{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2019:SDUTATTLQ2YEVFYQF4PPVH55G6","short_pith_number":"pith:SDUTATTL","schema_version":"1.0","canonical_sha256":"90e9304e6b86b04a97102f1efa9fbd379e1f12a5bb8456cdb977e3e98fe6d932","source":{"kind":"arxiv","id":"1909.00453","version":2},"attestation_state":"computed","paper":{"title":"Topics to Avoid: Demoting Latent Confounds in Text Classification","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"Noah A. Smith, Sachin Kumar, Shuly Wintner, Yulia Tsvetkov","submitted_at":"2019-09-01T19:18:44Z","abstract_excerpt":"Despite impressive performance on many text classification tasks, deep neural networks tend to learn frequent superficial patterns that are specific to the training data and do not always generalize well. In this work, we observe this limitation with respect to the task of native language identification. We find that standard text classifiers which perform well on the test set end up learning topical features which are confounds of the prediction task (e.g., if the input text mentions Sweden, the classifier predicts that the author's native language is Swedish). We propose a method that repres"},"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":"1909.00453","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2019-09-01T19:18:44Z","cross_cats_sorted":["cs.CL","stat.ML"],"title_canon_sha256":"24cda80d19cde26cb5d396063c49f0cc390c5222adccb8a104615953900a6db1","abstract_canon_sha256":"fc2ef3ce2b702952780c93d67b921d42f468c6a971c35a729e2946edda57d884"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:49:09.161093Z","signature_b64":"YJoaBlQr5ZJ6G05ePDHOto1KlfDc0aKrOducXSNj4UAkB6rFHOHI/AxNC6AVff76PZcEXRGRY69Zntxr9N5ZBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"90e9304e6b86b04a97102f1efa9fbd379e1f12a5bb8456cdb977e3e98fe6d932","last_reissued_at":"2026-07-05T02:49:09.160573Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:49:09.160573Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Topics to Avoid: Demoting Latent Confounds in Text Classification","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL","stat.ML"],"primary_cat":"cs.LG","authors_text":"Noah A. Smith, Sachin Kumar, Shuly Wintner, Yulia Tsvetkov","submitted_at":"2019-09-01T19:18:44Z","abstract_excerpt":"Despite impressive performance on many text classification tasks, deep neural networks tend to learn frequent superficial patterns that are specific to the training data and do not always generalize well. In this work, we observe this limitation with respect to the task of native language identification. We find that standard text classifiers which perform well on the test set end up learning topical features which are confounds of the prediction task (e.g., if the input text mentions Sweden, the classifier predicts that the author's native language is Swedish). We propose a method that repres"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"1909.00453","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/1909.00453/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":"1909.00453","created_at":"2026-07-05T02:49:09.160632+00:00"},{"alias_kind":"arxiv_version","alias_value":"1909.00453v2","created_at":"2026-07-05T02:49:09.160632+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.1909.00453","created_at":"2026-07-05T02:49:09.160632+00:00"},{"alias_kind":"pith_short_12","alias_value":"SDUTATTLQ2YE","created_at":"2026-07-05T02:49:09.160632+00:00"},{"alias_kind":"pith_short_16","alias_value":"SDUTATTLQ2YEVFYQ","created_at":"2026-07-05T02:49:09.160632+00:00"},{"alias_kind":"pith_short_8","alias_value":"SDUTATTL","created_at":"2026-07-05T02:49:09.160632+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2509.05230","citing_title":"CURE: Controlled Unlearning for Robust Embeddings -- Mitigating Conceptual Shortcuts in Pre-Trained Language Models","ref_index":14,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/SDUTATTLQ2YEVFYQF4PPVH55G6","json":"https://pith.science/pith/SDUTATTLQ2YEVFYQF4PPVH55G6.json","graph_json":"https://pith.science/api/pith-number/SDUTATTLQ2YEVFYQF4PPVH55G6/graph.json","events_json":"https://pith.science/api/pith-number/SDUTATTLQ2YEVFYQF4PPVH55G6/events.json","paper":"https://pith.science/paper/SDUTATTL"},"agent_actions":{"view_html":"https://pith.science/pith/SDUTATTLQ2YEVFYQF4PPVH55G6","download_json":"https://pith.science/pith/SDUTATTLQ2YEVFYQF4PPVH55G6.json","view_paper":"https://pith.science/paper/SDUTATTL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=1909.00453&json=true","fetch_graph":"https://pith.science/api/pith-number/SDUTATTLQ2YEVFYQF4PPVH55G6/graph.json","fetch_events":"https://pith.science/api/pith-number/SDUTATTLQ2YEVFYQF4PPVH55G6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/SDUTATTLQ2YEVFYQF4PPVH55G6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/SDUTATTLQ2YEVFYQF4PPVH55G6/action/storage_attestation","attest_author":"https://pith.science/pith/SDUTATTLQ2YEVFYQF4PPVH55G6/action/author_attestation","sign_citation":"https://pith.science/pith/SDUTATTLQ2YEVFYQF4PPVH55G6/action/citation_signature","submit_replication":"https://pith.science/pith/SDUTATTLQ2YEVFYQF4PPVH55G6/action/replication_record"}},"created_at":"2026-07-05T02:49:09.160632+00:00","updated_at":"2026-07-05T02:49:09.160632+00:00"}