{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2022:ULBQLMH3GPUDHOLMYLKAWVWU6D","short_pith_number":"pith:ULBQLMH3","schema_version":"1.0","canonical_sha256":"a2c305b0fb33e833b96cc2d40b56d4f0eea99af9816e3b83a636931d6f24aa62","source":{"kind":"arxiv","id":"2206.02608","version":1},"attestation_state":"computed","paper":{"title":"What do tokens know about their characters and how do they know it?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ayush Kaushal, Kyle Mahowald","submitted_at":"2022-06-06T13:27:26Z","abstract_excerpt":"Pre-trained language models (PLMs) that use subword tokenization schemes can succeed at a variety of language tasks that require character-level information, despite lacking explicit access to the character composition of tokens. Here, studying a range of models (e.g., GPT- J, BERT, RoBERTa, GloVe), we probe what word pieces encode about character-level information by training classifiers to predict the presence or absence of a particular alphabetical character in a token, based on its embedding (e.g., probing whether the model embedding for \"cat\" encodes that it contains the character \"a\"). W"},"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":"2206.02608","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2022-06-06T13:27:26Z","cross_cats_sorted":[],"title_canon_sha256":"a0b627cfbd43fe73d3ee8eb64d43c03d8afdf01c3eb532b1457c15ccd8e49676","abstract_canon_sha256":"828b41d62da5a677952996231cc2def3f07eeee41e2feee127b5a8e51e732de8"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T04:29:19.565530Z","signature_b64":"9DTl7EXk0HzVTlxxAwvEKL4y3AiBoElSCdr97Tbh1xjut70FIl0DR6EG3WgQw85q1CQv+7d4Qk+RLoRapyfMDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a2c305b0fb33e833b96cc2d40b56d4f0eea99af9816e3b83a636931d6f24aa62","last_reissued_at":"2026-07-05T04:29:19.565044Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T04:29:19.565044Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"What do tokens know about their characters and how do they know it?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ayush Kaushal, Kyle Mahowald","submitted_at":"2022-06-06T13:27:26Z","abstract_excerpt":"Pre-trained language models (PLMs) that use subword tokenization schemes can succeed at a variety of language tasks that require character-level information, despite lacking explicit access to the character composition of tokens. Here, studying a range of models (e.g., GPT- J, BERT, RoBERTa, GloVe), we probe what word pieces encode about character-level information by training classifiers to predict the presence or absence of a particular alphabetical character in a token, based on its embedding (e.g., probing whether the model embedding for \"cat\" encodes that it contains the character \"a\"). W"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2206.02608","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/2206.02608/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":"2206.02608","created_at":"2026-07-05T04:29:19.565133+00:00"},{"alias_kind":"arxiv_version","alias_value":"2206.02608v1","created_at":"2026-07-05T04:29:19.565133+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2206.02608","created_at":"2026-07-05T04:29:19.565133+00:00"},{"alias_kind":"pith_short_12","alias_value":"ULBQLMH3GPUD","created_at":"2026-07-05T04:29:19.565133+00:00"},{"alias_kind":"pith_short_16","alias_value":"ULBQLMH3GPUDHOLM","created_at":"2026-07-05T04:29:19.565133+00:00"},{"alias_kind":"pith_short_8","alias_value":"ULBQLMH3","created_at":"2026-07-05T04:29:19.565133+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2507.16323","citing_title":"SpeLLM: Character-Level Multi-Head Decoding","ref_index":11,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ULBQLMH3GPUDHOLMYLKAWVWU6D","json":"https://pith.science/pith/ULBQLMH3GPUDHOLMYLKAWVWU6D.json","graph_json":"https://pith.science/api/pith-number/ULBQLMH3GPUDHOLMYLKAWVWU6D/graph.json","events_json":"https://pith.science/api/pith-number/ULBQLMH3GPUDHOLMYLKAWVWU6D/events.json","paper":"https://pith.science/paper/ULBQLMH3"},"agent_actions":{"view_html":"https://pith.science/pith/ULBQLMH3GPUDHOLMYLKAWVWU6D","download_json":"https://pith.science/pith/ULBQLMH3GPUDHOLMYLKAWVWU6D.json","view_paper":"https://pith.science/paper/ULBQLMH3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2206.02608&json=true","fetch_graph":"https://pith.science/api/pith-number/ULBQLMH3GPUDHOLMYLKAWVWU6D/graph.json","fetch_events":"https://pith.science/api/pith-number/ULBQLMH3GPUDHOLMYLKAWVWU6D/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ULBQLMH3GPUDHOLMYLKAWVWU6D/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ULBQLMH3GPUDHOLMYLKAWVWU6D/action/storage_attestation","attest_author":"https://pith.science/pith/ULBQLMH3GPUDHOLMYLKAWVWU6D/action/author_attestation","sign_citation":"https://pith.science/pith/ULBQLMH3GPUDHOLMYLKAWVWU6D/action/citation_signature","submit_replication":"https://pith.science/pith/ULBQLMH3GPUDHOLMYLKAWVWU6D/action/replication_record"}},"created_at":"2026-07-05T04:29:19.565133+00:00","updated_at":"2026-07-05T04:29:19.565133+00:00"}