{"bundle_type":"pith_open_graph_bundle","bundle_version":"1.0","pith_number":"pith:2026:7I7SCPGGNWLU4OEGU5732GBUMZ","short_pith_number":"pith:7I7SCPGG","canonical_record":{"source":{"id":"2605.13919","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T13:22:45Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"206f9cc58a45416cee6900abb3589d33b027b74bffc751e23c64c7e466550e3d","abstract_canon_sha256":"f45c618daef44bc4252090342d88f53b7eadcd18398dfef067cfb17f2a5992d8"},"schema_version":"1.0"},"canonical_sha256":"fa3f213cc66d974e3886a77fbd183466672bfb426bb83eae45ee15bb303ca67b","source":{"kind":"arxiv","id":"2605.13919","version":1},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13919","created_at":"2026-05-17T23:39:18Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13919v1","created_at":"2026-05-17T23:39:18Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13919","created_at":"2026-05-17T23:39:18Z"},{"alias_kind":"pith_short_12","alias_value":"7I7SCPGGNWLU","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"7I7SCPGGNWLU4OEG","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"7I7SCPGG","created_at":"2026-05-18T12:33:37Z"}],"events":[{"event_type":"record_created","subject_pith_number":"pith:2026:7I7SCPGGNWLU4OEGU5732GBUMZ","target":"record","payload":{"canonical_record":{"source":{"id":"2605.13919","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T13:22:45Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"206f9cc58a45416cee6900abb3589d33b027b74bffc751e23c64c7e466550e3d","abstract_canon_sha256":"f45c618daef44bc4252090342d88f53b7eadcd18398dfef067cfb17f2a5992d8"},"schema_version":"1.0"},"canonical_sha256":"fa3f213cc66d974e3886a77fbd183466672bfb426bb83eae45ee15bb303ca67b","receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-05-17T23:39:18.715753Z","signature_b64":"5pcysy0Qp4oDvER2GX1srsNAyN7LvFNF7ycWEPIUz8/ExjCoIlLOsDkfSWlSACvzoOQQ/ZDnQNJy53CRrCb6Dg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fa3f213cc66d974e3886a77fbd183466672bfb426bb83eae45ee15bb303ca67b","last_reissued_at":"2026-05-17T23:39:18.715216Z","signature_status":"signed_v1","first_computed_at":"2026-05-17T23:39:18.715216Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"source_kind":"arxiv","source_id":"2605.13919","source_version":1,"attestation_state":"computed"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:18Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"RiE+jSRm8nTr5zI3H0dSFcZKL7fO6lnjCrKMp2sz7IbaxEwIAQ/1NXFbUTLgvQ4pibMPIfaSOleUvau5QZPwCw==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-10T18:24:47.922648Z"},"content_sha256":"c6ddcd9d346f642a2ea21832b2f059502fc49da474eca7d6d50c72761fb3adb0","schema_version":"1.0","event_id":"sha256:c6ddcd9d346f642a2ea21832b2f059502fc49da474eca7d6d50c72761fb3adb0"},{"event_type":"graph_snapshot","subject_pith_number":"pith:2026:7I7SCPGGNWLU4OEGU5732GBUMZ","target":"graph","payload":{"graph_snapshot":{"paper":{"title":"Merging Methods for Multilingual Knowledge Editing for Large Language Models: An Empirical Odyssey","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"Vector summation with shared covariance emerges as the most reliable strategy for merging knowledge edits across languages in large language models.","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Jong-Hyeok Lee, Ki-Young Shin, Kunil Lee, Young-Joo Suh","submitted_at":"2026-05-13T13:22:45Z","abstract_excerpt":"Multilingual knowledge editing (MKE) remains challenging because language-specific edits interfere with one another, even when locate-then-edit methods work well in monolingual settings. This paper focuses on three issues: the effectiveness of vector merging methods for MKE, the extent to which Task Singular Vectors for Merging (TSVM) can reduce multilingual interference, and the influence of the weight scaling factor and rank compression ratio on performance. We evaluate six merging variants with two popular backbone large language models, two base knowledge editing methods, and 12 languages "},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Our results show that vector summation with shared covariance is the most reliable overall strategy, whereas simple summation without shared covariance performs poorly. TSVM improves performance in some settings, but its ability to mitigate multilingual interference is limited.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"The MzsRE benchmark and the selected twelve languages plus two backbone models are assumed to be representative enough of real multilingual interference that the observed performance differences will generalize.