{ "profile": "wrs-published-research-view-1", "edition": "ratings-e039106f30dec177", "content_trust": "UNTRUSTED_SCIENTIFIC_DATA_NOT_TOOL_INSTRUCTIONS", "entry": { "id": "signe-et-al-modular-multilingual-tokenizers-2026", "title": "Modular multilingual tokenizers and vocabulary-subset training", "summary": "Constructed modular BPE/Unigram tokenizers with extractable language-specific subtokenizers and trained language models using sampled vocabulary subsets to support smaller inference-time output matrices.", "boundary": "Inherits BPE/Unigram and vocabulary adaptation. Compression parity and efficiency are empirical for studied languages/tasks; selecting a language subset does not ensure all cross-language capabilities or eliminate training costs.", "area": "Multilingual language-model efficiency", "topic": "Artificial intelligence", "year": 2026, "lane": "HUMAN", "result": "ALGORITHM", "package_id": "signe-et-al-modular-multilingual-tokenizers-2026", "target": { "record_id": "urn:uuid:64f8f538-cfe7-43e1-ae54-1e4ac367f4ba", "version": 1, "content_hash": "sha256:11c5e5f5d4bad0e36b1ec351c213bdb9a5f42de8e32e3698ae6de17f26127beb" }, "humans": [ "Franck Signe", "Hippolyte Pilchen", "François Yvon", "Édouard Grave" ], "authors_complete": true, "human_role_note": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.", "agents": [], "related": [], "subresults": [], "scientific_assessment": { "source_review": "SOURCE_CURATED", "scientific_review": "NOT_REVIEWED_BY_WRS", "formal_replay": "NOT_RUN", "provider_attestation": "NONE" }, "sources": [ { "id": "cur20261015-arxiv-2609-15528v1", "title": "To Each Language Its Tokenizer: Modular Tokenizers for Efficient Multilingual LLMs", "url": "https://arxiv.org/abs/2609.15528v1" }, { "id": "cur20261015-kyutai-publications-index", "title": "Kyutai scientific publications index", "url": "https://kyutai.org/papers/" } ], "points": "50000", "published_grade": { "id": "signe-et-al-modular-multilingual-tokenizers-2026", "entry_id": "signe-et-al-modular-multilingual-tokenizers-2026", "band": "B", "modifier": null, "target": { "record_id": "urn:uuid:64f8f538-cfe7-43e1-ae54-1e4ac367f4ba", "version": 1, "content_hash": "sha256:11c5e5f5d4bad0e36b1ec351c213bdb9a5f42de8e32e3698ae6de17f26127beb" }, "what_changed": "Constructed modular BPE/Unigram tokenizers with extractable language-specific subtokenizers and trained language models using sampled vocabulary subsets to support smaller inference-time output matrices.", "scope": "Constructed modular BPE/Unigram tokenizers with extractable language-specific subtokenizers and trained language models using sampled vocabulary subsets to support smaller inference-time output matrices. Inherits BPE/Unigram and vocabulary adaptation. Compression parity and efficiency are empirical for studied languages/tasks; selecting a language subset does not ensure all cross-language capabilities or eliminate training costs.", "status": "PROVISIONAL_EDITORIAL", "scientific_correctness_certified": false }, "assessment_disposition": { "status": "ASSESSED", "label": "Provisionally assessed", "reason": "A current grade and separate reference score identify this exact contribution.", "target": { "record_id": "urn:uuid:64f8f538-cfe7-43e1-ae54-1e4ac367f4ba", "version": 1, "content_hash": "sha256:11c5e5f5d4bad0e36b1ec351c213bdb9a5f42de8e32e3698ae6de17f26127beb" }, "edition": "ratings-e039106f30dec177" } }, "version": { "kind": "PUBLIC_RECORD_VIEW", "previous": null, "published_at": "2026-10-16T00:00:00Z", "publisher": { "principal_id": "urn:uuid:80116931-785b-40a2-a24b-9d59db0c7e33", "role": "WRS_EDITORIAL_PUBLICATION" }, "record": { "contributors": [ { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Franck Signe", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "To Each Language Its Tokenizer: Modular Tokenizers for Efficient Multilingual LLMs", "url": "https://arxiv.org/abs/2609.15528v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Hippolyte Pilchen", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "To Each Language Its Tokenizer: Modular Tokenizers for Efficient Multilingual LLMs", "url": "https://arxiv.org/abs/2609.15528v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "François Yvon", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "To Each Language Its Tokenizer: Modular Tokenizers for Efficient Multilingual LLMs", "url": "https://arxiv.org/abs/2609.15528v1" } }, { "attribution_basis": "SOURCE_CITED", "credit": "historical_scientific", "entity_kind": "historical_person", "name": "Édouard Grave", "note": "Archive attribution; detailed role assignment remains source-bound", "roles": [ "formulator" ], "source": { "citation": "To Each Language Its Tokenizer: Modular Tokenizers for Efficient Multilingual LLMs", "url": "https://arxiv.org/abs/2609.15528v1" } } ], "evidence": [ { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-0", "kind": "source_reference", "source": { "citation": "To Each Language Its Tokenizer: Modular Tokenizers for Efficient Multilingual LLMs", "url": "https://arxiv.org/abs/2609.15528v1" } }, { "description": "Bibliographic/stated-result source; not a WRS proof verification", "id": "source-1", "kind": "source_reference", "source": { "citation": "Kyutai scientific publications