{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2019:MMMH3ZIPPYIA4PRKN3P3QYSQGI","short_pith_number":"pith:MMMH3ZIP","canonical_record":{"source":{"id":"1904.09751","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2019-04-22T07:17:18Z","cross_cats_sorted":[],"title_canon_sha256":"129d205975a35cd33688c24c44d313f3ae68ca6cbc7d1714e8be66e73effee3e","abstract_canon_sha256":"0865715a7c2b1b50ff1338569cfaabd8b7b13c833b5130585c2667f391fad78f"},"schema_version":"1.0"},"canonical_sha256":"63187de50f7e100e3e2a6edfb862503203212d19850bc74f103d3ea21a3fa124","source":{"kind":"arxiv","id":"1904.09751","version":2},"source_aliases":[{"alias_kind":"arxiv","alias_value":"1904.09751","created_at":"2026-07-05T00:40:54Z"},{"alias_kind":"arxiv_version","alias_value":"1904.09751v2","created_at":"2026-07-05T00:40:54Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1904.09751","created_at":"2026-07-05T00:40:54Z"},{"alias_kind":"pith_short_12","alias_value":"MMMH3ZIPPYIA","created_at":"2026-07-05T00:40:54Z"},{"alias_kind":"pith_short_16","alias_value":"MMMH3ZIPPYIA4PRK","created_at":"2026-07-05T00:40:54Z"},{"alias_kind":"pith_short_8","alias_value":"MMMH3ZIP","created_at":"2026-07-05T00:40:54Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2019:MMMH3ZIPPYIA4PRKN3P3QYSQGI","target":"record","payload":{"canonical_record":{"source":{"id":"1904.09751","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2019-04-22T07:17:18Z","cross_cats_sorted":[],"title_canon_sha256":"129d205975a35cd33688c24c44d313f3ae68ca6cbc7d1714e8be66e73effee3e","abstract_canon_sha256":"0865715a7c2b1b50ff1338569cfaabd8b7b13c833b5130585c2667f391fad78f"},"schema_version":"1.0"},"canonical_sha256":"63187de50f7e100e3e2a6edfb862503203212d19850bc74f103d3ea21a3fa124","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T00:40:54.934801Z","signature_b64":"plEODnP7zyxBm2K0eyJ4BGI6IcT9Xnvw+1rDMSpHtQeu7h2Gxsjj+VrKxoC+21pJLc/oVoclNoP9S2bHkLxECw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"63187de50f7e100e3e2a6edfb862503203212d19850bc74f103d3ea21a3fa124","last_reissued_at":"2026-07-05T00:40:54.934288Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T00:40:54.934288Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"1904.09751","source_version":2,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T00:40:54Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"gOdh60kI2voZG3lWL/3QMiuOhzS6tb86bjX2ilpBgVt5Uo3QUAwqqN2zmL7Y2dlSFsc13UrWBF6QIhU+mSSZAA==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-09T05:43:13.455421Z"},"content_sha256":"1978ecb48d2e38fee215783ee177979e999bd1d72ac197155b826501727812c6","schema_version":"1.0","event_id":"sha256:1978ecb48d2e38fee215783ee177979e999bd1d72ac197155b826501727812c6"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2019:MMMH3ZIPPYIA4PRKN3P3QYSQGI","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"The Curious Case of Neural Text Degeneration","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Nucleus sampling draws from the dynamic high-probability set to generate more diverse and coherent text than beam search or top-k methods.","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ari Holtzman, Jan Buys, Li Du, Maxwell Forbes, Yejin Choi","submitted_at":"2019-04-22T07:17:18Z","abstract_excerpt":"Despite considerable advancements with deep neural language models, the enigma of neural text degeneration persists when these models are tested as text generators. The counter-intuitive empirical observation is that even though the use of likelihood as training objective leads to high quality models for a broad range of language understanding tasks, using likelihood as a decoding objective leads to text that is bland and strangely repetitive.\n  In this paper, we reveal surprising distributional differences between human text and machine text. In addition, we find that decoding strategies alon"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"By sampling text from the dynamic nucleus of the probability distribution, which allows for diversity while effectively truncating the less reliable tail of the distribution, the resulting text better demonstrates the quality of human text, yielding enhanced diversity without sacrificing fluency and coherence.