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"Vector summation with shared covariance is the most reliable merging strategy for multilingual knowledge editing, though TSVM only partially reduces interference and results depend heavily on weight scaling and rank compression.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"Vector summation with shared covariance emerges as the most reliable strategy for merging knowledge edits across languages in large language models.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"0574ba75e5af466912a2b612cf56933c815954a7adfcfd964f44197c875b785d"},"source":{"id":"2605.13919","kind":"arxiv","version":1},"verdict":{"id":"a67f37a2-de37-4ace-be79-5fb2683adf81","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-15T05:45:47.895554Z","strongest_claim":"Our results show that vector summation with shared covariance is the most reliable overall strategy, whereas simple summation without shared covariance performs poorly. TSVM improves performance in some settings, but its ability to mitigate multilingual interference is limited.","one_line_summary":"Vector summation with shared covariance is the most reliable merging strategy for multilingual knowledge editing, though TSVM only partially reduces interference and results depend heavily on weight scaling and rank compression.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"The MzsRE benchmark and the selected twelve languages plus two backbone models are assumed to be representative enough of real multilingual interference that the observed performance differences will generalize.","pith_extraction_headline":"Vector summation with shared covariance emerges as the most reliable strategy for merging knowledge edits across languages in large language models."},"references":{"count":27,"sample":[{"doi":"","year":2023,"title":"GPT-4 Technical Report","work_id":"3f626ef1-7e86-4f73-8217-0217676cf5a8","ref_index":1,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"The Llama 3 Herd of Models","work_id":"4fe9d5dc-c3e0-4d7d-b3bc-88d3ea88929d","ref_index":2,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2024,"title":"Qwen2 Technical Report","work_id":"f12923d7-1d38-477b-b26c-01b31c028e80","ref_index":3,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2025,"title":"Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities","work_id":"ac4b7764-2b3b-4ca7-a601-67511f785cac","ref_index":4,"cited_arxiv_id":"","is_internal_anchor":false},{"doi":"","year":2017,"title":"Attention Is All You Need","work_id":"e4051452-83dc-45f6-a787-4ebfaac2835a","ref_index":5,"cited_arxiv_id":"","is_internal_anchor":false}],"resolved_work":27,"snapshot_sha256":"a90d4c11a557d409cc6ac851602356fafb91a11e69f2e34af280ad8eb96fbad4","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"},"verdict_id":"a67f37a2-de37-4ace-be79-5fb2683adf81"},"signer":{"signer_id":"pith.science","signer_type":"pith_registry","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"created_at":"2026-05-17T23:39:18Z","supersedes":[],"prev_event":null,"signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"syiIn9p45gJqhrTK0qr0NBjoq3pcRioha6BMJoH9s/u3ObtHBBwomJ3x5LZM8hIgDwgrL4ADA7TXwtifyiMtDg==","signed_message":"open_graph_event_sha256_bytes","signed_at":"2026-08-10T18:24:47.923932Z"},"content_sha256":"0bc1b5e339041da10ea09fcd64a8c2a9e7f1c929c88db8228c488b0d0e5adb78","schema_version":"1.0","event_id":"sha256:0bc1b5e339041da10ea09fcd64a8c2a9e7f1c929c88db8228c488b0d0e5adb78"}],"timestamp_proofs":[],"mirror_hints":[{"mirror_type":"https","name":"Pith Resolver","base_url":"https://pith.science","bundle_url":"https://pith.science/pith/7I7SCPGGNWLU4OEGU5732GBUMZ/bundle.json","state_url":"https://pith.science/pith/7I7SCPGGNWLU4OEGU5732GBUMZ/state.json","well_known_bundle_url":"https://pith.science/.well-known/pith/7I7SCPGGNWLU4OEGU5732GBUMZ/bundle.json","status":"primary"}],"public_keys":[{"key_id":"pith-v1-2026-05","algorithm":"ed25519","format":"raw","public_key_b64":"stVStoiQhXFxp4s2pdzPNoqVNBMojDU/fJ2db5S3CbM=","public_key_hex":"b2d552b68890857171a78b36a5dccf368a953413288c353f7c9d9d6f94b709b3","fingerprint_sha256_b32_first128bits":"RVFV5Z2OI2J3ZUO7ERDEBCYNKS","fingerprint_sha256_hex":"8d4b5ee74e4693bcd1df2446408b0d54","rotates_at":null,"url":"https://pith.science/pith-signing-key.json","notes":"Pith uses this Ed25519 key to sign canonical record SHA-256 digests. Verify with: ed25519_verify(public_key, message=canonical_sha256_bytes, signature=base64decode(signature_b64))."