{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2019:G7P6DUPKW7OB6IQYCPAEJI46KR","short_pith_number":"pith:G7P6DUPK","schema_version":"1.0","canonical_sha256":"37dfe1d1eab7dc1f221813c044a39e545136783b2544c0d89c5e621f9a9dae4b","source":{"kind":"arxiv","id":"1909.03368","version":1},"attestation_state":"computed","paper":{"title":"Designing and Interpreting Probes with Control Tasks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"John Hewitt, Percy Liang","submitted_at":"2019-09-08T02:04:32Z","abstract_excerpt":"Probes, supervised models trained to predict properties (like parts-of-speech) from representations (like ELMo), have achieved high accuracy on a range of linguistic tasks. But does this mean that the representations encode linguistic structure or just that the probe has learned the linguistic task? In this paper, we propose control tasks, which associate word types with random outputs, to complement linguistic tasks. By construction, these tasks can only be learned by the probe itself. So a good probe, (one that reflects the representation), should be selective, achieving high linguistic task"},"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":"1909.03368","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2019-09-08T02:04:32Z","cross_cats_sorted":[],"title_canon_sha256":"268749e1cf97d5705e1513fb3c5120283e4108a45525851739d453cb70743973","abstract_canon_sha256":"d1aa9f449d7f7f16a1a9d06d44d11a13f49cca5b7c79757f75691a4e58a5cc6b"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T00:03:04.892673Z","signature_b64":"RLD8jQJhs0FxKnPp+GclFIQ5y1upZWRH0fI/O2VVwZQHiGZwgEKQVUjGV8xrxmOywE/zXnZNhnD7GeD5nhKfDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"37dfe1d1eab7dc1f221813c044a39e545136783b2544c0d89c5e621f9a9dae4b","last_reissued_at":"2026-07-05T00:03:04.892229Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T00:03:04.892229Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Designing and Interpreting Probes with Control Tasks","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"John Hewitt, Percy Liang","submitted_at":"2019-09-08T02:04:32Z","abstract_excerpt":"Probes, supervised models trained to predict properties (like parts-of-speech) from representations (like ELMo), have achieved high accuracy on a range of linguistic tasks. But does this mean that the representations encode linguistic structure or just that the probe has learned the linguistic task? In this paper, we propose control tasks, which associate word types with random outputs, to complement linguistic tasks. By construction, these tasks can only be learned by the probe itself. So a good probe, (one that reflects the representation), should be selective, achieving high linguistic task"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1909.03368","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/1909.03368/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":"1909.03368","created_at":"2026-07-05T00:03:04.892291+00:00"},{"alias_kind":"arxiv_version","alias_value":"1909.03368v1","created_at":"2026-07-05T00:03:04.892291+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1909.03368","created_at":"2026-07-05T00:03:04.892291+00:00"},{"alias_kind":"pith_short_12","alias_value":"G7P6DUPKW7OB","created_at":"2026-07-05T00:03:04.892291+00:00"},{"alias_kind":"pith_short_16","alias_value":"G7P6DUPKW7OB6IQY","created_at":"2026-07-05T00:03:04.892291+00:00"},{"alias_kind":"pith_short_8","alias_value":"G7P6DUPK","created_at":"2026-07-05T00:03:04.892291+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":6,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.09646","citing_title":"Do Video Foundation Models Understand Intuitive Physics? A Layerwise Probing Analysis","ref_index":12,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09881","citing_title":"Toward Calibrated, Fair, and accurate Deepfake Detection","ref_index":27,"is_internal_anchor":false},{"citing_arxiv_id":"2606.09563","citing_title":"PRISM: Recovering Instruction Sets from Language Model Activations","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2602.20338","citing_title":"Emergent Manifold Separability during Reasoning in Large Language Models","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2604.24693","citing_title":"Contextual Linear Activation Steering of Language Models","ref_index":18,"is_internal_anchor":false},{"citing_arxiv_id":"2605.05741","citing_title":"HyperLens: Quantifying Cognitive Effort in LLMs with Fine-grained Confidence Trajectory","ref_index":31,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/G7P6DUPKW7OB6IQYCPAEJI46KR","json":"https://pith.science/pith/G7P6DUPKW7OB6IQYCPAEJI46KR.json","graph_json":"https://pith.science/api/pith-number/G7P6DUPKW7OB6IQYCPAEJI46KR/graph.json","events_json":"https://pith.science/api/pith-number/G7P6DUPKW7OB6IQYCPAEJI46KR/events.json","paper":"https://pith.science/paper/G7P6DUPK"},"agent_actions":{"view_html":"https://pith.science/pith/G7P6DUPKW7OB6IQYCPAEJI46KR","download_json":"https://pith.science/pith/G7P6DUPKW7OB6IQYCPAEJI46KR.json","view_paper":"https://pith.science/paper/G7P6DUPK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1909.03368&json=true","fetch_graph":"https://pith.science/api/pith-number/G7P6DUPKW7OB6IQYCPAEJI46KR/graph.json","fetch_events":"https://pith.science/api/pith-number/G7P6DUPKW7OB6IQYCPAEJI46KR/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/G7P6DUPKW7OB6IQYCPAEJI46KR/action/timestamp_anchor","attest_storage":"https://pith.science/pith/G7P6DUPKW7OB6IQYCPAEJI46KR/action/storage_attestation","attest_author":"https://pith.science/pith/G7P6DUPKW7OB6IQYCPAEJI46KR/action/author_attestation","sign_citation":"https://pith.science/pith/G7P6DUPKW7OB6IQYCPAEJI46KR/action/citation_signature","submit_replication":"https://pith.science/pith/G7P6DUPKW7OB6IQYCPAEJI46KR/action/replication_record"}},"created_at":"2026-07-05T00:03:04.892291+00:00","updated_at":"2026-07-05T00:03:04.892291+00:00"}