{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:T5RTKUS3PH42JRC65ZVTI2JTIO","short_pith_number":"pith:T5RTKUS3","schema_version":"1.0","canonical_sha256":"9f6335525b79f9a4c45eee6b346933438b32f15c2663897784eba4ea010ff709","source":{"kind":"arxiv","id":"2111.09509","version":1},"attestation_state":"computed","paper":{"title":"How much do language models copy from their training data? Evaluating linguistic novelty in text generation using RAVEN","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Asli Celikyilmaz, Jianfeng Gao, Paul Smolensky, R. Thomas McCoy, Tal Linzen","submitted_at":"2021-11-18T04:07:09Z","abstract_excerpt":"Current language models can generate high-quality text. Are they simply copying text they have seen before, or have they learned generalizable linguistic abstractions? To tease apart these possibilities, we introduce RAVEN, a suite of analyses for assessing the novelty of generated text, focusing on sequential structure (n-grams) and syntactic structure. We apply these analyses to four neural language models (an LSTM, a Transformer, Transformer-XL, and GPT-2). For local structure - e.g., individual dependencies - model-generated text is substantially less novel than our baseline of human-gener"},"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":"2111.09509","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2021-11-18T04:07:09Z","cross_cats_sorted":[],"title_canon_sha256":"a1c629a351f0d7dbe738f6dde2eeaa5a6850a7d7283f7af81373cc9e2511dd81","abstract_canon_sha256":"bd0eafad71f48f5fef6a1c1b4a18c4fea7caf412bab2de10c94698fc27cab089"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T03:33:12.409371Z","signature_b64":"43Zk0bYuagmu9mhFNu3TA+5uF6HY81/x3L0MYNPAXavmwUhUUUlGZWCcHalDldd37bcSR9xZ2UDI519GSzOOAg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"9f6335525b79f9a4c45eee6b346933438b32f15c2663897784eba4ea010ff709","last_reissued_at":"2026-07-05T03:33:12.408960Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T03:33:12.408960Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"How much do language models copy from their training data? Evaluating linguistic novelty in text generation using RAVEN","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Asli Celikyilmaz, Jianfeng Gao, Paul Smolensky, R. Thomas McCoy, Tal Linzen","submitted_at":"2021-11-18T04:07:09Z","abstract_excerpt":"Current language models can generate high-quality text. Are they simply copying text they have seen before, or have they learned generalizable linguistic abstractions? To tease apart these possibilities, we introduce RAVEN, a suite of analyses for assessing the novelty of generated text, focusing on sequential structure (n-grams) and syntactic structure. We apply these analyses to four neural language models (an LSTM, a Transformer, Transformer-XL, and GPT-2). For local structure - e.g., individual dependencies - model-generated text is substantially less novel than our baseline of human-gener"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2111.09509","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/2111.09509/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":"2111.09509","created_at":"2026-07-05T03:33:12.409018+00:00"},{"alias_kind":"arxiv_version","alias_value":"2111.09509v1","created_at":"2026-07-05T03:33:12.409018+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2111.09509","created_at":"2026-07-05T03:33:12.409018+00:00"},{"alias_kind":"pith_short_12","alias_value":"T5RTKUS3PH42","created_at":"2026-07-05T03:33:12.409018+00:00"},{"alias_kind":"pith_short_16","alias_value":"T5RTKUS3PH42JRC6","created_at":"2026-07-05T03:33:12.409018+00:00"},{"alias_kind":"pith_short_8","alias_value":"T5RTKUS3","created_at":"2026-07-05T03:33:12.409018+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":5,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.30815","citing_title":"When transformers learn \"impossible\" languages, what do they learn?","ref_index":47,"is_internal_anchor":false},{"citing_arxiv_id":"2507.06056","citing_title":"Data Compressibility Quantifies LLM Memorization","ref_index":46,"is_internal_anchor":false},{"citing_arxiv_id":"2308.05374","citing_title":"Trustworthy LLMs: a Survey and Guideline for Evaluating Large Language Models' Alignment","ref_index":192,"is_internal_anchor":false},{"citing_arxiv_id":"2202.07646","citing_title":"Quantifying Memorization Across Neural Language Models","ref_index":15,"is_internal_anchor":false},{"citing_arxiv_id":"2309.07864","citing_title":"The Rise and Potential of Large Language Model Based Agents: A Survey","ref_index":217,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/T5RTKUS3PH42JRC65ZVTI2JTIO","json":"https://pith.science/pith/T5RTKUS3PH42JRC65ZVTI2JTIO.json","graph_json":"https://pith.science/api/pith-number/T5RTKUS3PH42JRC65ZVTI2JTIO/graph.json","events_json":"https://pith.science/api/pith-number/T5RTKUS3PH42JRC65ZVTI2JTIO/events.json","paper":"https://pith.science/paper/T5RTKUS3"},"agent_actions":{"view_html":"https://pith.science/pith/T5RTKUS3PH42JRC65ZVTI2JTIO","download_json":"https://pith.science/pith/T5RTKUS3PH42JRC65ZVTI2JTIO.json","view_paper":"https://pith.science/paper/T5RTKUS3","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2111.09509&json=true","fetch_graph":"https://pith.science/api/pith-number/T5RTKUS3PH42JRC65ZVTI2JTIO/graph.json","fetch_events":"https://pith.science/api/pith-number/T5RTKUS3PH42JRC65ZVTI2JTIO/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/T5RTKUS3PH42JRC65ZVTI2JTIO/action/timestamp_anchor","attest_storage":"https://pith.science/pith/T5RTKUS3PH42JRC65ZVTI2JTIO/action/storage_attestation","attest_author":"https://pith.science/pith/T5RTKUS3PH42JRC65ZVTI2JTIO/action/author_attestation","sign_citation":"https://pith.science/pith/T5RTKUS3PH42JRC65ZVTI2JTIO/action/citation_signature","submit_replication":"https://pith.science/pith/T5RTKUS3PH42JRC65ZVTI2JTIO/action/replication_record"}},"created_at":"2026-07-05T03:33:12.409018+00:00","updated_at":"2026-07-05T03:33:12.409018+00:00"}