{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:4HPTLOEBBOLTYBB6BRJ7E6XO6G","short_pith_number":"pith:4HPTLOEB","schema_version":"1.0","canonical_sha256":"e1df35b8810b973c043e0c53f27aeef1943ceecb3969172ff682c9992ad6b9fc","source":{"kind":"arxiv","id":"2507.20279","version":1},"attestation_state":"computed","paper":{"title":"What Language(s) Does Aya-23 Think In? How Multilinguality Affects Internal Language Representations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Katharina Trinley, Tanja Baeumel, Tatiana Anikina, Toshiki Nakai","submitted_at":"2025-07-27T13:53:45Z","abstract_excerpt":"Large language models (LLMs) excel at multilingual tasks, yet their internal language processing remains poorly understood. We analyze how Aya-23-8B, a decoder-only LLM trained on balanced multilingual data, handles code-mixed, cloze, and translation tasks compared to predominantly monolingual models like Llama 3 and Chinese-LLaMA-2. Using logit lens and neuron specialization analyses, we find: (1) Aya-23 activates typologically related language representations during translation, unlike English-centric models that rely on a single pivot language; (2) code-mixed neuron activation patterns vary"},"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":"2507.20279","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-07-27T13:53:45Z","cross_cats_sorted":[],"title_canon_sha256":"35da56fe1870397425e4ca98107e7a96636a9e6a0c383faed8e1c9a0ca91ba5a","abstract_canon_sha256":"efd7cf1cc9ebe8ce136b0e00573f3c54f120043591be71284e9921e3f0d28633"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:44:16.256125Z","signature_b64":"mZhUN8/U87rlKFtoCgAwuMTS34bibQ7i08Ao9oh2qcVeCkJsCF6doZyQH2cZ8GqDMPLg2ozqnV0q/N/1jm+GDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"e1df35b8810b973c043e0c53f27aeef1943ceecb3969172ff682c9992ad6b9fc","last_reissued_at":"2026-07-05T11:44:16.255643Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:44:16.255643Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"What Language(s) Does Aya-23 Think In? How Multilinguality Affects Internal Language Representations","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Katharina Trinley, Tanja Baeumel, Tatiana Anikina, Toshiki Nakai","submitted_at":"2025-07-27T13:53:45Z","abstract_excerpt":"Large language models (LLMs) excel at multilingual tasks, yet their internal language processing remains poorly understood. We analyze how Aya-23-8B, a decoder-only LLM trained on balanced multilingual data, handles code-mixed, cloze, and translation tasks compared to predominantly monolingual models like Llama 3 and Chinese-LLaMA-2. Using logit lens and neuron specialization analyses, we find: (1) Aya-23 activates typologically related language representations during translation, unlike English-centric models that rely on a single pivot language; (2) code-mixed neuron activation patterns vary"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.20279","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/2507.20279/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":"2507.20279","created_at":"2026-07-05T11:44:16.255706+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.20279v1","created_at":"2026-07-05T11:44:16.255706+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.20279","created_at":"2026-07-05T11:44:16.255706+00:00"},{"alias_kind":"pith_short_12","alias_value":"4HPTLOEBBOLT","created_at":"2026-07-05T11:44:16.255706+00:00"},{"alias_kind":"pith_short_16","alias_value":"4HPTLOEBBOLTYBB6","created_at":"2026-07-05T11:44:16.255706+00:00"},{"alias_kind":"pith_short_8","alias_value":"4HPTLOEB","created_at":"2026-07-05T11:44:16.255706+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2607.00664","citing_title":"YOMI-Bench: A Benchmark for Evaluating Kanji Reading and Phonological Understanding of LLMs for Japanese","ref_index":5,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/4HPTLOEBBOLTYBB6BRJ7E6XO6G","json":"https://pith.science/pith/4HPTLOEBBOLTYBB6BRJ7E6XO6G.json","graph_json":"https://pith.science/api/pith-number/4HPTLOEBBOLTYBB6BRJ7E6XO6G/graph.json","events_json":"https://pith.science/api/pith-number/4HPTLOEBBOLTYBB6BRJ7E6XO6G/events.json","paper":"https://pith.science/paper/4HPTLOEB"},"agent_actions":{"view_html":"https://pith.science/pith/4HPTLOEBBOLTYBB6BRJ7E6XO6G","download_json":"https://pith.science/pith/4HPTLOEBBOLTYBB6BRJ7E6XO6G.json","view_paper":"https://pith.science/paper/4HPTLOEB","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.20279&json=true","fetch_graph":"https://pith.science/api/pith-number/4HPTLOEBBOLTYBB6BRJ7E6XO6G/graph.json","fetch_events":"https://pith.science/api/pith-number/4HPTLOEBBOLTYBB6BRJ7E6XO6G/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/4HPTLOEBBOLTYBB6BRJ7E6XO6G/action/timestamp_anchor","attest_storage":"https://pith.science/pith/4HPTLOEBBOLTYBB6BRJ7E6XO6G/action/storage_attestation","attest_author":"https://pith.science/pith/4HPTLOEBBOLTYBB6BRJ7E6XO6G/action/author_attestation","sign_citation":"https://pith.science/pith/4HPTLOEBBOLTYBB6BRJ7E6XO6G/action/citation_signature","submit_replication":"https://pith.science/pith/4HPTLOEBBOLTYBB6BRJ7E6XO6G/action/replication_record"}},"created_at":"2026-07-05T11:44:16.255706+00:00","updated_at":"2026-07-05T11:44:16.255706+00:00"}