{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:5QGGTNODICGZQ67O7FDJIEDYPJ","short_pith_number":"pith:5QGGTNOD","schema_version":"1.0","canonical_sha256":"ec0c69b5c3408d987beef9469410787a500b6bba2e6e83cef56105d35e598596","source":{"kind":"arxiv","id":"2407.08818","version":2},"attestation_state":"computed","paper":{"title":"MAGNET: Improving the Multilingual Fairness of Language Models with Adaptive Gradient-Based Tokenization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Hila Gonen, Noah A. Smith, Orevaoghene Ahia, Sachin Kumar, Tomasz Limisiewicz, Valentin Hofmann, Yulia Tsvetkov","submitted_at":"2024-07-11T18:59:21Z","abstract_excerpt":"In multilingual settings, non-Latin scripts and low-resource languages are usually disadvantaged in terms of language models' utility, efficiency, and cost. Specifically, previous studies have reported multiple modeling biases that the current tokenization algorithms introduce to non-Latin script languages, the main one being over-segmentation. In this work, we propose MAGNET; multilingual adaptive gradient-based tokenization to reduce over-segmentation via adaptive gradient-based subword tokenization. MAGNET learns to predict segment boundaries between byte tokens in a sequence via sub-module"},"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":"2407.08818","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2024-07-11T18:59:21Z","cross_cats_sorted":[],"title_canon_sha256":"668b8900bb0ff0b84f9d7c4f19982f4089c1e4d954200b772161c37a44932502","abstract_canon_sha256":"d2e9e7d0182b296a05d3b51f5cd71e25a30be98a8db20b80795bd9459539f58a"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T09:36:13.089365Z","signature_b64":"4VrCAXDo189uMCrpmizQ73SeNnWHv0MFEDhIcaflIvgPDz7+jM1GocBHoGxw4rNXzM4iadl0L0nt5cCn8SAwCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"ec0c69b5c3408d987beef9469410787a500b6bba2e6e83cef56105d35e598596","last_reissued_at":"2026-07-05T09:36:13.088911Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T09:36:13.088911Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"MAGNET: Improving the Multilingual Fairness of Language Models with Adaptive Gradient-Based Tokenization","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Hila Gonen, Noah A. Smith, Orevaoghene Ahia, Sachin Kumar, Tomasz Limisiewicz, Valentin Hofmann, Yulia Tsvetkov","submitted_at":"2024-07-11T18:59:21Z","abstract_excerpt":"In multilingual settings, non-Latin scripts and low-resource languages are usually disadvantaged in terms of language models' utility, efficiency, and cost. Specifically, previous studies have reported multiple modeling biases that the current tokenization algorithms introduce to non-Latin script languages, the main one being over-segmentation. In this work, we propose MAGNET; multilingual adaptive gradient-based tokenization to reduce over-segmentation via adaptive gradient-based subword tokenization. MAGNET learns to predict segment boundaries between byte tokens in a sequence via sub-module"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2407.08818","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/2407.08818/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":"2407.08818","created_at":"2026-07-05T09:36:13.088973+00:00"},{"alias_kind":"arxiv_version","alias_value":"2407.08818v2","created_at":"2026-07-05T09:36:13.088973+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2407.08818","created_at":"2026-07-05T09:36:13.088973+00:00"},{"alias_kind":"pith_short_12","alias_value":"5QGGTNODICGZ","created_at":"2026-07-05T09:36:13.088973+00:00"},{"alias_kind":"pith_short_16","alias_value":"5QGGTNODICGZQ67O","created_at":"2026-07-05T09:36:13.088973+00:00"},{"alias_kind":"pith_short_8","alias_value":"5QGGTNOD","created_at":"2026-07-05T09:36:13.088973+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2602.04289","citing_title":"Proxy Compression for Language Modeling","ref_index":1,"is_internal_anchor":false},{"citing_arxiv_id":"2605.09630","citing_title":"Scratchpad Patching: Decoupling Compute from Patch Size in Byte-Level Language Models","ref_index":1,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/5QGGTNODICGZQ67O7FDJIEDYPJ","json":"https://pith.science/pith/5QGGTNODICGZQ67O7FDJIEDYPJ.json","graph_json":"https://pith.science/api/pith-number/5QGGTNODICGZQ67O7FDJIEDYPJ/graph.json","events_json":"https://pith.science/api/pith-number/5QGGTNODICGZQ67O7FDJIEDYPJ/events.json","paper":"https://pith.science/paper/5QGGTNOD"},"agent_actions":{"view_html":"https://pith.science/pith/5QGGTNODICGZQ67O7FDJIEDYPJ","download_json":"https://pith.science/pith/5QGGTNODICGZQ67O7FDJIEDYPJ.json","view_paper":"https://pith.science/paper/5QGGTNOD","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2407.08818&json=true","fetch_graph":"https://pith.science/api/pith-number/5QGGTNODICGZQ67O7FDJIEDYPJ/graph.json","fetch_events":"https://pith.science/api/pith-number/5QGGTNODICGZQ67O7FDJIEDYPJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/5QGGTNODICGZQ67O7FDJIEDYPJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/5QGGTNODICGZQ67O7FDJIEDYPJ/action/storage_attestation","attest_author":"https://pith.science/pith/5QGGTNODICGZQ67O7FDJIEDYPJ/action/author_attestation","sign_citation":"https://pith.science/pith/5QGGTNODICGZQ67O7FDJIEDYPJ/action/citation_signature","submit_replication":"https://pith.science/pith/5QGGTNODICGZQ67O7FDJIEDYPJ/action/replication_record"}},"created_at":"2026-07-05T09:36:13.088973+00:00","updated_at":"2026-07-05T09:36:13.088973+00:00"}