{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:3TYEJPGUDSUCGLXQ3G75LYS3W4","short_pith_number":"pith:3TYEJPGU","schema_version":"1.0","canonical_sha256":"dcf044bcd41ca8232ef0d9bfd5e25bb73fa178ccc8e03df7aec6b39e1a72f930","source":{"kind":"arxiv","id":"2105.14652","version":1},"attestation_state":"computed","paper":{"title":"Attention Flows are Shapley Value Explanations","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dan Jurafsky, Kawin Ethayarajh","submitted_at":"2021-05-31T00:06:10Z","abstract_excerpt":"Shapley Values, a solution to the credit assignment problem in cooperative game theory, are a popular type of explanation in machine learning, having been used to explain the importance of features, embeddings, and even neurons. In NLP, however, leave-one-out and attention-based explanations still predominate. Can we draw a connection between these different methods? We formally prove that -- save for the degenerate case -- attention weights and leave-one-out values cannot be Shapley Values. $\\textit{Attention flow}$ is a post-processed variant of attention weights obtained by running the max-"},"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":"2105.14652","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-05-31T00:06:10Z","cross_cats_sorted":[],"title_canon_sha256":"d302733b30e9d06e2857851654d0b44a4c9843c5860df248e571f7c9c0f43ff4","abstract_canon_sha256":"6fb01bfa23dca581eb8af63ae4762675b7c2a59ca0c6e43d3ea69de9d3d866c6"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:44:39.822174Z","signature_b64":"QUlmgAdNPatvjUs31Nyyf2M23oJkfyAsS+/eXQFiZsjcXkjQNBFqCxneG41HVPegKCdhZHVwlw8U/QqzJmolDw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"dcf044bcd41ca8232ef0d9bfd5e25bb73fa178ccc8e03df7aec6b39e1a72f930","last_reissued_at":"2026-07-05T02:44:39.821763Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:44:39.821763Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Attention Flows are Shapley Value Explanations","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dan Jurafsky, Kawin Ethayarajh","submitted_at":"2021-05-31T00:06:10Z","abstract_excerpt":"Shapley Values, a solution to the credit assignment problem in cooperative game theory, are a popular type of explanation in machine learning, having been used to explain the importance of features, embeddings, and even neurons. In NLP, however, leave-one-out and attention-based explanations still predominate. Can we draw a connection between these different methods? We formally prove that -- save for the degenerate case -- attention weights and leave-one-out values cannot be Shapley Values. $\\textit{Attention flow}$ is a post-processed variant of attention weights obtained by running the max-"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2105.14652","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/2105.14652/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":"2105.14652","created_at":"2026-07-05T02:44:39.821824+00:00"},{"alias_kind":"arxiv_version","alias_value":"2105.14652v1","created_at":"2026-07-05T02:44:39.821824+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2105.14652","created_at":"2026-07-05T02:44:39.821824+00:00"},{"alias_kind":"pith_short_12","alias_value":"3TYEJPGUDSUC","created_at":"2026-07-05T02:44:39.821824+00:00"},{"alias_kind":"pith_short_16","alias_value":"3TYEJPGUDSUCGLXQ","created_at":"2026-07-05T02:44:39.821824+00:00"},{"alias_kind":"pith_short_8","alias_value":"3TYEJPGU","created_at":"2026-07-05T02:44:39.821824+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2501.06248","citing_title":"Utility-inspired Reward Transformations Improve Reinforcement Learning Training of Language Models","ref_index":7,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/3TYEJPGUDSUCGLXQ3G75LYS3W4","json":"https://pith.science/pith/3TYEJPGUDSUCGLXQ3G75LYS3W4.json","graph_json":"https://pith.science/api/pith-number/3TYEJPGUDSUCGLXQ3G75LYS3W4/graph.json","events_json":"https://pith.science/api/pith-number/3TYEJPGUDSUCGLXQ3G75LYS3W4/events.json","paper":"https://pith.science/paper/3TYEJPGU"},"agent_actions":{"view_html":"https://pith.science/pith/3TYEJPGUDSUCGLXQ3G75LYS3W4","download_json":"https://pith.science/pith/3TYEJPGUDSUCGLXQ3G75LYS3W4.json","view_paper":"https://pith.science/paper/3TYEJPGU","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2105.14652&json=true","fetch_graph":"https://pith.science/api/pith-number/3TYEJPGUDSUCGLXQ3G75LYS3W4/graph.json","fetch_events":"https://pith.science/api/pith-number/3TYEJPGUDSUCGLXQ3G75LYS3W4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/3TYEJPGUDSUCGLXQ3G75LYS3W4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/3TYEJPGUDSUCGLXQ3G75LYS3W4/action/storage_attestation","attest_author":"https://pith.science/pith/3TYEJPGUDSUCGLXQ3G75LYS3W4/action/author_attestation","sign_citation":"https://pith.science/pith/3TYEJPGUDSUCGLXQ3G75LYS3W4/action/citation_signature","submit_replication":"https://pith.science/pith/3TYEJPGUDSUCGLXQ3G75LYS3W4/action/replication_record"}},"created_at":"2026-07-05T02:44:39.821824+00:00","updated_at":"2026-07-05T02:44:39.821824+00:00"}