{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:2KFIOIM25NIAZ4BDBJI2AZNZA4","short_pith_number":"pith:2KFIOIM2","schema_version":"1.0","canonical_sha256":"d28a87219aeb500cf0230a51a065b907087a714be78b8ca60287d572ff383b94","source":{"kind":"arxiv","id":"2408.10811","version":1},"attestation_state":"computed","paper":{"title":"Beyond English-Centric LLMs: What Language Do Multilingual Language Models Think in?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Chengzhi Zhong, Chenhui Chu, Fei Cheng, Junfeng Jiang, Qianying Liu, Sadao Kurohashi, Yugo Murawaki, Zhen Wan","submitted_at":"2024-08-20T13:05:41Z","abstract_excerpt":"In this study, we investigate whether non-English-centric LLMs, despite their strong performance, `think' in their respective dominant language: more precisely, `think' refers to how the representations of intermediate layers, when un-embedded into the vocabulary space, exhibit higher probabilities for certain dominant languages during generation. We term such languages as internal $\\textbf{latent languages}$.\n  We examine the latent language of three typical categories of models for Japanese processing: Llama2, an English-centric model; Swallow, an English-centric model with continued pre-tra"},"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":"2408.10811","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-08-20T13:05:41Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"a67c12798d463392d838d9f035b9536a27e619330175b606940093fa599816ee","abstract_canon_sha256":"85eb8c81673cca138b5778dfc5b9346f43946ae7c6e2d3aa89072c91bb975982"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:57:17.807695Z","signature_b64":"qf0oX3CX9hdzW419UDBZmbtBN6JMmIDkslENUvtdjxDQOkhKxh8cJo3Ypz2zKR4nHqFs5aS/31d4RKsmHe2eBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"d28a87219aeb500cf0230a51a065b907087a714be78b8ca60287d572ff383b94","last_reissued_at":"2026-07-05T08:57:17.807193Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:57:17.807193Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Beyond English-Centric LLMs: What Language Do Multilingual Language Models Think in?","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.CL","authors_text":"Chengzhi Zhong, Chenhui Chu, Fei Cheng, Junfeng Jiang, Qianying Liu, Sadao Kurohashi, Yugo Murawaki, Zhen Wan","submitted_at":"2024-08-20T13:05:41Z","abstract_excerpt":"In this study, we investigate whether non-English-centric LLMs, despite their strong performance, `think' in their respective dominant language: more precisely, `think' refers to how the representations of intermediate layers, when un-embedded into the vocabulary space, exhibit higher probabilities for certain dominant languages during generation. We term such languages as internal $\\textbf{latent languages}$.\n  We examine the latent language of three typical categories of models for Japanese processing: Llama2, an English-centric model; Swallow, an English-centric model with continued pre-tra"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2408.10811","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/2408.10811/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":"2408.10811","created_at":"2026-07-05T08:57:17.807248+00:00"},{"alias_kind":"arxiv_version","alias_value":"2408.10811v1","created_at":"2026-07-05T08:57:17.807248+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2408.10811","created_at":"2026-07-05T08:57:17.807248+00:00"},{"alias_kind":"pith_short_12","alias_value":"2KFIOIM25NIA","created_at":"2026-07-05T08:57:17.807248+00:00"},{"alias_kind":"pith_short_16","alias_value":"2KFIOIM25NIAZ4BD","created_at":"2026-07-05T08:57:17.807248+00:00"},{"alias_kind":"pith_short_8","alias_value":"2KFIOIM2","created_at":"2026-07-05T08:57:17.807248+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.17669","citing_title":"DeSRPA: Decoupled Speech Role-Playing Agent via Inference-Time Intervention","ref_index":22,"is_internal_anchor":false},{"citing_arxiv_id":"2505.14990","citing_title":"Language Specific Knowledge: Do Models Know Better in X than in English?","ref_index":23,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/2KFIOIM25NIAZ4BDBJI2AZNZA4","json":"https://pith.science/pith/2KFIOIM25NIAZ4BDBJI2AZNZA4.json","graph_json":"https://pith.science/api/pith-number/2KFIOIM25NIAZ4BDBJI2AZNZA4/graph.json","events_json":"https://pith.science/api/pith-number/2KFIOIM25NIAZ4BDBJI2AZNZA4/events.json","paper":"https://pith.science/paper/2KFIOIM2"},"agent_actions":{"view_html":"https://pith.science/pith/2KFIOIM25NIAZ4BDBJI2AZNZA4","download_json":"https://pith.science/pith/2KFIOIM25NIAZ4BDBJI2AZNZA4.json","view_paper":"https://pith.science/paper/2KFIOIM2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2408.10811&json=true","fetch_graph":"https://pith.science/api/pith-number/2KFIOIM25NIAZ4BDBJI2AZNZA4/graph.json","fetch_events":"https://pith.science/api/pith-number/2KFIOIM25NIAZ4BDBJI2AZNZA4/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/2KFIOIM25NIAZ4BDBJI2AZNZA4/action/timestamp_anchor","attest_storage":"https://pith.science/pith/2KFIOIM25NIAZ4BDBJI2AZNZA4/action/storage_attestation","attest_author":"https://pith.science/pith/2KFIOIM25NIAZ4BDBJI2AZNZA4/action/author_attestation","sign_citation":"https://pith.science/pith/2KFIOIM25NIAZ4BDBJI2AZNZA4/action/citation_signature","submit_replication":"https://pith.science/pith/2KFIOIM25NIAZ4BDBJI2AZNZA4/action/replication_record"}},"created_at":"2026-07-05T08:57:17.807248+00:00","updated_at":"2026-07-05T08:57:17.807248+00:00"}