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the model's learned probability distribution is sufficiently well-calibrated that low-probability tokens in the tail are reliably lower quality, so truncating them improves rather than harms the output.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Nucleus sampling draws from the smallest set of tokens whose cumulative probability exceeds threshold p, yielding more human-like diversity and coherence than beam search or full-distribution sampling.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Nucleus sampling draws from the dynamic high-probability set to generate more diverse and coherent text than beam search or top-k methods.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"3b8532c10b46334fa9e442cbb9917dfb0faa1cccd3bb127d09f6208d16be5d6b"},"source":{"id":"1904.09751","kind":"arxiv","version":2},"verdict":{"id":"7fd82606-aaec-4bc9-ba7e-d74e59f4114a","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-12T06:13:35.190353Z","strongest_claim":"By sampling text from the dynamic nucleus of the probability distribution, which allows for diversity while effectively truncating the less reliable tail of the distribution, the resulting text better demonstrates the quality of human text, yielding enhanced diversity without sacrificing fluency and coherence.","one_line_summary":"Nucleus sampling draws from the smallest set of tokens whose cumulative probability exceeds threshold p, yielding more human-like diversity and coherence than beam search or full-distribution sampling.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the model's learned probability distribution is sufficiently well-calibrated that low-probability tokens in the tail are reliably lower quality, so truncating them improves rather than harms the output.","pith_extraction_headline":"Nucleus sampling draws from the dynamic high-probability set to generate more diverse and coherent text than beam search or top-k methods."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/1904.09751/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":16,"sample":[{"doi":"","year":2015,"title":"Neural machine translation by jointly learning to align and translate","work_id":"68cc9ea9-5ed8-4ad2-9cd8-3ecaba36c66d","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Language gans falling short","work_id":"7f7fd007-0f9d-4b67-8792-b16d15391b72","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Language GANs falling short","work_id":"851f20c8-1e5a-4d79-8bac-46b38d0a46aa","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2018,"title":"Elizabeth Clark, Yangfeng Ji, and Noah A. Smith. Neural text generation in stories using entity rep- resentations as context. In Proceedings of the 2018 Conference of the North American Chapter of the","work_id":"eadf665e-7076-4bb7-bd5f-17e33ca5ea81","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2020,"title":"Hierarchical neural story generation","work_id":"493a14d4-4c20-40e4-90a0-6be4b605230e","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":16,"snapshot_sha256":"80b9779dd1ee31b6c918039d8cec74b03e32a904c6d45633339b09e7518b698d","internal_anchors":2},"formal_canon":{"evidence_count":2,"snapshot_sha256":"000a34424d54f510ae578f1e7c406f74e33ea795dfc02fc73bc21b058b807fa1"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"verdict_id":"7fd82606-aaec-4bc9-ba7e-d74e59f4114a"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-07-05T00:40:54Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"t1H+m/QJvq0rVqqhzZaVAQhiPYc2WSlWpdjEplLnCSwkPEQ1Sx4EujSkyILyPv1/ZiNbyxNqNEW6pAUkfL8rDw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-07-09T05:43:13.455998Z"},"content_sha256":"15b8fa65dd0d14947fcb6b5bce32dd7253d2875e8eef5545775c962a3e54994a","schema_version":"1.0","event_id":"sha256:15b8fa65dd0d14947fcb6b5bce32dd7253d2875e8eef5545775c962a3e54994a"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/MMMH3ZIPPYIA4PRKN3P3QYSQGI/bundle.json","state_url":"https://pith.science/pith/MMMH3ZIPPYIA4PRKN3P3QYSQGI/