{"paper":{"title":"Distributional Open-Ended Evaluation of LLM Cultural Value Alignment Based on Value Codebook","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"DOVE evaluates LLM cultural alignment by mapping texts to a learned value codebook and comparing distributions with unbalanced optimal transport.","cross_cats":["cs.AI","cs.CY","cs.LG"],"primary_cat":"cs.CL","authors_text":"Hyunjin Hwang, Jaehyeok Lee, Jing Yao, JinYeong Bak, Roy Ka-Wei Lee, Xiaoyuan Yi, Xing Xie","submitted_at":"2026-03-16T08:33:10Z","abstract_excerpt":"As LLMs are globally deployed, aligning their cultural value orientations is critical for safety and user engagement. However, existing benchmarks face the Construct-Composition-Context ($C^3$) challenge: relying on discriminative, multiple-choice formats that probe value knowledge rather than true orientations, overlook subcultural heterogeneity, and mismatch with real-world open-ended generation. We introduce DOVE, a distributional evaluation framework that directly compares human-written text distributions with LLM-generated outputs. DOVE utilizes a rate-distortion variational optimization "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"DOVE achieves superior predictive validity, attaining a 31.56% correlation with downstream tasks, while maintaining high reliability with as few as 500 samples per culture.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The value codebook constructed from 10K documents via rate-distortion optimization accurately captures true cultural value orientations rather than surface-level semantic patterns, and that distributional differences measured by unbalanced optimal transport reflect genuine alignment rather than generation style artifacts.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"DOVE constructs a value codebook via rate-distortion variational optimization from 10K documents and measures LLM-human cultural alignment through unbalanced optimal transport, showing 31.56% correlation with downstream tasks and reliability at 500 samples per culture.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"DOVE evaluates LLM cultural alignment by mapping texts to a learned value codebook and comparing distributions with unbalanced optimal transport.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"2de6e300927b5723887d6b61144477ccf8d76556ac2fb571bddcdc163204fc81"},"source":{"id":"2604.06210","kind":"arxiv","version":3},"verdict":{"id":"20fd7bed-436b-4f9a-91da-08b117a11d14","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T10:49:22.828737Z","strongest_claim":"DOVE achieves superior predictive validity, attaining a 31.56% correlation with downstream tasks, while maintaining high reliability with as few as 500 samples per culture.","one_line_summary":"DOVE constructs a value codebook via rate-distortion variational optimization from 10K documents and measures LLM-human cultural alignment through unbalanced optimal transport, showing 31.56% correlation with downstream tasks and reliability at 500 samples per culture.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The value codebook constructed from 10K documents via rate-distortion optimization accurately captures true cultural value orientations rather than surface-level semantic patterns, and that distributional differences measured by unbalanced optimal transport reflect genuine alignment rather than generation style artifacts.","pith_extraction_headline":"DOVE evaluates LLM cultural alignment by mapping texts to a learned value codebook and comparing distributions with unbalanced optimal transport."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.06210/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":2,"snapshot_sha256":"d8e5314fcefacb6f44eab036776b7d75e14a5826aaa64222b2df1f59bcb849d2"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}