{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:WWBTLEBQBHQQAJMAD2FR2VWHUA","short_pith_number":"pith:WWBTLEBQ","schema_version":"1.0","canonical_sha256":"b58335903009e10025801e8b1d56c7a03a0da5566a2cebd60b1e8f6a22014b8b","source":{"kind":"arxiv","id":"2508.11017","version":2},"attestation_state":"computed","paper":{"title":"Beyond the Rosetta Stone: Unification Forces in Generalization Dynamics","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Ann Yuan, Asma Ghandeharioun, Carter Blum, Fred Zhang, Jessica Hoffmann, Julian Zimmert, Katja Filippova, Lucas Dixon, Martin Wattenberg, Mor Geva, Tal Linzen","submitted_at":"2025-08-14T18:44:13Z","abstract_excerpt":"Large language models (LLMs) struggle with cross-lingual knowledge transfer: they hallucinate when asked in one language about facts expressed in a different language during training. This work introduces a controlled setting to study the causes and dynamics of this phenomenon by training small Transformer models from scratch on synthetic multilingual datasets. We identify a learning phase wherein a model develops either separate or unified representations of the same facts across languages, and show that unification is essential for cross-lingual transfer. We also show that the degree of unif"},"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":"2508.11017","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-08-14T18:44:13Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"7cb6f891f78f9a8763cc996d5856e3efd0da4a05cb9e1706d81eafc2f25d012a","abstract_canon_sha256":"4d590bc8148bd87735f468e8b09304da41e1bc5443afae1d1dbee05ad1396347"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T12:00:51.632796Z","signature_b64":"noJGBK8HezpjHItnxKeKPgF1ua7CH7KuTfgy2FcQqPNVDD4V72cQmT8DFn7N8vNGSA9Wp7I6mwTEeJy49O7BAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"b58335903009e10025801e8b1d56c7a03a0da5566a2cebd60b1e8f6a22014b8b","last_reissued_at":"2026-07-05T12:00:51.632222Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T12:00:51.632222Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Beyond the Rosetta Stone: Unification Forces in Generalization Dynamics","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Ann Yuan, Asma Ghandeharioun, Carter Blum, Fred Zhang, Jessica Hoffmann, Julian Zimmert, Katja Filippova, Lucas Dixon, Martin Wattenberg, Mor Geva, Tal Linzen","submitted_at":"2025-08-14T18:44:13Z","abstract_excerpt":"Large language models (LLMs) struggle with cross-lingual knowledge transfer: they hallucinate when asked in one language about facts expressed in a different language during training. This work introduces a controlled setting to study the causes and dynamics of this phenomenon by training small Transformer models from scratch on synthetic multilingual datasets. We identify a learning phase wherein a model develops either separate or unified representations of the same facts across languages, and show that unification is essential for cross-lingual transfer. We also show that the degree of unif"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2508.11017","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/2508.11017/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":"2508.11017","created_at":"2026-07-05T12:00:51.632280+00:00"},{"alias_kind":"arxiv_version","alias_value":"2508.11017v2","created_at":"2026-07-05T12:00:51.632280+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2508.11017","created_at":"2026-07-05T12:00:51.632280+00:00"},{"alias_kind":"pith_short_12","alias_value":"WWBTLEBQBHQQ","created_at":"2026-07-05T12:00:51.632280+00:00"},{"alias_kind":"pith_short_16","alias_value":"WWBTLEBQBHQQAJMA","created_at":"2026-07-05T12:00:51.632280+00:00"},{"alias_kind":"pith_short_8","alias_value":"WWBTLEBQ","created_at":"2026-07-05T12:00:51.632280+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.27237","citing_title":"LMs as Task-Specific Knowledge Bases: An Interpretability Analysis","ref_index":111,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/WWBTLEBQBHQQAJMAD2FR2VWHUA","json":"https://pith.science/pith/WWBTLEBQBHQQAJMAD2FR2VWHUA.json","graph_json":"https://pith.science/api/pith-number/WWBTLEBQBHQQAJMAD2FR2VWHUA/graph.json","events_json":"https://pith.science/api/pith-number/WWBTLEBQBHQQAJMAD2FR2VWHUA/events.json","paper":"https://pith.science/paper/WWBTLEBQ"},"agent_actions":{"view_html":"https://pith.science/pith/WWBTLEBQBHQQAJMAD2FR2VWHUA","download_json":"https://pith.science/pith/WWBTLEBQBHQQAJMAD2FR2VWHUA.json","view_paper":"https://pith.science/paper/WWBTLEBQ","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2508.11017&json=true","fetch_graph":"https://pith.science/api/pith-number/WWBTLEBQBHQQAJMAD2FR2VWHUA/graph.json","fetch_events":"https://pith.science/api/pith-number/WWBTLEBQBHQQAJMAD2FR2VWHUA/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/WWBTLEBQBHQQAJMAD2FR2VWHUA/action/timestamp_anchor","attest_storage":"https://pith.science/pith/WWBTLEBQBHQQAJMAD2FR2VWHUA/action/storage_attestation","attest_author":"https://pith.science/pith/WWBTLEBQBHQQAJMAD2FR2VWHUA/action/author_attestation","sign_citation":"https://pith.science/pith/WWBTLEBQBHQQAJMAD2FR2VWHUA/action/citation_signature","submit_replication":"https://pith.science/pith/WWBTLEBQBHQQAJMAD2FR2VWHUA/action/replication_record"}},"created_at":"2026-07-05T12:00:51.632280+00:00","updated_at":"2026-07-05T12:00:51.632280+00:00"}