{"paper":{"title":"Creativity Bias: How Machine Evaluation Struggles with Creativity in Literary Translations","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"Automatic evaluation metrics and LLM judges correlate poorly with professional translators on creativity in literary texts and bias toward machine outputs.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ana Guerberof Arenas, Kyo Gerrits, Rik van Noord","submitted_at":"2026-05-13T14:30:41Z","abstract_excerpt":"This article investigates the performance of automatic evaluation metrics (AEMs) and LLM-as-a-judge evaluation on literary translation across multiple languages, genres, and translation modalities. The aim is to assess how well these tools align with professionals when evaluating translation, creativity (creative shifts & errors), and see if they can substitute laborious manual annotations. A dataset of literary translations across three modalities (human translation, machine translation, and post-editing), three genres and three language pairs was created and annotated in detail for creativit"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"both AEMs and LLM-as-a-judge evaluations correlate poorly with professional evaluations on creativity, with LLM-as-a-judge showing a systematic bias in favour of machine-translated texts and penalising creative and culturally appropriate solutions.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That detailed annotations by experienced professional literary translators constitute an objective and reliable ground truth for measuring creativity and translation quality across genres and modalities.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Automatic evaluation tools for literary translations correlate poorly with expert human judgments on creativity and exhibit bias favoring machine-translated texts.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Automatic evaluation metrics and LLM judges correlate poorly with professional translators on creativity in literary texts and bias toward machine outputs.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"0f55518bf682a68175241d0a9116fc2265d3e705ad605523ef6c62f1f069aa88"},"source":{"id":"2605.13596","kind":"arxiv","version":1},"verdict":{"id":"f076e1eb-f3a1-43df-ab63-f0e1a5031fc1","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-14T20:02:14.689918Z","strongest_claim":"both AEMs and LLM-as-a-judge evaluations correlate poorly with professional evaluations on creativity, with LLM-as-a-judge showing a systematic bias in favour of machine-translated texts and penalising creative and culturally appropriate solutions.","one_line_summary":"Automatic evaluation tools for literary translations correlate poorly with expert human judgments on creativity and exhibit bias favoring machine-translated texts.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That detailed annotations by experienced professional literary translators constitute an objective and reliable ground truth for measuring creativity and translation quality across genres and modalities.","pith_extraction_headline":"Automatic evaluation metrics and LLM judges correlate poorly with professional translators on creativity in literary texts and bias toward machine outputs."},"references":{"count":130,"sample":[{"doi":"","year":1972,"title":"and Ullman, Jeffrey D","work_id":"9f4d095b-5cc6-4ad8-a102-3e3ec582f2e1","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2006,"title":"Interspeech 2006 --- Ninth International Conference on Spoken Language Processing , address=","work_id":"dca1746a-b0f4-4744-96a6-703b6c190994","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1983,"title":"Publications. 1983 , publisher=","work_id":"653ffea5-4aac-4cbc-ac57-775ffd7e5df3","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":1981,"title":"Chandra and Dexter C","work_id":"30e59c52-947d-4f04-ac44-aa8b8453b2b3","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2008,"title":"Coling 2008, 22nd International Conference on Computational Linguistics , address=","work_id":"5e2dd450-37f6-41fa-a3fc-30ae4540c645","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":130,"snapshot_sha256":"df677d842fccc16f555bd598783cf37afe9ed33b9044b9285f069c74e23edfcb","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}