{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:5R6MM4LRYYSNCWVITYTQ5MY36Q","short_pith_number":"pith:5R6MM4LR","schema_version":"1.0","canonical_sha256":"ec7cc67171c624d15aa89e270eb31bf407b4149280f70a76250050262357b8e3","source":{"kind":"arxiv","id":"2505.12584","version":2},"attestation_state":"computed","paper":{"title":"Improving Multilingual Language Models by Aligning Representations through Steering","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Buddhika Laknath Semage, Omar Mahmoud, Santu Rana, Thommen George Karimpanal","submitted_at":"2025-05-19T00:14:43Z","abstract_excerpt":"This paper investigates how Large Language Models (LLMs) represent non-English tokens -- a question that remains underexplored despite recent progress. We propose a lightweight intervention method using representation steering, where a learned vector is added to the residual stream at a single model layer to enhance multilingual performance. Through extensive experiments across seven competitive baselines -- including prompt optimization, supervised fine-tuning (SFT), in-context learning, cross-lingual transfer, and translation-based methods-we show that our approach consistently outperforms m"},"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":"2505.12584","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-05-19T00:14:43Z","cross_cats_sorted":[],"title_canon_sha256":"0361e9c73d3974557b3b3897d85b6cbbe5f7f0ca7ee970dc1bd7e796ae1b957b","abstract_canon_sha256":"08b99b2e7cc1ca0f0667dc956af5fb5ef72559a78c6fb1f037d00edda62cad74"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:59:14.666970Z","signature_b64":"TJnPIPcXw/c479eVeTEyHViUM/3BXmd3Hhx+z+2BAKv16OFfTOEgacipM+1OHVcJj2mRl9YX5aIa9NoEw+7sDQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ec7cc67171c624d15aa89e270eb31bf407b4149280f70a76250050262357b8e3","last_reissued_at":"2026-07-05T11:59:14.666499Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:59:14.666499Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Improving Multilingual Language Models by Aligning Representations through Steering","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Buddhika Laknath Semage, Omar Mahmoud, Santu Rana, Thommen George Karimpanal","submitted_at":"2025-05-19T00:14:43Z","abstract_excerpt":"This paper investigates how Large Language Models (LLMs) represent non-English tokens -- a question that remains underexplored despite recent progress. We propose a lightweight intervention method using representation steering, where a learned vector is added to the residual stream at a single model layer to enhance multilingual performance. Through extensive experiments across seven competitive baselines -- including prompt optimization, supervised fine-tuning (SFT), in-context learning, cross-lingual transfer, and translation-based methods-we show that our approach consistently outperforms m"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.12584","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/2505.12584/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":"2505.12584","created_at":"2026-07-05T11:59:14.666558+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.12584v2","created_at":"2026-07-05T11:59:14.666558+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.12584","created_at":"2026-07-05T11:59:14.666558+00:00"},{"alias_kind":"pith_short_12","alias_value":"5R6MM4LRYYSN","created_at":"2026-07-05T11:59:14.666558+00:00"},{"alias_kind":"pith_short_16","alias_value":"5R6MM4LRYYSNCWVI","created_at":"2026-07-05T11:59:14.666558+00:00"},{"alias_kind":"pith_short_8","alias_value":"5R6MM4LR","created_at":"2026-07-05T11:59:14.666558+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.22473","citing_title":"Interleaved Speech Language Models Latently Work In Text","ref_index":17,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/5R6MM4LRYYSNCWVITYTQ5MY36Q","json":"https://pith.science/pith/5R6MM4LRYYSNCWVITYTQ5MY36Q.json","graph_json":"https://pith.science/api/pith-number/5R6MM4LRYYSNCWVITYTQ5MY36Q/graph.json","events_json":"https://pith.science/api/pith-number/5R6MM4LRYYSNCWVITYTQ5MY36Q/events.json","paper":"https://pith.science/paper/5R6MM4LR"},"agent_actions":{"view_html":"https://pith.science/pith/5R6MM4LRYYSNCWVITYTQ5MY36Q","download_json":"https://pith.science/pith/5R6MM4LRYYSNCWVITYTQ5MY36Q.json","view_paper":"https://pith.science/paper/5R6MM4LR","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.12584&json=true","fetch_graph":"https://pith.science/api/pith-number/5R6MM4LRYYSNCWVITYTQ5MY36Q/graph.json","fetch_events":"https://pith.science/api/pith-number/5R6MM4LRYYSNCWVITYTQ5MY36Q/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/5R6MM4LRYYSNCWVITYTQ5MY36Q/action/timestamp_anchor","attest_storage":"https://pith.science/pith/5R6MM4LRYYSNCWVITYTQ5MY36Q/action/storage_attestation","attest_author":"https://pith.science/pith/5R6MM4LRYYSNCWVITYTQ5MY36Q/action/author_attestation","sign_citation":"https://pith.science/pith/5R6MM4LRYYSNCWVITYTQ5MY36Q/action/citation_signature","submit_replication":"https://pith.science/pith/5R6MM4LRYYSNCWVITYTQ5MY36Q/action/replication_record"}},"created_at":"2026-07-05T11:59:14.666558+00:00","updated_at":"2026-07-05T11:59:14.666558+00:00"}