{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:ZRHW2P7NNKUSL4HH6EABBUOM62","short_pith_number":"pith:ZRHW2P7N","schema_version":"1.0","canonical_sha256":"cc4f6d3fed6aa925f0e7f10010d1ccf69cbe6dd1a08f7ffbc3f37b5fbb87dac4","source":{"kind":"arxiv","id":"2506.04373","version":2},"attestation_state":"computed","paper":{"title":"Mechanistic Decomposition of Sentence Representations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Jonathan Michala, Juri Opitz, Matthieu Tehenan, Milton Lin, Vikram Natarajan","submitted_at":"2025-06-04T18:42:57Z","abstract_excerpt":"Sentence embeddings are central to modern NLP and AI systems, yet little is known about their internal structure. While we can compare these embeddings using measures such as cosine similarity, the contributing features are not human-interpretable, and the content of an embedding seems untraceable, as it is masked by complex neural transformations and a final pooling operation that combines individual token embeddings. To alleviate this issue, we propose a new method to mechanistically decompose sentence embeddings into interpretable components, by using dictionary learning on token-level repr"},"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":"2506.04373","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-06-04T18:42:57Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"801f10e58b36f1f80046ce37cbe3fcdfd29ea335a8c3a947aaaef935e6141b5c","abstract_canon_sha256":"2048a59660b6424162281b20a6658fe82d87c9fc79c2a6505d9e07c75b753d05"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:19:07.895435Z","signature_b64":"sAsXYoBJO6sBdvTDJg8Z+wOeeOGu6XJdbUKcArUWHQsgPx4ZLcgbigaR5ucTiZTYYlzNsGPr0QioMj6uc0o1Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cc4f6d3fed6aa925f0e7f10010d1ccf69cbe6dd1a08f7ffbc3f37b5fbb87dac4","last_reissued_at":"2026-07-05T11:19:07.894930Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:19:07.894930Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Mechanistic Decomposition of Sentence Representations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Jonathan Michala, Juri Opitz, Matthieu Tehenan, Milton Lin, Vikram Natarajan","submitted_at":"2025-06-04T18:42:57Z","abstract_excerpt":"Sentence embeddings are central to modern NLP and AI systems, yet little is known about their internal structure. While we can compare these embeddings using measures such as cosine similarity, the contributing features are not human-interpretable, and the content of an embedding seems untraceable, as it is masked by complex neural transformations and a final pooling operation that combines individual token embeddings. To alleviate this issue, we propose a new method to mechanistically decompose sentence embeddings into interpretable components, by using dictionary learning on token-level repr"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2506.04373","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/2506.04373/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":"2506.04373","created_at":"2026-07-05T11:19:07.894981+00:00"},{"alias_kind":"arxiv_version","alias_value":"2506.04373v2","created_at":"2026-07-05T11:19:07.894981+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2506.04373","created_at":"2026-07-05T11:19:07.894981+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZRHW2P7NNKUS","created_at":"2026-07-05T11:19:07.894981+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZRHW2P7NNKUSL4HH","created_at":"2026-07-05T11:19:07.894981+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZRHW2P7N","created_at":"2026-07-05T11:19:07.894981+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.07047","citing_title":"Riemannian Geometry for Pre-trained Language Model Embeddings","ref_index":52,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZRHW2P7NNKUSL4HH6EABBUOM62","json":"https://pith.science/pith/ZRHW2P7NNKUSL4HH6EABBUOM62.json","graph_json":"https://pith.science/api/pith-number/ZRHW2P7NNKUSL4HH6EABBUOM62/graph.json","events_json":"https://pith.science/api/pith-number/ZRHW2P7NNKUSL4HH6EABBUOM62/events.json","paper":"https://pith.science/paper/ZRHW2P7N"},"agent_actions":{"view_html":"https://pith.science/pith/ZRHW2P7NNKUSL4HH6EABBUOM62","download_json":"https://pith.science/pith/ZRHW2P7NNKUSL4HH6EABBUOM62.json","view_paper":"https://pith.science/paper/ZRHW2P7N","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2506.04373&json=true","fetch_graph":"https://pith.science/api/pith-number/ZRHW2P7NNKUSL4HH6EABBUOM62/graph.json","fetch_events":"https://pith.science/api/pith-number/ZRHW2P7NNKUSL4HH6EABBUOM62/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZRHW2P7NNKUSL4HH6EABBUOM62/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZRHW2P7NNKUSL4HH6EABBUOM62/action/storage_attestation","attest_author":"https://pith.science/pith/ZRHW2P7NNKUSL4HH6EABBUOM62/action/author_attestation","sign_citation":"https://pith.science/pith/ZRHW2P7NNKUSL4HH6EABBUOM62/action/citation_signature","submit_replication":"https://pith.science/pith/ZRHW2P7NNKUSL4HH6EABBUOM62/action/replication_record"}},"created_at":"2026-07-05T11:19:07.894981+00:00","updated_at":"2026-07-05T11:19:07.894981+00:00"}