{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:EH5NFG6BJ7MYYKZUDVIX2M5C2H","short_pith_number":"pith:EH5NFG6B","schema_version":"1.0","canonical_sha256":"21fad29bc14fd98c2b341d517d33a2d1f0261b7f7b0c0382df5375719280c544","source":{"kind":"arxiv","id":"2309.10966","version":6},"attestation_state":"computed","paper":{"title":"MBR and QE Finetuning: Training-time Distillation of the Best and Most Expensive Decoding Methods","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Apurva Shah, Mara Finkelstein, Markus Freitag, Mehdi Mirzazadeh, Subhajit Naskar","submitted_at":"2023-09-19T23:39:07Z","abstract_excerpt":"Recent research in decoding methods for Natural Language Generation (NLG) tasks has shown that MAP decoding is not optimal, because model probabilities do not always align with human preferences. Stronger decoding methods, including Quality Estimation (QE) reranking and Minimum Bayes' Risk (MBR) decoding, have since been proposed to mitigate the model-perplexity-vs-quality mismatch. While these decoding methods achieve state-of-the-art performance, they are prohibitively expensive to compute. In this work, we propose MBR finetuning and QE finetuning which distill the quality gains from these d"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2309.10966","kind":"arxiv","version":6},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2023-09-19T23:39:07Z","cross_cats_sorted":[],"title_canon_sha256":"33da07b63b7a8dbc1427499d3253dcd44e97b8e7e630aa79f64ffb81c15d68a3","abstract_canon_sha256":"f7057ac6691bd08e9ead742c8b98924fd402495ca63ea568d57bf2ea5acc2194"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:00:36.026912Z","signature_b64":"R/f7Cqrh1h2NGHipMJ37uHyLjYucSXzxmSdagOgi7InyYSB96YfbWngZMsH3oKbcSK5hiO1h6f4wIqjAf6PjDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"21fad29bc14fd98c2b341d517d33a2d1f0261b7f7b0c0382df5375719280c544","last_reissued_at":"2026-07-05T08:00:36.026442Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:00:36.026442Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MBR and QE Finetuning: Training-time Distillation of the Best and Most Expensive Decoding Methods","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Apurva Shah, Mara Finkelstein, Markus Freitag, Mehdi Mirzazadeh, Subhajit Naskar","submitted_at":"2023-09-19T23:39:07Z","abstract_excerpt":"Recent research in decoding methods for Natural Language Generation (NLG) tasks has shown that MAP decoding is not optimal, because model probabilities do not always align with human preferences. Stronger decoding methods, including Quality Estimation (QE) reranking and Minimum Bayes' Risk (MBR) decoding, have since been proposed to mitigate the model-perplexity-vs-quality mismatch. While these decoding methods achieve state-of-the-art performance, they are prohibitively expensive to compute. In this work, we propose MBR finetuning and QE finetuning which distill the quality gains from these d"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2309.10966","kind":"arxiv","version":6},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2309.10966/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":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2309.10966","created_at":"2026-07-05T08:00:36.026500+00:00"},{"alias_kind":"arxiv_version","alias_value":"2309.10966v6","created_at":"2026-07-05T08:00:36.026500+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2309.10966","created_at":"2026-07-05T08:00:36.026500+00:00"},{"alias_kind":"pith_short_12","alias_value":"EH5NFG6BJ7MY","created_at":"2026-07-05T08:00:36.026500+00:00"},{"alias_kind":"pith_short_16","alias_value":"EH5NFG6BJ7MYYKZU","created_at":"2026-07-05T08:00:36.026500+00:00"},{"alias_kind":"pith_short_8","alias_value":"EH5NFG6B","created_at":"2026-07-05T08:00:36.026500+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.23306","citing_title":"The Anatomy of the CTC Oracle Gap: Acoustic Exhaustion and Linguistic Recovery","ref_index":1,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EH5NFG6BJ7MYYKZUDVIX2M5C2H","json":"https://pith.science/pith/EH5NFG6BJ7MYYKZUDVIX2M5C2H.json","graph_json":"https://pith.science/api/pith-number/EH5NFG6BJ7MYYKZUDVIX2M5C2H/graph.json","events_json":"https://pith.science/api/pith-number/EH5NFG6BJ7MYYKZUDVIX2M5C2H/events.json","paper":"https://pith.science/paper/EH5NFG6B"},"agent_actions":{"view_html":"https://pith.science/pith/EH5NFG6BJ7MYYKZUDVIX2M5C2H","download_json":"https://pith.science/pith/EH5NFG6BJ7MYYKZUDVIX2M5C2H.json","view_paper":"https://pith.science/paper/EH5NFG6B","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2309.10966&json=true","fetch_graph":"https://pith.science/api/pith-number/EH5NFG6BJ7MYYKZUDVIX2M5C2H/graph.json","fetch_events":"https://pith.science/api/pith-number/EH5NFG6BJ7MYYKZUDVIX2M5C2H/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EH5NFG6BJ7MYYKZUDVIX2M5C2H/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EH5NFG6BJ7MYYKZUDVIX2M5C2H/action/storage_attestation","attest_author":"https://pith.science/pith/EH5NFG6BJ7MYYKZUDVIX2M5C2H/action/author_attestation","sign_citation":"https://pith.science/pith/EH5NFG6BJ7MYYKZUDVIX2M5C2H/action/citation_signature","submit_replication":"https://pith.science/pith/EH5NFG6BJ7MYYKZUDVIX2M5C2H/action/replication_record"}},"created_at":"2026-07-05T08:00:36.026500+00:00","updated_at":"2026-07-05T08:00:36.026500+00:00"}