{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:Q2TKKCDKRHVKRJWDONHTBCS6SK","short_pith_number":"pith:Q2TKKCDK","schema_version":"1.0","canonical_sha256":"86a6a5086a89eaa8a6c3734f308a5e9296af1cc1348c4f37a5b321646270b9b7","source":{"kind":"arxiv","id":"2103.07601","version":3},"attestation_state":"computed","paper":{"title":"Approximating How Single Head Attention Learns","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Charlie Snell, Dan Klein, Jacob Steinhardt, Ruiqi Zhong","submitted_at":"2021-03-13T02:32:19Z","abstract_excerpt":"Why do models often attend to salient words, and how does this evolve throughout training? We approximate model training as a two stage process: early on in training when the attention weights are uniform, the model learns to translate individual input word `i` to `o` if they co-occur frequently. Later, the model learns to attend to `i` while the correct output is $o$ because it knows `i` translates to `o`. To formalize, we define a model property, Knowledge to Translate Individual Words (KTIW) (e.g. knowing that `i` translates to `o`), and claim that it drives the learning of the attention. T"},"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":"2103.07601","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2021-03-13T02:32:19Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"98dfc9ed6f47cc8c3fc187f3fec26fb284e9dabb3ee3312a1246aa1bd64e3a0c","abstract_canon_sha256":"2e6ea1754fd06e00258ead336f77e6cbfc0de6a4fc32d7f0273776e35561bc89"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:24:28.701849Z","signature_b64":"FcPCkbCkn/jfyZMdORRgwj+039qxsTAh3ONU0dL6rs67mVJdRuWBFLz47w56H7CYtQt6XvhB3DzSJeFMzThhBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"86a6a5086a89eaa8a6c3734f308a5e9296af1cc1348c4f37a5b321646270b9b7","last_reissued_at":"2026-07-05T03:24:28.701302Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:24:28.701302Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Approximating How Single Head Attention Learns","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Charlie Snell, Dan Klein, Jacob Steinhardt, Ruiqi Zhong","submitted_at":"2021-03-13T02:32:19Z","abstract_excerpt":"Why do models often attend to salient words, and how does this evolve throughout training? We approximate model training as a two stage process: early on in training when the attention weights are uniform, the model learns to translate individual input word `i` to `o` if they co-occur frequently. Later, the model learns to attend to `i` while the correct output is $o$ because it knows `i` translates to `o`. To formalize, we define a model property, Knowledge to Translate Individual Words (KTIW) (e.g. knowing that `i` translates to `o`), and claim that it drives the learning of the attention. T"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2103.07601","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/2103.07601/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":"2103.07601","created_at":"2026-07-05T03:24:28.701395+00:00"},{"alias_kind":"arxiv_version","alias_value":"2103.07601v3","created_at":"2026-07-05T03:24:28.701395+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2103.07601","created_at":"2026-07-05T03:24:28.701395+00:00"},{"alias_kind":"pith_short_12","alias_value":"Q2TKKCDKRHVK","created_at":"2026-07-05T03:24:28.701395+00:00"},{"alias_kind":"pith_short_16","alias_value":"Q2TKKCDKRHVKRJWD","created_at":"2026-07-05T03:24:28.701395+00:00"},{"alias_kind":"pith_short_8","alias_value":"Q2TKKCDK","created_at":"2026-07-05T03:24:28.701395+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/Q2TKKCDKRHVKRJWDONHTBCS6SK","json":"https://pith.science/pith/Q2TKKCDKRHVKRJWDONHTBCS6SK.json","graph_json":"https://pith.science/api/pith-number/Q2TKKCDKRHVKRJWDONHTBCS6SK/graph.json","events_json":"https://pith.science/api/pith-number/Q2TKKCDKRHVKRJWDONHTBCS6SK/events.json","paper":"https://pith.science/paper/Q2TKKCDK"},"agent_actions":{"view_html":"https://pith.science/pith/Q2TKKCDKRHVKRJWDONHTBCS6SK","download_json":"https://pith.science/pith/Q2TKKCDKRHVKRJWDONHTBCS6SK.json","view_paper":"https://pith.science/paper/Q2TKKCDK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2103.07601&json=true","fetch_graph":"https://pith.science/api/pith-number/Q2TKKCDKRHVKRJWDONHTBCS6SK/graph.json","fetch_events":"https://pith.science/api/pith-number/Q2TKKCDKRHVKRJWDONHTBCS6SK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/Q2TKKCDKRHVKRJWDONHTBCS6SK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/Q2TKKCDKRHVKRJWDONHTBCS6SK/action/storage_attestation","attest_author":"https://pith.science/pith/Q2TKKCDKRHVKRJWDONHTBCS6SK/action/author_attestation","sign_citation":"https://pith.science/pith/Q2TKKCDKRHVKRJWDONHTBCS6SK/action/citation_signature","submit_replication":"https://pith.science/pith/Q2TKKCDKRHVKRJWDONHTBCS6SK/action/replication_record"}},"created_at":"2026-07-05T03:24:28.701395+00:00","updated_at":"2026-07-05T03:24:28.701395+00:00"}