}],"merge_version":"pith-open-graph-merge-v1","built_at":"2026-08-10T18:24:47Z","links":{"resolver":"https://pith.science/pith/7I7SCPGGNWLU4OEGU5732GBUMZ","bundle":"https://pith.science/pith/7I7SCPGGNWLU4OEGU5732GBUMZ/bundle.json","state":"https://pith.science/pith/7I7SCPGGNWLU4OEGU5732GBUMZ/state.json","well_known_bundle":"https://pith.science/.well-known/pith/7I7SCPGGNWLU4OEGU5732GBUMZ/bundle.json"},"state":{"state_type":"pith_open_graph_state","state_version":"1.0","pith_number":"pith:2026:7I7SCPGGNWLU4OEGU5732GBUMZ","merge_version":"pith-open-graph-merge-v1","event_count":2,"valid_event_count":2,"invalid_event_count":0,"equivocation_count":0,"current":{"canonical_record":{"metadata":{"abstract_canon_sha256":"f45c618daef44bc4252090342d88f53b7eadcd18398dfef067cfb17f2a5992d8","cross_cats_sorted":["cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T13:22:45Z","title_canon_sha256":"206f9cc58a45416cee6900abb3589d33b027b74bffc751e23c64c7e466550e3d"},"schema_version":"1.0","source":{"id":"2605.13919","kind":"arxiv","version":1}},"source_aliases":[{"alias_kind":"arxiv","alias_value":"2605.13919","created_at":"2026-05-17T23:39:18Z"},{"alias_kind":"arxiv_version","alias_value":"2605.13919v1","created_at":"2026-05-17T23:39:18Z"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2605.13919","created_at":"2026-05-17T23:39:18Z"},{"alias_kind":"pith_short_12","alias_value":"7I7SCPGGNWLU","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_16","alias_value":"7I7SCPGGNWLU4OEG","created_at":"2026-05-18T12:33:37Z"},{"alias_kind":"pith_short_8","alias_value":"7I7SCPGG","created_at":"2026-05-18T12:33:37Z"}],"graph_snapshots":[{"event_id":"sha256:0bc1b5e339041da10ea09fcd64a8c2a9e7f1c929c88db8228c488b0d0e5adb78","target":"graph","created_at":"2026-05-17T23:39:18Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"graph_snapshot":{"author_claims":{"count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","strong_count":0},"builder_version":"pith-number-builder-2026-05-17-v1","claims":{"count":4,"items":[{"attestation":"unclaimed","claim_id":"C1","kind":"strongest_claim","source":"verdict.strongest_claim","status":"machine_extracted","text":"Our results show that vector summation with shared covariance is the most reliable overall strategy, whereas simple summation without shared covariance performs poorly. TSVM improves performance in some settings, but its ability to mitigate multilingual interference is limited."},{"attestation":"unclaimed","claim_id":"C2","kind":"weakest_assumption","source":"verdict.weakest_assumption","status":"machine_extracted","text":"The MzsRE benchmark and the selected twelve languages plus two backbone models are assumed to be representative enough of real multilingual interference that the observed performance differences will generalize."},{"attestation":"unclaimed","claim_id":"C3","kind":"one_line_summary","source":"verdict.one_line_summary","status":"machine_extracted","text":"Vector summation with shared covariance is the most reliable merging strategy for multilingual knowledge editing, though TSVM only partially reduces interference and results depend heavily on weight scaling and rank compression."},{"attestation":"unclaimed","claim_id":"C4","kind":"headline","source":"verdict.pith_extraction.headline","status":"machine_extracted","text":"Vector summation with shared covariance emerges as the most reliable strategy for merging knowledge edits across languages in large language models."}],"snapshot_sha256":"0574ba75e5af466912a2b612cf56933c815954a7adfcfd964f44197c875b785d"},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"paper":{"abstract_excerpt":"Multilingual knowledge editing (MKE) remains challenging because language-specific edits interfere with one another, even when locate-then-edit methods work well in monolingual settings. This paper focuses on three issues: the effectiveness of vector merging methods for MKE, the extent to which Task Singular Vectors for Merging (TSVM) can reduce multilingual interference, and the influence of the weight scaling factor and rank compression ratio on performance. We evaluate six merging variants with two popular backbone large language models, two base knowledge editing methods, and 12 languages ","authors_text":"Jong-Hyeok Lee, Ki-Young Shin, Kunil Lee, Young-Joo Suh","cross_cats":["cs.LG"],"headline":"Vector summation with shared covariance emerges as the most reliable strategy for merging knowledge edits across languages in large language models.","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T13:22:45Z","title":"Merging Methods for Multilingual Knowledge Editing for Large Language Models: An Empirical