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/MMMH3ZIPPYIA4PRKN3P3QYSQGI/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-07-09T05:43:13Z","links":{"resolver":"https://pith.science/pith/MMMH3ZIPPYIA4PRKN3P3QYSQGI","bundle":"https://pith.science/pith/MMMH3ZIPPYIA4PRKN3P3QYSQGI/bundle.json","state":"https://pith.science/pith/MMMH3ZIPPYIA4PRKN3P3QYSQGI/state.json","well_known_bundle":"https://pith.science/.well-known/pith/MMMH3ZIPPYIA4PRKN3P3QYSQGI/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2019:MMMH3ZIPPYIA4PRKN3P3QYSQGI","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"0865715a7c2b1b50ff1338569cfaabd8b7b13c833b5130585c2667f391fad78f","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2019-04-22T07:17:18Z","title_canon_sha256":"129d205975a35cd33688c24c44d313f3ae68ca6cbc7d1714e8be66e73effee3e"},"schema_version":"1.0","source":{"id":"1904.09751","kind":"arxiv","version":2}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"1904.09751","created_at":"2026-07-05T00:40:54Z"},{"alias_kind":"arxiv_version","alias_value":"1904.09751v2","created_at":"2026-07-05T00:40:54Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1904.09751","created_at":"2026-07-05T00:40:54Z"},{"alias_kind":"pith_short_12","alias_value":"MMMH3ZIPPYIA","created_at":"2026-07-05T00:40:54Z"},{"alias_kind":"pith_short_16","alias_value":"MMMH3ZIPPYIA4PRK","created_at":"2026-07-05T00:40:54Z"},{"alias_kind":"pith_short_8","alias_value":"MMMH3ZIP","created_at":"2026-07-05T00:40:54Z"}],"graph_snapshots":[{"event_id":"sha256:15b8fa65dd0d14947fcb6b5bce32dd7253d2875e8eef5545775c962a3e54994a","target":"graph","created_at":"2026-07-05T00:40:54Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"By sampling text from the dynamic nucleus of the probability distribution, which allows for diversity while effectively truncating the less reliable tail of the distribution, the resulting text better demonstrates the quality of human text, yielding enhanced diversity without sacrificing fluency and coherence."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"That the model's learned probability distribution is sufficiently well-calibrated that low-probability tokens in the tail are reliably lower quality, so truncating them improves rather than harms the output."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Nucleus sampling draws from the smallest set of tokens whose cumulative probability exceeds threshold p, yielding more human-like diversity and coherence than beam search or full-distribution sampling."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Nucleus sampling draws from the dynamic high-probability set to generate more diverse and coherent text than beam search or top-k methods."}],"snapshot_sha256":"3b8532c10b46334fa9e442cbb9917dfb0faa1cccd3bb127d09f6208d16be5d6b"},"formal_canon":{"evidence_count":2,"snapshot_sha256":"000a34424d54f510ae578f1e7c406f74e33ea795dfc02fc73bc21b058b807fa1"},"integrity":{"available":true,"clean":true,"detectors_run":[],"endpoint":"/pith/1904.09751/integrity.json","findings":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938","summary":{"advisory":0,"by_detector":{},"critical":0,"informational":0}},"paper":{"abstract_excerpt":"Despite considerable advancements with deep neural language models, the enigma of neural text degeneration persists when these models are tested as text generators. The counter-intuitive empirical observation is that even though the use of likelihood as training objective leads to high quality models for a broad range of language understanding tasks, using likelihood as a decoding objective leads to text that is bland and strangely repetitive.\n  In this paper, we reveal surprising distributional differences between human text and machine text. In addition, we find that decoding strategies alon","authors_text":"Ari Holtzman, Jan Buys, Li Du, Maxwell Forbes, Yejin Choi","cross_cats":[],"headline":"Nucleus sampling draws from the dynamic high-probability set to generate more diverse and coherent text than beam search or top-k methods.