{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:7C5LXJ5F7C76VGIYX2IF4XCEEX","short_pith_number":"pith:7C5LXJ5F","schema_version":"1.0","canonical_sha256":"f8babba7a5f8bfea9918be905e5c4425f17de7745874a2cfc9805159d957ab86","source":{"kind":"arxiv","id":"2505.19116","version":2},"attestation_state":"computed","paper":{"title":"Controlling Language Confusion in Multilingual LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Guijin Son, Hyunwoo Ko, Nahyun Lee, Yeongseo Woo","submitted_at":"2025-05-25T12:15:31Z","abstract_excerpt":"Large language models often suffer from language confusion, a phenomenon in which responses are partially or entirely generated in unintended languages. This critically degrades the user experience, especially in low-resource settings. We hypothesize that this issue stems from limitations in conventional fine-tuning objectives, such as supervised learning, which optimize the likelihood of correct tokens without explicitly penalizing undesired outputs such as cross-lingual mixing. Analysis of loss trajectories during pretraining further reveals that models fail to distinguish between monolingua"},"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":"2505.19116","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-05-25T12:15:31Z","cross_cats_sorted":[],"title_canon_sha256":"eefccfcd5653c93a9ebe7c388c9c12aabbf6225e640458c3f13fc9449a89523f","abstract_canon_sha256":"a167a7fda7da4716ee5a450442ba5424472906e39bb4e2ffcf2b14ef9176e52c"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:39:54.279125Z","signature_b64":"stCak3Ku8zVkQdXCmBVsm31/yry/5t0Qqb5o0bk8pV9cT+AHZwbNvltjoyk7QLAnyG2TTHmIeaqiIASMULWvDA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"f8babba7a5f8bfea9918be905e5c4425f17de7745874a2cfc9805159d957ab86","last_reissued_at":"2026-07-05T11:39:54.278603Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:39:54.278603Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Controlling Language Confusion in Multilingual LLMs","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Guijin Son, Hyunwoo Ko, Nahyun Lee, Yeongseo Woo","submitted_at":"2025-05-25T12:15:31Z","abstract_excerpt":"Large language models often suffer from language confusion, a phenomenon in which responses are partially or entirely generated in unintended languages. This critically degrades the user experience, especially in low-resource settings. We hypothesize that this issue stems from limitations in conventional fine-tuning objectives, such as supervised learning, which optimize the likelihood of correct tokens without explicitly penalizing undesired outputs such as cross-lingual mixing. Analysis of loss trajectories during pretraining further reveals that models fail to distinguish between monolingua"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.19116","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/2505.19116/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":"2505.19116","created_at":"2026-07-05T11:39:54.278662+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.19116v2","created_at":"2026-07-05T11:39:54.278662+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.19116","created_at":"2026-07-05T11:39:54.278662+00:00"},{"alias_kind":"pith_short_12","alias_value":"7C5LXJ5F7C76","created_at":"2026-07-05T11:39:54.278662+00:00"},{"alias_kind":"pith_short_16","alias_value":"7C5LXJ5F7C76VGIY","created_at":"2026-07-05T11:39:54.278662+00:00"},{"alias_kind":"pith_short_8","alias_value":"7C5LXJ5F","created_at":"2026-07-05T11:39:54.278662+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7C5LXJ5F7C76VGIYX2IF4XCEEX","json":"https://pith.science/pith/7C5LXJ5F7C76VGIYX2IF4XCEEX.json","graph_json":"https://pith.science/api/pith-number/7C5LXJ5F7C76VGIYX2IF4XCEEX/graph.json","events_json":"https://pith.science/api/pith-number/7C5LXJ5F7C76VGIYX2IF4XCEEX/events.json","paper":"https://pith.science/paper/7C5LXJ5F"},"agent_actions":{"view_html":"https://pith.science/pith/7C5LXJ5F7C76VGIYX2IF4XCEEX","download_json":"https://pith.science/pith/7C5LXJ5F7C76VGIYX2IF4XCEEX.json","view_paper":"https://pith.science/paper/7C5LXJ5F","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.19116&json=true","fetch_graph":"https://pith.science/api/pith-number/7C5LXJ5F7C76VGIYX2IF4XCEEX/graph.json","fetch_events":"https://pith.science/api/pith-number/7C5LXJ5F7C76VGIYX2IF4XCEEX/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7C5LXJ5F7C76VGIYX2IF4XCEEX/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7C5LXJ5F7C76VGIYX2IF4XCEEX/action/storage_attestation","attest_author":"https://pith.science/pith/7C5LXJ5F7C76VGIYX2IF4XCEEX/action/author_attestation","sign_citation":"https://pith.science/pith/7C5LXJ5F7C76VGIYX2IF4XCEEX/action/citation_signature","submit_replication":"https://pith.science/pith/7C5LXJ5F7C76VGIYX2IF4XCEEX/action/replication_record"}},"created_at":"2026-07-05T11:39:54.278662+00:00","updated_at":"2026-07-05T11:39:54.278662+00:00"}