{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:YQXA5CKTSKXW5IK7DJEHEPOXQA","short_pith_number":"pith:YQXA5CKT","schema_version":"1.0","canonical_sha256":"c42e0e895392af6ea15f1a48723dd78016fb7c234e45c945dc191e274c6fbdd7","source":{"kind":"arxiv","id":"2012.15524","version":3},"attestation_state":"computed","paper":{"title":"Fast WordPiece Tokenization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Alex Salcianu, Dave Dopson, Denny Zhou, Xinying Song, Yang Song","submitted_at":"2020-12-31T10:01:29Z","abstract_excerpt":"Tokenization is a fundamental preprocessing step for almost all NLP tasks. In this paper, we propose efficient algorithms for the WordPiece tokenization used in BERT, from single-word tokenization to general text (e.g., sentence) tokenization. When tokenizing a single word, WordPiece uses a longest-match-first strategy, known as maximum matching. The best known algorithms so far are O(n^2) (where n is the input length) or O(nm) (where m is the maximum vocabulary token length). We propose a novel algorithm whose tokenization complexity is strictly O(n). Our method is inspired by the Aho-Corasic"},"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":"2012.15524","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2020-12-31T10:01:29Z","cross_cats_sorted":[],"title_canon_sha256":"f63178ae4b76b1af4ca92e76a4453e6f52bab295c5b8528faf7e3d02e7b336a1","abstract_canon_sha256":"2cb6fc0aa705af632cdfb46b943c8dde9bbe5da99344c9aaaaefc8c31ebb6c52"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:20:19.448583Z","signature_b64":"CxsTX2VHE/Oi7PT4/6Yvad4nw1GoyWFhxaW0zhMNL1c2fiPNASK6j7VQQDlxXHNmQWzOwlQDb/4yS50fa5h8Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"c42e0e895392af6ea15f1a48723dd78016fb7c234e45c945dc191e274c6fbdd7","last_reissued_at":"2026-07-05T03:20:19.448191Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:20:19.448191Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Fast WordPiece Tokenization","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Alex Salcianu, Dave Dopson, Denny Zhou, Xinying Song, Yang Song","submitted_at":"2020-12-31T10:01:29Z","abstract_excerpt":"Tokenization is a fundamental preprocessing step for almost all NLP tasks. In this paper, we propose efficient algorithms for the WordPiece tokenization used in BERT, from single-word tokenization to general text (e.g., sentence) tokenization. When tokenizing a single word, WordPiece uses a longest-match-first strategy, known as maximum matching. The best known algorithms so far are O(n^2) (where n is the input length) or O(nm) (where m is the maximum vocabulary token length). We propose a novel algorithm whose tokenization complexity is strictly O(n). Our method is inspired by the Aho-Corasic"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2012.15524","kind":"arxiv","version":3},"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/2012.15524/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":"2012.15524","created_at":"2026-07-05T03:20:19.448244+00:00"},{"alias_kind":"arxiv_version","alias_value":"2012.15524v3","created_at":"2026-07-05T03:20:19.448244+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2012.15524","created_at":"2026-07-05T03:20:19.448244+00:00"},{"alias_kind":"pith_short_12","alias_value":"YQXA5CKTSKXW","created_at":"2026-07-05T03:20:19.448244+00:00"},{"alias_kind":"pith_short_16","alias_value":"YQXA5CKTSKXW5IK7","created_at":"2026-07-05T03:20:19.448244+00:00"},{"alias_kind":"pith_short_8","alias_value":"YQXA5CKT","created_at":"2026-07-05T03:20:19.448244+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2509.07025","citing_title":"1 bit is all we need: binary normalized neural networks","ref_index":20,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/YQXA5CKTSKXW5IK7DJEHEPOXQA","json":"https://pith.science/pith/YQXA5CKTSKXW5IK7DJEHEPOXQA.json","graph_json":"https://pith.science/api/pith-number/YQXA5CKTSKXW5IK7DJEHEPOXQA/graph.json","events_json":"https://pith.science/api/pith-number/YQXA5CKTSKXW5IK7DJEHEPOXQA/events.json","paper":"https://pith.science/paper/YQXA5CKT"},"agent_actions":{"view_html":"https://pith.science/pith/YQXA5CKTSKXW5IK7DJEHEPOXQA","download_json":"https://pith.science/pith/YQXA5CKTSKXW5IK7DJEHEPOXQA.json","view_paper":"https://pith.science/paper/YQXA5CKT","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2012.15524&json=true","fetch_graph":"https://pith.science/api/pith-number/YQXA5CKTSKXW5IK7DJEHEPOXQA/graph.json","fetch_events":"https://pith.science/api/pith-number/YQXA5CKTSKXW5IK7DJEHEPOXQA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/YQXA5CKTSKXW5IK7DJEHEPOXQA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/YQXA5CKTSKXW5IK7DJEHEPOXQA/action/storage_attestation","attest_author":"https://pith.science/pith/YQXA5CKTSKXW5IK7DJEHEPOXQA/action/author_attestation","sign_citation":"https://pith.science/pith/YQXA5CKTSKXW5IK7DJEHEPOXQA/action/citation_signature","submit_replication":"https://pith.science/pith/YQXA5CKTSKXW5IK7DJEHEPOXQA/action/replication_record"}},"created_at":"2026-07-05T03:20:19.448244+00:00","updated_at":"2026-07-05T03:20:19.448244+00:00"}