{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:M3MP4PLBGL32CSJZEVP7P7SUCD","short_pith_number":"pith:M3MP4PLB","schema_version":"1.0","canonical_sha256":"66d8fe3d6132f7a14939255ff7fe5410e7ce6d68dee196ba18542d1fdb6485b5","source":{"kind":"arxiv","id":"2410.12924","version":3},"attestation_state":"computed","paper":{"title":"Interpreting token compositionality in LLMs: A robustness analysis","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Andr\\'e Freitas, Danilo S. Carvalho, Nura Aljaafari","submitted_at":"2024-10-16T18:10:50Z","abstract_excerpt":"Understanding the internal mechanisms of large language models (LLMs) is integral to enhancing their reliability, interpretability, and inference processes. We present Constituent-Aware Pooling (CAP), a methodology designed to analyse how LLMs process compositional linguistic structures. Grounded in principles of compositionality, mechanistic interpretability, and information theory, CAP systematically intervenes in model activations through constituent-based pooling at various model levels. Our experiments on inverse definition modelling, hypernym and synonym prediction reveal critical insigh"},"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":"2410.12924","kind":"arxiv","version":3},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","primary_cat":"cs.CL","submitted_at":"2024-10-16T18:10:50Z","cross_cats_sorted":[],"title_canon_sha256":"848d56e70c01e024dc2e2da516cd65f98d93b8208d01d5c4a7cc0552231c2f47","abstract_canon_sha256":"8e86f62f7c349f50a286c0a554f997640978361a5f7625303a28573bcf132ef2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:05:35.612919Z","signature_b64":"IspVx+NTGnJWJNZHp9HRLWFeIa44fh81dCg0tlMXj073yeDnYr2MCg98jUoz1db2wKEemBDvw9SpaNlKPVF0Aw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"66d8fe3d6132f7a14939255ff7fe5410e7ce6d68dee196ba18542d1fdb6485b5","last_reissued_at":"2026-07-05T11:05:35.612466Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:05:35.612466Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Interpreting token compositionality in LLMs: A robustness analysis","license":"http://creativecommons.org/licenses/by-nc-sa/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Andr\\'e Freitas, Danilo S. Carvalho, Nura Aljaafari","submitted_at":"2024-10-16T18:10:50Z","abstract_excerpt":"Understanding the internal mechanisms of large language models (LLMs) is integral to enhancing their reliability, interpretability, and inference processes. We present Constituent-Aware Pooling (CAP), a methodology designed to analyse how LLMs process compositional linguistic structures. Grounded in principles of compositionality, mechanistic interpretability, and information theory, CAP systematically intervenes in model activations through constituent-based pooling at various model levels. Our experiments on inverse definition modelling, hypernym and synonym prediction reveal critical insigh"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2410.12924","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/2410.12924/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":"2410.12924","created_at":"2026-07-05T11:05:35.612523+00:00"},{"alias_kind":"arxiv_version","alias_value":"2410.12924v3","created_at":"2026-07-05T11:05:35.612523+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2410.12924","created_at":"2026-07-05T11:05:35.612523+00:00"},{"alias_kind":"pith_short_12","alias_value":"M3MP4PLBGL32","created_at":"2026-07-05T11:05:35.612523+00:00"},{"alias_kind":"pith_short_16","alias_value":"M3MP4PLBGL32CSJZ","created_at":"2026-07-05T11:05:35.612523+00:00"},{"alias_kind":"pith_short_8","alias_value":"M3MP4PLB","created_at":"2026-07-05T11:05:35.612523+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2506.15629","citing_title":"Revisiting Compositional Generalization Capability of Large Language Models Considering Instruction Following Ability","ref_index":2024,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/M3MP4PLBGL32CSJZEVP7P7SUCD","json":"https://pith.science/pith/M3MP4PLBGL32CSJZEVP7P7SUCD.json","graph_json":"https://pith.science/api/pith-number/M3MP4PLBGL32CSJZEVP7P7SUCD/graph.json","events_json":"https://pith.science/api/pith-number/M3MP4PLBGL32CSJZEVP7P7SUCD/events.json","paper":"https://pith.science/paper/M3MP4PLB"},"agent_actions":{"view_html":"https://pith.science/pith/M3MP4PLBGL32CSJZEVP7P7SUCD","download_json":"https://pith.science/pith/M3MP4PLBGL32CSJZEVP7P7SUCD.json","view_paper":"https://pith.science/paper/M3MP4PLB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2410.12924&json=true","fetch_graph":"https://pith.science/api/pith-number/M3MP4PLBGL32CSJZEVP7P7SUCD/graph.json","fetch_events":"https://pith.science/api/pith-number/M3MP4PLBGL32CSJZEVP7P7SUCD/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/M3MP4PLBGL32CSJZEVP7P7SUCD/action/timestamp_anchor","attest_storage":"https://pith.science/pith/M3MP4PLBGL32CSJZEVP7P7SUCD/action/storage_attestation","attest_author":"https://pith.science/pith/M3MP4PLBGL32CSJZEVP7P7SUCD/action/author_attestation","sign_citation":"https://pith.science/pith/M3MP4PLBGL32CSJZEVP7P7SUCD/action/citation_signature","submit_replication":"https://pith.science/pith/M3MP4PLBGL32CSJZEVP7P7SUCD/action/replication_record"}},"created_at":"2026-07-05T11:05:35.612523+00:00","updated_at":"2026-07-05T11:05:35.612523+00:00"}