{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2020:VR47DM43J2AT5CMVTUNSZTQ3PG","short_pith_number":"pith:VR47DM43","schema_version":"1.0","canonical_sha256":"ac79f1b39b4e813e89959d1b2cce1b79a02a1003f1244085d56d69bbbed855ee","source":{"kind":"arxiv","id":"2004.14602","version":4},"attestation_state":"computed","paper":{"title":"Look at the First Sentence: Position Bias in Question Answering","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Gangwoo Kim, Hyunjae Kim, Jaewoo Kang, Jinhyuk Lee, Miyoung Ko","submitted_at":"2020-04-30T06:25:16Z","abstract_excerpt":"Many extractive question answering models are trained to predict start and end positions of answers. The choice of predicting answers as positions is mainly due to its simplicity and effectiveness. In this study, we hypothesize that when the distribution of the answer positions is highly skewed in the training set (e.g., answers lie only in the k-th sentence of each passage), QA models predicting answers as positions can learn spurious positional cues and fail to give answers in different positions. We first illustrate this position bias in popular extractive QA models such as BiDAF and BERT a"},"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":"2004.14602","kind":"arxiv","version":4},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2020-04-30T06:25:16Z","cross_cats_sorted":[],"title_canon_sha256":"c26f0057c217575c210942fedbb7dccc44c76d55407cd9cb44be01b9fbdb61e0","abstract_canon_sha256":"c045b47d7bfdb270095b1db588db1b0b94a49a6bd38bae25da4c5a28dd523086"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:20:54.393070Z","signature_b64":"RcQjwGiod1GBH/BmXC+KshcuBbUXZOPgTOegS3E7OETrZWsC3ZYAmLFYuSpiMH3mW/QiCbxBB7jRqAoihWUhBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ac79f1b39b4e813e89959d1b2cce1b79a02a1003f1244085d56d69bbbed855ee","last_reissued_at":"2026-07-05T02:20:54.392579Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:20:54.392579Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Look at the First Sentence: Position Bias in Question Answering","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Gangwoo Kim, Hyunjae Kim, Jaewoo Kang, Jinhyuk Lee, Miyoung Ko","submitted_at":"2020-04-30T06:25:16Z","abstract_excerpt":"Many extractive question answering models are trained to predict start and end positions of answers. The choice of predicting answers as positions is mainly due to its simplicity and effectiveness. In this study, we hypothesize that when the distribution of the answer positions is highly skewed in the training set (e.g., answers lie only in the k-th sentence of each passage), QA models predicting answers as positions can learn spurious positional cues and fail to give answers in different positions. We first illustrate this position bias in popular extractive QA models such as BiDAF and BERT a"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2004.14602","kind":"arxiv","version":4},"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/2004.14602/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":"2004.14602","created_at":"2026-07-05T02:20:54.392635+00:00"},{"alias_kind":"arxiv_version","alias_value":"2004.14602v4","created_at":"2026-07-05T02:20:54.392635+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2004.14602","created_at":"2026-07-05T02:20:54.392635+00:00"},{"alias_kind":"pith_short_12","alias_value":"VR47DM43J2AT","created_at":"2026-07-05T02:20:54.392635+00:00"},{"alias_kind":"pith_short_16","alias_value":"VR47DM43J2AT5CMV","created_at":"2026-07-05T02:20:54.392635+00:00"},{"alias_kind":"pith_short_8","alias_value":"VR47DM43","created_at":"2026-07-05T02:20:54.392635+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.22329","citing_title":"BabelJudge: Measuring LLM-as-a-Judge Reliability Across Languages and Agent Trajectories","ref_index":8,"is_internal_anchor":false},{"citing_arxiv_id":"2412.05579","citing_title":"LLMs-as-Judges: A Comprehensive Survey on LLM-based Evaluation Methods","ref_index":115,"is_internal_anchor":false},{"citing_arxiv_id":"2306.05685","citing_title":"Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena","ref_index":22,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VR47DM43J2AT5CMVTUNSZTQ3PG","json":"https://pith.science/pith/VR47DM43J2AT5CMVTUNSZTQ3PG.json","graph_json":"https://pith.science/api/pith-number/VR47DM43J2AT5CMVTUNSZTQ3PG/graph.json","events_json":"https://pith.science/api/pith-number/VR47DM43J2AT5CMVTUNSZTQ3PG/events.json","paper":"https://pith.science/paper/VR47DM43"},"agent_actions":{"view_html":"https://pith.science/pith/VR47DM43J2AT5CMVTUNSZTQ3PG","download_json":"https://pith.science/pith/VR47DM43J2AT5CMVTUNSZTQ3PG.json","view_paper":"https://pith.science/paper/VR47DM43","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2004.14602&json=true","fetch_graph":"https://pith.science/api/pith-number/VR47DM43J2AT5CMVTUNSZTQ3PG/graph.json","fetch_events":"https://pith.science/api/pith-number/VR47DM43J2AT5CMVTUNSZTQ3PG/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VR47DM43J2AT5CMVTUNSZTQ3PG/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VR47DM43J2AT5CMVTUNSZTQ3PG/action/storage_attestation","attest_author":"https://pith.science/pith/VR47DM43J2AT5CMVTUNSZTQ3PG/action/author_attestation","sign_citation":"https://pith.science/pith/VR47DM43J2AT5CMVTUNSZTQ3PG/action/citation_signature","submit_replication":"https://pith.science/pith/VR47DM43J2AT5CMVTUNSZTQ3PG/action/replication_record"}},"created_at":"2026-07-05T02:20:54.392635+00:00","updated_at":"2026-07-05T02:20:54.392635+00:00"}