{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:J6TN5AR2K2WGHXA3DMPYV4YWH3","short_pith_number":"pith:J6TN5AR2","schema_version":"1.0","canonical_sha256":"4fa6de823a56ac63dc1b1b1f8af3163ec7b8caccf7e3653f221c6672ac79a502","source":{"kind":"arxiv","id":"2212.01349","version":2},"attestation_state":"computed","paper":{"title":"Nonparametric Masked Language Modeling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Hannaneh Hajishirzi, Luke Zettlemoyer, Mike Lewis, Sewon Min, Weijia Shi, Wen-tau Yih, Xilun Chen","submitted_at":"2022-12-02T18:10:42Z","abstract_excerpt":"Existing language models (LMs) predict tokens with a softmax over a finite vocabulary, which can make it difficult to predict rare tokens or phrases. We introduce NPM, the first nonparametric masked language model that replaces this softmax with a nonparametric distribution over every phrase in a reference corpus. NPM fills in the [MASK] solely from retrieving a token from a text corpus. We show that NPM can be efficiently trained with a contrastive objective and an in-batch approximation to full corpus retrieval. Zero-shot evaluation on 16 tasks including classification, fact probing and ques"},"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":"2212.01349","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2022-12-02T18:10:42Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"56c9942700c807e4c52131c909da299051a850b9a69d8d76f4f95924b31df651","abstract_canon_sha256":"d77594ae513987edde9341c73b5a776fd8287a31e3c66683dede979a99f87d57"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T06:14:07.023350Z","signature_b64":"Uy8e5hd9gn/aqiho4tZHxQgHjqXv2mKbQdLiW1q7+bm/DPKimD/6+hgY21H+EriJALbiDrvn3b0Qkwv42u6KCQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4fa6de823a56ac63dc1b1b1f8af3163ec7b8caccf7e3653f221c6672ac79a502","last_reissued_at":"2026-07-05T06:14:07.022926Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T06:14:07.022926Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Nonparametric Masked Language Modeling","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Hannaneh Hajishirzi, Luke Zettlemoyer, Mike Lewis, Sewon Min, Weijia Shi, Wen-tau Yih, Xilun Chen","submitted_at":"2022-12-02T18:10:42Z","abstract_excerpt":"Existing language models (LMs) predict tokens with a softmax over a finite vocabulary, which can make it difficult to predict rare tokens or phrases. We introduce NPM, the first nonparametric masked language model that replaces this softmax with a nonparametric distribution over every phrase in a reference corpus. NPM fills in the [MASK] solely from retrieving a token from a text corpus. We show that NPM can be efficiently trained with a contrastive objective and an in-batch approximation to full corpus retrieval. Zero-shot evaluation on 16 tasks including classification, fact probing and ques"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2212.01349","kind":"arxiv","version":2},"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/2212.01349/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":"2212.01349","created_at":"2026-07-05T06:14:07.022986+00:00"},{"alias_kind":"arxiv_version","alias_value":"2212.01349v2","created_at":"2026-07-05T06:14:07.022986+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2212.01349","created_at":"2026-07-05T06:14:07.022986+00:00"},{"alias_kind":"pith_short_12","alias_value":"J6TN5AR2K2WG","created_at":"2026-07-05T06:14:07.022986+00:00"},{"alias_kind":"pith_short_16","alias_value":"J6TN5AR2K2WGHXA3","created_at":"2026-07-05T06:14:07.022986+00:00"},{"alias_kind":"pith_short_8","alias_value":"J6TN5AR2","created_at":"2026-07-05T06:14:07.022986+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2301.12652","citing_title":"REPLUG: Retrieval-Augmented Black-Box Language Models","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2310.11511","citing_title":"Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection","ref_index":113,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/J6TN5AR2K2WGHXA3DMPYV4YWH3","json":"https://pith.science/pith/J6TN5AR2K2WGHXA3DMPYV4YWH3.json","graph_json":"https://pith.science/api/pith-number/J6TN5AR2K2WGHXA3DMPYV4YWH3/graph.json","events_json":"https://pith.science/api/pith-number/J6TN5AR2K2WGHXA3DMPYV4YWH3/events.json","paper":"https://pith.science/paper/J6TN5AR2"},"agent_actions":{"view_html":"https://pith.science/pith/J6TN5AR2K2WGHXA3DMPYV4YWH3","download_json":"https://pith.science/pith/J6TN5AR2K2WGHXA3DMPYV4YWH3.json","view_paper":"https://pith.science/paper/J6TN5AR2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2212.01349&json=true","fetch_graph":"https://pith.science/api/pith-number/J6TN5AR2K2WGHXA3DMPYV4YWH3/graph.json","fetch_events":"https://pith.science/api/pith-number/J6TN5AR2K2WGHXA3DMPYV4YWH3/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/J6TN5AR2K2WGHXA3DMPYV4YWH3/action/timestamp_anchor","attest_storage":"https://pith.science/pith/J6TN5AR2K2WGHXA3DMPYV4YWH3/action/storage_attestation","attest_author":"https://pith.science/pith/J6TN5AR2K2WGHXA3DMPYV4YWH3/action/author_attestation","sign_citation":"https://pith.science/pith/J6TN5AR2K2WGHXA3DMPYV4YWH3/action/citation_signature","submit_replication":"https://pith.science/pith/J6TN5AR2K2WGHXA3DMPYV4YWH3/action/replication_record"}},"created_at":"2026-07-05T06:14:07.022986+00:00","updated_at":"2026-07-05T06:14:07.022986+00:00"}