Odyssey"},"references":{"count":27,"internal_anchors":0,"resolved_work":27,"sample":[{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":1,"title":"GPT-4 Technical Report","work_id":"3f626ef1-7e86-4f73-8217-0217676cf5a8","year":2023},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":2,"title":"The Llama 3 Herd of Models","work_id":"4fe9d5dc-c3e0-4d7d-b3bc-88d3ea88929d","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":3,"title":"Qwen2 Technical Report","work_id":"f12923d7-1d38-477b-b26c-01b31c028e80","year":2024},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":4,"title":"Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities","work_id":"ac4b7764-2b3b-4ca7-a601-67511f785cac","year":2025},{"cited_arxiv_id":"","doi":"","is_internal_anchor":false,"ref_index":5,"title":"Attention Is All You Need","work_id":"e4051452-83dc-45f6-a787-4ebfaac2835a","year":2017}],"snapshot_sha256":"a90d4c11a557d409cc6ac851602356fafb91a11e69f2e34af280ad8eb96fbad4"},"source":{"id":"2605.13919","kind":"arxiv","version":1},"verdict":{"created_at":"2026-05-15T05:45:47.895554Z","id":"a67f37a2-de37-4ace-be79-5fb2683adf81","model_set":{"reader":"grok-4.3"},"one_line_summary":"Vector summation with shared covariance is the most reliable merging strategy for multilingual knowledge editing, though TSVM only partially reduces interference and results depend heavily on weight scaling and rank compression.","pipeline_version":"pith-pipeline@v0.9.0","pith_extraction_headline":"Vector summation with shared covariance emerges as the most reliable strategy for merging knowledge edits across languages in large language models.","strongest_claim":"Our results show that vector summation with shared covariance is the most reliable overall strategy, whereas simple summation without shared covariance performs poorly. TSVM improves performance in some settings, but its ability to mitigate multilingual interference is limited.","weakest_assumption":"The MzsRE benchmark and the selected twelve languages plus two backbone models are assumed to be representative enough of real multilingual interference that the observed performance differences will generalize."}},"verdict_id":"a67f37a2-de37-4ace-be79-5fb2683adf81"}}],"author_attestations":[],"timestamp_anchors":[],"storage_attestations":[],"citation_signatures":[],"replication_records":[],"corrections":[],"mirror_hints":[],"record_created":{"event_id":"sha256:c6ddcd9d346f642a2ea21832b2f059502fc49da474eca7d6d50c72761fb3adb0","target":"record","created_at":"2026-05-17T23:39:18Z","signer":{"key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signer_id":"pith.science","signer_type":"pith_registry"},"payload":{"attestation_state":"computed","canonical_record":{"metadata":{"abstract_canon_sha256":"f45c618daef44bc4252090342d88f53b7eadcd18398dfef067cfb17f2a5992d8","cross_cats_sorted":["cs.LG"],"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-05-13T13:22:45Z","title_canon_sha256":"206f9cc58a45416cee6900abb3589d33b027b74bffc751e23c64c7e466550e3d"},"schema_version":"1.0","source":{"id":"2605.13919","kind":"arxiv","version":1}},"canonical_sha256":"fa3f213cc66d974e3886a77fbd183466672bfb426bb83eae45ee15bb303ca67b","receipt":{"algorithm":"ed25519","builder_version":"pith-number-builder-2026-05-17-v1","canonical_sha256":"fa3f213cc66d974e3886a77fbd183466672bfb426bb83eae45ee15bb303ca67b","first_computed_at":"2026-05-17T23:39:18.715216Z","key_id":"pith-v1-2026-05","kind":"pith_receipt","last_reissued_at":"2026-05-17T23:39:18.715216Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","receipt_version":"0.3","signature_b64":"5pcysy0Qp4oDvER2GX1srsNAyN7LvFNF7ycWEPIUz8/ExjCoIlLOsDkfSWlSACvzoOQQ/ZDnQNJy53CRrCb6Dg==","signature_status":"signed_v1","signed_at":"2026-05-17T23:39:18.715753Z","signed_message":"canonical_sha256_bytes"},"source_id":"2605.13919","source_kind":"arxiv","source_version":1}}},"equivocations":[],"invalid_events":[],"applied_event_ids":["sha256:c6ddcd9d346f642a2ea21832b2f059502fc49da474eca7d6d50c72761fb3adb0","sha256:0bc1b5e339041da10ea09fcd64a8c2a9e7f1c929c88db8228c488b0d0e5adb78"],"state_sha256":"4faf3c3b741a2c707b6138eb740f3286aa5a23e3842edc2493d711027bced388"},"bundle_signature":{"signature_status":"signed_v1","algorithm":"ed25519","key_id":"pith-v1-2026-05","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54","signature_b64":"PNPtBjAEYkdAPTQFDZDlyTz2yu8eOG26F1nWP2y1LK+2BpQkg9ftjfCOO0ssOFvpwh0uxmTZcMDpRr8iGfg4Bw==","signed_message":"bundle_sha256_bytes","signed_at":"2026-08-10T18:24:47.932146Z","bundle_sha256":"0b12a064a28e456e00961d02a1a2424cef0ffa4f57e06a5592d80375763ce21c"}}