{
  "abstract": "Across various non-verifiable tasks, human evaluation is reliable but expensive, while automatic metrics are more scalable but often biased. Building on prediction-powered inference (PPI), we propose prediction-powered evaluation, a framework that combines limited human judgments with large-scale automatic scores to obtain data-efficient system comparisons that are provably unbiased. We develop parametric and non-parametric procedures, analyze the efficiency trade-off between paired and unpaired designs, and validate the framework on six WMT datasets. We further introduce the Prediction-Powered Saving Ratio (PPSR), a meta-metric that measures how much human annotation an automatic metric can save when used within prediction-powered evaluation. PPSR directly targets metric utility for prediction-powered evaluation and yields more discriminative and stable metric rankings than existing system-level meta-metrics. Overall, our new paradigm reframes automatic metrics as tools for reducing human annotation cost rather than replacing human judgment, and applies broadly to non-verifiable tasks.",
  "abstract_shown": true,
  "affiliations": [],
  "article_type": null,
  "arxiv": "2608.26638",
  "author_total": 3,
  "authors": [
    {
      "name": "Mingqi Gao"
    },
    {
      "name": "Anthony Sicilia"
    },
    {
      "name": "Weiyan Shi"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/71bk0e6hqa9rn.bib",
  "cited_by": 0,
  "doi": "10.48550/arxiv.2608.26638",
  "fulltext_url": null,
  "id": "71bk0e6hqa9rn",
  "issns": [],
  "landing_url": "https://arxiv.org/abs/2608.26638",
  "language": null,
  "licence": "LicenseRef-arXiv-nonexclusive-1.0",
  "licence_label": "arXiv licence",
  "licence_source": "article",
  "licence_url": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
  "metadata_licence": "CC0-1.0",
  "native_subjects": [
    [
      "arxiv",
      "cs.CL"
    ],
    [
      "arxiv",
      "stat.ML"
    ],
    [
      "openalex",
      "1702|Natural Language Processing Techniques"
    ]
  ],
  "openalex": "W7204555598",
  "pdf_url": "https://arxiv.org/pdf/2608.26638",
  "pmcid": null,
  "pmid": null,
  "published": "2026-08-30",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/71bk0e6hqa9rn.ris",
  "sources": [
    {
      "first_seen": "2026-10-30T12:17:54Z",
      "native_id": "2608.26638",
      "source": "arxiv",
      "url": "https://arxiv.org/abs/2608.26638"
    }
  ],
  "subjects": [
    {
      "field": "mathematics",
      "origin": "arxiv:stat.ML",
      "subfield": "statistics"
    }
  ],
  "title": "Which Metrics Save the Most Human Annotation? Prediction-Powered Evaluation and Meta-Evaluation",
  "title_text": "Which Metrics Save the Most Human Annotation? Prediction-Powered Evaluation and Meta-Evaluation",
  "type": "preprint",
  "updated_at": "2026-10-30T12:17:54Z",
  "url": "https://research.draconitic.com/paper/71bk0e6hqa9rn",
  "venue": "arXiv",
  "versions": []
}