{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:BNIWH4BSQ5YQKRYB4IFYEIAUAV","short_pith_number":"pith:BNIWH4BS","schema_version":"1.0","canonical_sha256":"0b5163f0328771054701e20b822014057f247849b12140c58c298747a655fc50","source":{"kind":"arxiv","id":"2305.14481","version":2},"attestation_state":"computed","paper":{"title":"FOCUS: Effective Embedding Initialization for Monolingual Specialization of Multilingual Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Gerard de Melo, Konstantin Dobler","submitted_at":"2023-05-23T19:21:53Z","abstract_excerpt":"Using model weights pretrained on a high-resource language as a warm start can reduce the need for data and compute to obtain high-quality language models for other, especially low-resource, languages. However, if we want to use a new tokenizer specialized for the target language, we cannot transfer the source model's embedding matrix. In this paper, we propose FOCUS - Fast Overlapping Token Combinations Using Sparsemax, a novel embedding initialization method that initializes the embedding matrix effectively for a new tokenizer based on information in the source model's embedding matrix. FOCU"},"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":"2305.14481","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2023-05-23T19:21:53Z","cross_cats_sorted":[],"title_canon_sha256":"1f0379fb453c1aad6611b76de4aae9a3076e9b06f8c654ecf85add12d327b0f0","abstract_canon_sha256":"cb462006c445d44e67438aec56f39d0575587b9e7f1307a78110bcfc7c64a207"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T08:04:12.608936Z","signature_b64":"87HFDsqQvM/In1IgwC7hBuvrb9i7xgBq0k+C59mkYy+wcM9luIMcRMiAttx1AhtjdQ0sT/R6XkUwzEnTsey1Ag==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0b5163f0328771054701e20b822014057f247849b12140c58c298747a655fc50","last_reissued_at":"2026-07-05T08:04:12.608482Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T08:04:12.608482Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"FOCUS: Effective Embedding Initialization for Monolingual Specialization of Multilingual Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Gerard de Melo, Konstantin Dobler","submitted_at":"2023-05-23T19:21:53Z","abstract_excerpt":"Using model weights pretrained on a high-resource language as a warm start can reduce the need for data and compute to obtain high-quality language models for other, especially low-resource, languages. However, if we want to use a new tokenizer specialized for the target language, we cannot transfer the source model's embedding matrix. In this paper, we propose FOCUS - Fast Overlapping Token Combinations Using Sparsemax, a novel embedding initialization method that initializes the embedding matrix effectively for a new tokenizer based on information in the source model's embedding matrix. FOCU"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2305.14481","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/2305.14481/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":"2305.14481","created_at":"2026-07-05T08:04:12.608547+00:00"},{"alias_kind":"arxiv_version","alias_value":"2305.14481v2","created_at":"2026-07-05T08:04:12.608547+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2305.14481","created_at":"2026-07-05T08:04:12.608547+00:00"},{"alias_kind":"pith_short_12","alias_value":"BNIWH4BSQ5YQ","created_at":"2026-07-05T08:04:12.608547+00:00"},{"alias_kind":"pith_short_16","alias_value":"BNIWH4BSQ5YQKRYB","created_at":"2026-07-05T08:04:12.608547+00:00"},{"alias_kind":"pith_short_8","alias_value":"BNIWH4BS","created_at":"2026-07-05T08:04:12.608547+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2505.14297","citing_title":"Cross-Lingual Optimization for Language Transfer in Large Language Models","ref_index":9,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BNIWH4BSQ5YQKRYB4IFYEIAUAV","json":"https://pith.science/pith/BNIWH4BSQ5YQKRYB4IFYEIAUAV.json","graph_json":"https://pith.science/api/pith-number/BNIWH4BSQ5YQKRYB4IFYEIAUAV/graph.json","events_json":"https://pith.science/api/pith-number/BNIWH4BSQ5YQKRYB4IFYEIAUAV/events.json","paper":"https://pith.science/paper/BNIWH4BS"},"agent_actions":{"view_html":"https://pith.science/pith/BNIWH4BSQ5YQKRYB4IFYEIAUAV","download_json":"https://pith.science/pith/BNIWH4BSQ5YQKRYB4IFYEIAUAV.json","view_paper":"https://pith.science/paper/BNIWH4BS","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2305.14481&json=true","fetch_graph":"https://pith.science/api/pith-number/BNIWH4BSQ5YQKRYB4IFYEIAUAV/graph.json","fetch_events":"https://pith.science/api/pith-number/BNIWH4BSQ5YQKRYB4IFYEIAUAV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BNIWH4BSQ5YQKRYB4IFYEIAUAV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BNIWH4BSQ5YQKRYB4IFYEIAUAV/action/storage_attestation","attest_author":"https://pith.science/pith/BNIWH4BSQ5YQKRYB4IFYEIAUAV/action/author_attestation","sign_citation":"https://pith.science/pith/BNIWH4BSQ5YQKRYB4IFYEIAUAV/action/citation_signature","submit_replication":"https://pith.science/pith/BNIWH4BSQ5YQKRYB4IFYEIAUAV/action/replication_record"}},"created_at":"2026-07-05T08:04:12.608547+00:00","updated_at":"2026-07-05T08:04:12.608547+00:00"}