{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2018:FK5RK5H7FQVZYLPJTSKKMS2VNK","short_pith_number":"pith:FK5RK5H7","schema_version":"1.0","canonical_sha256":"2abb1574ff2c2b9c2de99c94a64b556aa7e2007d886704817b56440b7189a72b","source":{"kind":"arxiv","id":"1808.08079","version":3},"attestation_state":"computed","paper":{"title":"Under the Hood: Using Diagnostic Classifiers to Investigate and Improve how Language Models Track Agreement Information","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Dieuwke Hupkes, Florian Mohnert, Jacqueline Harding, Mario Giulianelli, Willem Zuidema","submitted_at":"2018-08-24T10:29:45Z","abstract_excerpt":"How do neural language models keep track of number agreement between subject and verb? We show that `diagnostic classifiers', trained to predict number from the internal states of a language model, provide a detailed understanding of how, when, and where this information is represented. Moreover, they give us insight into when and where number information is corrupted in cases where the language model ends up making agreement errors. To demonstrate the causal role played by the representations we find, we then use agreement information to influence the course of the LSTM during the processing "},"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":"1808.08079","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2018-08-24T10:29:45Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"b6baa76e735c321cd9a09d7b0ac91026533ac0f8ae223c9be3bc59fa141e14b2","abstract_canon_sha256":"5eaf3a1d73134ec0bbb391dbc0631a0f37ee56bd6274e61b5f1ac277e1f301a9"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:33:01.322279Z","signature_b64":"kKQAzwsLY+zRoPjWZ4gzvBMYNGsBaRWvY3vQAbbb8QFCacrxqfiEe9kZ1/KVp65hR0dOCRP+gGcatDf5vmIvCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2abb1574ff2c2b9c2de99c94a64b556aa7e2007d886704817b56440b7189a72b","last_reissued_at":"2026-07-05T03:33:01.321867Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:33:01.321867Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Under the Hood: Using Diagnostic Classifiers to Investigate and Improve how Language Models Track Agreement Information","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Dieuwke Hupkes, Florian Mohnert, Jacqueline Harding, Mario Giulianelli, Willem Zuidema","submitted_at":"2018-08-24T10:29:45Z","abstract_excerpt":"How do neural language models keep track of number agreement between subject and verb? We show that `diagnostic classifiers', trained to predict number from the internal states of a language model, provide a detailed understanding of how, when, and where this information is represented. Moreover, they give us insight into when and where number information is corrupted in cases where the language model ends up making agreement errors. To demonstrate the causal role played by the representations we find, we then use agreement information to influence the course of the LSTM during the processing "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1808.08079","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/1808.08079/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":"1808.08079","created_at":"2026-07-05T03:33:01.321924+00:00"},{"alias_kind":"arxiv_version","alias_value":"1808.08079v3","created_at":"2026-07-05T03:33:01.321924+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1808.08079","created_at":"2026-07-05T03:33:01.321924+00:00"},{"alias_kind":"pith_short_12","alias_value":"FK5RK5H7FQVZ","created_at":"2026-07-05T03:33:01.321924+00:00"},{"alias_kind":"pith_short_16","alias_value":"FK5RK5H7FQVZYLPJ","created_at":"2026-07-05T03:33:01.321924+00:00"},{"alias_kind":"pith_short_8","alias_value":"FK5RK5H7","created_at":"2026-07-05T03:33:01.321924+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2605.18537","citing_title":"Probing for Representation Manifolds in Superposition","ref_index":65,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05892","citing_title":"Beyond Steering Vector: Flow-based Activation Steering for Inference-Time Intervention","ref_index":13,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FK5RK5H7FQVZYLPJTSKKMS2VNK","json":"https://pith.science/pith/FK5RK5H7FQVZYLPJTSKKMS2VNK.json","graph_json":"https://pith.science/api/pith-number/FK5RK5H7FQVZYLPJTSKKMS2VNK/graph.json","events_json":"https://pith.science/api/pith-number/FK5RK5H7FQVZYLPJTSKKMS2VNK/events.json","paper":"https://pith.science/paper/FK5RK5H7"},"agent_actions":{"view_html":"https://pith.science/pith/FK5RK5H7FQVZYLPJTSKKMS2VNK","download_json":"https://pith.science/pith/FK5RK5H7FQVZYLPJTSKKMS2VNK.json","view_paper":"https://pith.science/paper/FK5RK5H7","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1808.08079&json=true","fetch_graph":"https://pith.science/api/pith-number/FK5RK5H7FQVZYLPJTSKKMS2VNK/graph.json","fetch_events":"https://pith.science/api/pith-number/FK5RK5H7FQVZYLPJTSKKMS2VNK/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FK5RK5H7FQVZYLPJTSKKMS2VNK/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FK5RK5H7FQVZYLPJTSKKMS2VNK/action/storage_attestation","attest_author":"https://pith.science/pith/FK5RK5H7FQVZYLPJTSKKMS2VNK/action/author_attestation","sign_citation":"https://pith.science/pith/FK5RK5H7FQVZYLPJTSKKMS2VNK/action/citation_signature","submit_replication":"https://pith.science/pith/FK5RK5H7FQVZYLPJTSKKMS2VNK/action/replication_record"}},"created_at":"2026-07-05T03:33:01.321924+00:00","updated_at":"2026-07-05T03:33:01.321924+00:00"}