{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:ZUBCXTJFLJZPN24ERRBDC3YWZZ","short_pith_number":"pith:ZUBCXTJF","schema_version":"1.0","canonical_sha256":"cd022bcd255a72f6eb848c42316f16ce5e002219f6aefbd6f73c2a2d79b10476","source":{"kind":"arxiv","id":"2402.12530","version":1},"attestation_state":"computed","paper":{"title":"Parallel Structures in Pre-training Data Yield In-Context Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chen Zhao, He He, Kathleen McKeown, Yanda Chen, Zhou Yu","submitted_at":"2024-02-19T20:40:48Z","abstract_excerpt":"Pre-trained language models (LMs) are capable of in-context learning (ICL): they can adapt to a task with only a few examples given in the prompt without any parameter update. However, it is unclear where this capability comes from as there is a stark distribution shift between pre-training text and ICL prompts. In this work, we study what patterns of the pre-training data contribute to ICL. We find that LMs' ICL ability depends on $\\textit{parallel structures}$ in the pre-training data -- pairs of phrases following similar templates in the same context window. Specifically, we detect parallel"},"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":"2402.12530","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2024-02-19T20:40:48Z","cross_cats_sorted":["cs.AI","cs.LG"],"title_canon_sha256":"a6424b9db8cecb69994698f964590bcecba31ec028d920b5c0ec624f3fb12f24","abstract_canon_sha256":"4980a5da7803f64b34fb1846338acb7626898b4701be5f05ea96bc0ac82627c3"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:47:12.654311Z","signature_b64":"WEg+/M54ULw0UPO5V8fKQbRehnmLk5QZll0z8dNJ/uE/bjiS+ca0av6Z1NDgI/k+UwC9kYP2Qkk1ioX6LcpgDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"cd022bcd255a72f6eb848c42316f16ce5e002219f6aefbd6f73c2a2d79b10476","last_reissued_at":"2026-07-05T07:47:12.653902Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:47:12.653902Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Parallel Structures in Pre-training Data Yield In-Context Learning","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG"],"primary_cat":"cs.CL","authors_text":"Chen Zhao, He He, Kathleen McKeown, Yanda Chen, Zhou Yu","submitted_at":"2024-02-19T20:40:48Z","abstract_excerpt":"Pre-trained language models (LMs) are capable of in-context learning (ICL): they can adapt to a task with only a few examples given in the prompt without any parameter update. However, it is unclear where this capability comes from as there is a stark distribution shift between pre-training text and ICL prompts. In this work, we study what patterns of the pre-training data contribute to ICL. We find that LMs' ICL ability depends on $\\textit{parallel structures}$ in the pre-training data -- pairs of phrases following similar templates in the same context window. Specifically, we detect parallel"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.12530","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/2402.12530/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":"2402.12530","created_at":"2026-07-05T07:47:12.653961+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.12530v1","created_at":"2026-07-05T07:47:12.653961+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.12530","created_at":"2026-07-05T07:47:12.653961+00:00"},{"alias_kind":"pith_short_12","alias_value":"ZUBCXTJFLJZP","created_at":"2026-07-05T07:47:12.653961+00:00"},{"alias_kind":"pith_short_16","alias_value":"ZUBCXTJFLJZPN24E","created_at":"2026-07-05T07:47:12.653961+00:00"},{"alias_kind":"pith_short_8","alias_value":"ZUBCXTJF","created_at":"2026-07-05T07:47:12.653961+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2504.17068","citing_title":"In-Context Learning can distort the relationship between sequence likelihoods and biological fitness","ref_index":41,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/ZUBCXTJFLJZPN24ERRBDC3YWZZ","json":"https://pith.science/pith/ZUBCXTJFLJZPN24ERRBDC3YWZZ.json","graph_json":"https://pith.science/api/pith-number/ZUBCXTJFLJZPN24ERRBDC3YWZZ/graph.json","events_json":"https://pith.science/api/pith-number/ZUBCXTJFLJZPN24ERRBDC3YWZZ/events.json","paper":"https://pith.science/paper/ZUBCXTJF"},"agent_actions":{"view_html":"https://pith.science/pith/ZUBCXTJFLJZPN24ERRBDC3YWZZ","download_json":"https://pith.science/pith/ZUBCXTJFLJZPN24ERRBDC3YWZZ.json","view_paper":"https://pith.science/paper/ZUBCXTJF","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.12530&json=true","fetch_graph":"https://pith.science/api/pith-number/ZUBCXTJFLJZPN24ERRBDC3YWZZ/graph.json","fetch_events":"https://pith.science/api/pith-number/ZUBCXTJFLJZPN24ERRBDC3YWZZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/ZUBCXTJFLJZPN24ERRBDC3YWZZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/ZUBCXTJFLJZPN24ERRBDC3YWZZ/action/storage_attestation","attest_author":"https://pith.science/pith/ZUBCXTJFLJZPN24ERRBDC3YWZZ/action/author_attestation","sign_citation":"https://pith.science/pith/ZUBCXTJFLJZPN24ERRBDC3YWZZ/action/citation_signature","submit_replication":"https://pith.science/pith/ZUBCXTJFLJZPN24ERRBDC3YWZZ/action/replication_record"}},"created_at":"2026-07-05T07:47:12.653961+00:00","updated_at":"2026-07-05T07:47:12.653961+00:00"}