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2019-04-22T07:17:18Z","title":"The Curious Case of Neural Text Degeneration"},"references":{"count":16,"internal_anchors":2,"resolved_work":16,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"Neural machine translation by jointly learning to align and translate","work_id":"68cc9ea9-5ed8-4ad2-9cd8-3ecaba36c66d","year":2015},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"Language gans falling short","work_id":"7f7fd007-0f9d-4b67-8792-b16d15391b72","year":2018},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Language GANs falling short","work_id":"851f20c8-1e5a-4d79-8bac-46b38d0a46aa","year":2018},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Elizabeth Clark, Yangfeng Ji, and Noah A. Smith. Neural text generation in stories using entity rep- resentations as context. In Proceedings of the 2018 Conference of the North American Chapter of the","work_id":"eadf665e-7076-4bb7-bd5f-17e33ca5ea81","year":2018},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Hierarchical neural story generation","work_id":"493a14d4-4c20-40e4-90a0-6be4b605230e","year":2020}],"snapshot_sha256":"80b9779dd1ee31b6c918039d8cec74b03e32a904c6d45633339b09e7518b698d"},"source":{"id":"1904.09751","kind":"arxiv","version":2},"verdict":{"created_at":"2026-05-12T06:13:35.190353Z","id":"7fd82606-aaec-4bc9-ba7e-d74e59f4114a","model_set":{"reader":"grok-4.3"},"one_line_summary":"Nucleus sampling draws from the smallest set of tokens whose cumulative probability exceeds threshold p, yielding more human-like diversity and coherence than beam search or full-distribution sampling.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Nucleus sampling draws from the dynamic high-probability set to generate more diverse and coherent text than beam search or top-k methods.","strongest_claim":"By sampling text from the dynamic nucleus of the probability distribution, which allows for diversity while effectively truncating the less reliable tail of the distribution, the resulting text better demonstrates the quality of human text, yielding enhanced diversity without sacrificing fluency and coherence.","weakest_assumption":"That the model's learned probability distribution is sufficiently well-calibrated that low-probability tokens in the tail are reliably lower quality, so truncating them improves rather than harms the output."}},"verdict_id":"7fd82606-aaec-4bc9-ba7e-d74e59f4114a"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:1978ecb48d2e38fee215783ee177979e999bd1d72ac197155b826501727812c6","target":"record","created_at":"2026-07-05T00:40:54Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"0865715a7c2b1b50ff1338569cfaabd8b7b13c833b5130585c2667f391fad78f","cross_cats_sorted":[],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2019-04-22T07:17:18Z","title_canon_sha256":"129d205975a35cd33688c24c44d313f3ae68ca6cbc7d1714e8be66e73effee3e"},"schema_version":"1.0","source":{"id":"1904.09751","kind":"arxiv","version":2}},"canonical_sha256":"63187de50f7e100e3e2a6edfb862503203212d19850bc74f103d3ea21a3fa124","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"63187de50f7e100e3e2a6edfb862503203212d19850bc74f103d3ea21a3fa124","first_computed_at":"2026-07-05T00:40:54.934288Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-07-05T00:40:54.934288Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"plEODnP7zyxBm2K0eyJ4BGI6IcT9Xnvw+1rDMSpHtQeu7h2Gxsjj+VrKxoC+21pJLc/oVoclNoP9S2bHkLxECw==","signature_status":"signed_v1","signed_at":"2026-07-05T00:40:54.934801Z","signed_message":"canonical_sha256_bytes"},"source_id":"1904.09751","source_kind":"arxiv","source_version":2}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:1978ecb48d2e38fee215783ee177979e999bd1d72ac197155b826501727812c6","sha256:15b8fa65dd0d14947fcb6b5bce32dd7253d2875e8eef5545775c962a3e54994a"],"state_sha256":"2198297a3f9d26210311da86948a4d6f99192d327ad26e14e642ada84ccb810b"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"EtbsNsJNJSy9azCP88YBo/kCVcL4hy7W0duH4ceF6BgVFMMPPHJQHI2137YjybW5GIJAfy9PpGEAVoJ1lb6gDQ==","signed_message":"bundle_sha256_bytes","signed_at":"2026-07-09T05:43:13.459227Z","bundle_sha256":"fc13b9b9800e995fbb684905c21ae92cc57b4378bdd7c4f926140a2e666c1bc9"}}