Mihir Kale, Emiel van Miltenburg, Sebastian Gehrmann
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Evaluate on COVID-19-related news instead of the standard test distribution | mT5 generation models · MLSum and XSum COVID-19 test sets | BERTScore: drops of almost 40% | Hurt |
| Test on inputs with properties unseen during training | mT5 WebNLG models · WebNLG English and Russian | performance | Hurt |
| Apply systematic input transformations, including back-translation, typos, punctuation removal, or component reordering | mT5 generation models · English text-to-text, dialog, and data-to-text challenge sets | performance scores | Hurt |
| Increase the number of input properties | mT5 Base · WebNLG English | BLEURT: 0.20 -> 0.08 -> 0.03 -> -0.04 -> -0.10 -> -0.10 -> -0.14 | Hurt |
| Increase input size | generation models · Czech Restaurants and ToTTo | performance | No clear effect |
| Scramble input component order | mT5 generation models · data-to-text and Schema-guided Dialog | local recall | Hurt |
| Evaluate on inputs about female people rather than male people | generation models · ToTTo people subpopulations | BLEURT | Hurt |
| Evaluate African American examples rather than the all-USA subset | generation models · ToTTo people subpopulations | BLEURT | Hurt |
Closest papers by meaning
- Generative Judge for Evaluating Alignment (ICLR 2024 · similarity 0.87)
- Neither Valid nor Reliable? Investigating the Use of LLMs as Judges (NeurIPS 2025 · similarity 0.86)
- Reliable and Efficient Amortized Model-based Evaluation (ICML 2025 · similarity 0.86)
- Benchmarking LLMs' Judgments with No Gold Standard (ICLR 2025 · similarity 0.85)
- Automating Dataset Updates Towards Reliable and Timely Evaluation of Large Language Models (NeurIPS 2024 · similarity 0.85)
- SimpleStrat: Diversifying Language Model Generation with Stratification (NeurIPS 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
