{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:MF57I23IESNIVLYFTYFFY5ECUS","short_pith_number":"pith:MF57I23I","schema_version":"1.0","canonical_sha256":"617bf46b68249a8aaf059e0a5c7482a4b3b72a5817826c5917c400dd69093e3a","source":{"kind":"arxiv","id":"2106.01087","version":2},"attestation_state":"computed","paper":{"title":"Is Sparse Attention more Interpretable?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Clara Meister, Isabelle Augenstein, Ryan Cotterell, Stefan Lazov","submitted_at":"2021-06-02T11:42:56Z","abstract_excerpt":"Sparse attention has been claimed to increase model interpretability under the assumption that it highlights influential inputs. Yet the attention distribution is typically over representations internal to the model rather than the inputs themselves, suggesting this assumption may not have merit. We build on the recent work exploring the interpretability of attention; we design a set of experiments to help us understand how sparsity affects our ability to use attention as an explainability tool. On three text classification tasks, we verify that only a weak relationship between inputs and co-i"},"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":"2106.01087","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-06-02T11:42:56Z","cross_cats_sorted":[],"title_canon_sha256":"e34388c0138acbf0ff8b1459102b7b940740fb255eb24b2428a181e6c5fdf486","abstract_canon_sha256":"5afb489488933fa00f9055534475c123d61f403a7d604d74564f89435ca59b52"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:47:08.498020Z","signature_b64":"lVro4GT6vj2HublY9eVjOsQ6++HlkOoO1mu5viISHgt2upS53/l6XcCJF+vkhHwONk/OyXuPkCH/4dKbXVPgBw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"617bf46b68249a8aaf059e0a5c7482a4b3b72a5817826c5917c400dd69093e3a","last_reissued_at":"2026-07-05T02:47:08.497623Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:47:08.497623Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Is Sparse Attention more Interpretable?","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Clara Meister, Isabelle Augenstein, Ryan Cotterell, Stefan Lazov","submitted_at":"2021-06-02T11:42:56Z","abstract_excerpt":"Sparse attention has been claimed to increase model interpretability under the assumption that it highlights influential inputs. Yet the attention distribution is typically over representations internal to the model rather than the inputs themselves, suggesting this assumption may not have merit. We build on the recent work exploring the interpretability of attention; we design a set of experiments to help us understand how sparsity affects our ability to use attention as an explainability tool. On three text classification tasks, we verify that only a weak relationship between inputs and co-i"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2106.01087","kind":"arxiv","version":2},"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/2106.01087/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":"2106.01087","created_at":"2026-07-05T02:47:08.497683+00:00"},{"alias_kind":"arxiv_version","alias_value":"2106.01087v2","created_at":"2026-07-05T02:47:08.497683+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2106.01087","created_at":"2026-07-05T02:47:08.497683+00:00"},{"alias_kind":"pith_short_12","alias_value":"MF57I23IESNI","created_at":"2026-07-05T02:47:08.497683+00:00"},{"alias_kind":"pith_short_16","alias_value":"MF57I23IESNIVLYF","created_at":"2026-07-05T02:47:08.497683+00:00"},{"alias_kind":"pith_short_8","alias_value":"MF57I23I","created_at":"2026-07-05T02:47:08.497683+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2411.11424","citing_title":"Membership Inference Attack against Long-Context Large Language Models","ref_index":17,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/MF57I23IESNIVLYFTYFFY5ECUS","json":"https://pith.science/pith/MF57I23IESNIVLYFTYFFY5ECUS.json","graph_json":"https://pith.science/api/pith-number/MF57I23IESNIVLYFTYFFY5ECUS/graph.json","events_json":"https://pith.science/api/pith-number/MF57I23IESNIVLYFTYFFY5ECUS/events.json","paper":"https://pith.science/paper/MF57I23I"},"agent_actions":{"view_html":"https://pith.science/pith/MF57I23IESNIVLYFTYFFY5ECUS","download_json":"https://pith.science/pith/MF57I23IESNIVLYFTYFFY5ECUS.json","view_paper":"https://pith.science/paper/MF57I23I","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2106.01087&json=true","fetch_graph":"https://pith.science/api/pith-number/MF57I23IESNIVLYFTYFFY5ECUS/graph.json","fetch_events":"https://pith.science/api/pith-number/MF57I23IESNIVLYFTYFFY5ECUS/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/MF57I23IESNIVLYFTYFFY5ECUS/action/timestamp_anchor","attest_storage":"https://pith.science/pith/MF57I23IESNIVLYFTYFFY5ECUS/action/storage_attestation","attest_author":"https://pith.science/pith/MF57I23IESNIVLYFTYFFY5ECUS/action/author_attestation","sign_citation":"https://pith.science/pith/MF57I23IESNIVLYFTYFFY5ECUS/action/citation_signature","submit_replication":"https://pith.science/pith/MF57I23IESNIVLYFTYFFY5ECUS/action/replication_record"}},"created_at":"2026-07-05T02:47:08.497683+00:00","updated_at":"2026-07-05T02:47:08.497683+00:00"}