{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:DPJBQGP7IAZNGJBIVCXOALP2VU","short_pith_number":"pith:DPJBQGP7","schema_version":"1.0","canonical_sha256":"1bd21819ff4032d32428a8aee02dfaad0cb841499a6074ab12bdf01d7a3fab19","source":{"kind":"arxiv","id":"2506.03149","version":1},"attestation_state":"computed","paper":{"title":"Causal Estimation of Tokenisation Bias","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Andreas Vlachos, Clara Meister, Pietro Lesci, Thomas Hofmann, Tiago Pimentel","submitted_at":"2025-06-03T17:59:47Z","abstract_excerpt":"Modern language models are typically trained over subword sequences, but ultimately define probabilities over character-strings. Ideally, the choice of the tokeniser -- which maps character-strings to subwords -- should not affect the probability assigned to the underlying character-string; in practice, it does. We define this mismatch as tokenisation bias. In this work, we quantify one particular type of tokenisation bias: the effect of including or not a subword (e.g., $\\langle hello \\rangle$) in a tokeniser's vocabulary on the probability a trained model assigns to the corresponding charact"},"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":"2506.03149","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-03T17:59:47Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"94bcea0c91351337a1a823d179313b29d41c6d7ae9a9b1b40cad595ae3c397bd","abstract_canon_sha256":"eaab52ca775cb3ab00fdfdf70f83d84eeab1661316f183e2254635a4a2cd7344"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:15:15.732023Z","signature_b64":"VPtIy//iiwYxVilKfq9x04w7HMuHHUr80R25cNz+TiWs7b6jU6tcnWxboUkzQmDzu/bYAxqgRH3yaEiMyk4KCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"1bd21819ff4032d32428a8aee02dfaad0cb841499a6074ab12bdf01d7a3fab19","last_reissued_at":"2026-07-05T11:15:15.731538Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:15:15.731538Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Causal Estimation of Tokenisation Bias","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Andreas Vlachos, Clara Meister, Pietro Lesci, Thomas Hofmann, Tiago Pimentel","submitted_at":"2025-06-03T17:59:47Z","abstract_excerpt":"Modern language models are typically trained over subword sequences, but ultimately define probabilities over character-strings. Ideally, the choice of the tokeniser -- which maps character-strings to subwords -- should not affect the probability assigned to the underlying character-string; in practice, it does. We define this mismatch as tokenisation bias. In this work, we quantify one particular type of tokenisation bias: the effect of including or not a subword (e.g., $\\langle hello \\rangle$) in a tokeniser's vocabulary on the probability a trained model assigns to the corresponding charact"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.03149","kind":"arxiv","version":1},"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/2506.03149/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":"2506.03149","created_at":"2026-07-05T11:15:15.731595+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.03149v1","created_at":"2026-07-05T11:15:15.731595+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.03149","created_at":"2026-07-05T11:15:15.731595+00:00"},{"alias_kind":"pith_short_12","alias_value":"DPJBQGP7IAZN","created_at":"2026-07-05T11:15:15.731595+00:00"},{"alias_kind":"pith_short_16","alias_value":"DPJBQGP7IAZNGJBI","created_at":"2026-07-05T11:15:15.731595+00:00"},{"alias_kind":"pith_short_8","alias_value":"DPJBQGP7","created_at":"2026-07-05T11:15:15.731595+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.05029","citing_title":"Validity Threats for Foundation Model Research","ref_index":55,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/DPJBQGP7IAZNGJBIVCXOALP2VU","json":"https://pith.science/pith/DPJBQGP7IAZNGJBIVCXOALP2VU.json","graph_json":"https://pith.science/api/pith-number/DPJBQGP7IAZNGJBIVCXOALP2VU/graph.json","events_json":"https://pith.science/api/pith-number/DPJBQGP7IAZNGJBIVCXOALP2VU/events.json","paper":"https://pith.science/paper/DPJBQGP7"},"agent_actions":{"view_html":"https://pith.science/pith/DPJBQGP7IAZNGJBIVCXOALP2VU","download_json":"https://pith.science/pith/DPJBQGP7IAZNGJBIVCXOALP2VU.json","view_paper":"https://pith.science/paper/DPJBQGP7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.03149&json=true","fetch_graph":"https://pith.science/api/pith-number/DPJBQGP7IAZNGJBIVCXOALP2VU/graph.json","fetch_events":"https://pith.science/api/pith-number/DPJBQGP7IAZNGJBIVCXOALP2VU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/DPJBQGP7IAZNGJBIVCXOALP2VU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/DPJBQGP7IAZNGJBIVCXOALP2VU/action/storage_attestation","attest_author":"https://pith.science/pith/DPJBQGP7IAZNGJBIVCXOALP2VU/action/author_attestation","sign_citation":"https://pith.science/pith/DPJBQGP7IAZNGJBIVCXOALP2VU/action/citation_signature","submit_replication":"https://pith.science/pith/DPJBQGP7IAZNGJBIVCXOALP2VU/action/replication_record"}},"created_at":"2026-07-05T11:15:15.731595+00:00","updated_at":"2026-07-05T11:15:15.731595+00:00"}