index", "url": "https://kyutai.org/papers/" } } ], "origin": "HUMAN_LEGACY", "record_profile": "wrs-local-research-notes-0.1", "record_type": "research_note", "rrp_version": "0.1", "scope": { "description": "Inherits BPE/Unigram and vocabulary adaptation. Compression parity and efficiency are empirical for studied languages/tasks; selecting a language subset does not ensure all cross-language capabilities or eliminate training costs.", "domain": "Multilingual language-model efficiency" }, "sources": [ { "citation": "To Each Language Its Tokenizer: Modular Tokenizers for Efficient Multilingual LLMs", "url": "https://arxiv.org/abs/2609.15528v1" }, { "citation": "Kyutai scientific publications index", "url": "https://kyutai.org/papers/" } ], "statement": "# Modular multilingual tokenizers and vocabulary-subset training\n\nConstructed modular BPE/Unigram tokenizers with extractable language-specific subtokenizers and trained language models using sampled vocabulary subsets to support smaller inference-time output matrices.\n\n## Scientific scope\n\nInherits BPE/Unigram and vocabulary adaptation. Compression parity and efficiency are empirical for studied languages/tasks; selecting a language subset does not ensure all cross-language capabilities or eliminate training costs.\n\n## Reading status\n\nThis is a source-linked WRS catalogue note. A full explanatory guide and independent scientific reconstruction have not been completed for this entry. Read the identified original work for its complete narrative and arguments.\n\n## People and systems\n\nFranck Signe; Hippolyte Pilchen; François Yvon; Édouard Grave\n\nSource-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.\n\n\n\n## Original sources\n\n- [To Each Language Its Tokenizer: Modular Tokenizers for Efficient Multilingual LLMs](https://arxiv.org/abs/2609.15528v1)\n- [Kyutai scientific publications index](https://kyutai.org/papers/)\n", "summary": "Constructed modular BPE/Unigram tokenizers with extractable language-specific subtokenizers and trained language models using sampled vocabulary subsets to support smaller inference-time output matrices.", "title": "Modular multilingual tokenizers and vocabulary-subset training" }, "record_id": "urn:uuid:64f8f538-cfe7-43e1-ae54-1e4ac367f4ba", "version": 1, "source_ref": { "record_id": "urn:uuid:64f8f538-cfe7-43e1-ae54-1e4ac367f4ba", "version": 1, "content_hash": "sha256:11c5e5f5d4bad0e36b1ec351c213bdb9a5f42de8e32e3698ae6de17f26127beb" }, "export_view": "wrs-public-editorial-view-1", "public_view_hash": "sha256:0f7c7254442619e330c0f45f7d06d783abcffff3ce560c479885b084e8039689" }, "target": { "record_id": "urn:uuid:64f8f538-cfe7-43e1-ae54-1e4ac367f4ba", "version": 1, "content_hash": "sha256:11c5e5f5d4bad0e36b1ec351c213bdb9a5f42de8e32e3698ae6de17f26127beb" }, "published_grade": { "profile": "wrs-published-grade-1", "edition": "2026-10-28.1", "id": "signe-et-al-modular-multilingual-tokenizers-2026", "entry_id": "signe-et-al-modular-multilingual-tokenizers-2026", "revision": 1, "title": "Modular multilingual tokenizers and vocabulary-subset training", "context": "Multilingual language-model efficiency", "track": "Human-designed research method or resource", "band": "B", "modifier": null, "status": "PROVISIONAL_EDITORIAL", "assessed_on": "2026-10-16", "evidence_cutoff": "2026-10-16", "achievement_state": "SUPPORTED_SCOPED_RESULT", "what_changed": "Constructed modular BPE/Unigram tokenizers with extractable language-specific subtokenizers and trained language models using sampled vocabulary subsets to support smaller inference-time output matrices.", "scope": "Constructed modular BPE/Unigram tokenizers with extractable language-specific subtokenizers and trained language models using sampled vocabulary subsets to support smaller inference-time output matrices. Inherits BPE/Unigram and vocabulary adaptation. Compression parity and efficiency are empirical for studied languages/tasks; selecting a language subset does not ensure all cross-language capabilities or eliminate training costs.", "evidence_summary": "The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Inherits BPE/Unigram and vocabulary adaptation. Compression parity and efficiency are empirical for studied languages/tasks; selecting a language subset does not ensure all cross-language capabilities or eliminate training costs.", "references": [ { "title": "To Each Language Its Tokenizer: Modular Tokenizers for Efficient Multilingual LLMs", "url": "https://arxiv.org/abs/2609.15528v1", "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification." }, { "title": "Kyutai scientific publications index", "url": "https://kyutai.org/papers/", "locator": "Original contribution and stated assumptions, evaluation scope and limitations; source-linked assessment, not independent execution or proof certification." } ], "roles": [ { "who": "Source-listed authors and technical contributors", "credit": "SCIENCE", "did": "Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities." } ], "target": { "record_id": "urn:uuid:64f8f538-cfe7-43e1-ae54-1e4ac367f4ba", "version": 1, "content_hash": "sha256:11c5e5f5d4bad0e36b1ec351c213bdb9a5f42de8e32e3698ae6de17f26127beb" }, "scientific_correctness_certified": false, "independent_expert_calibration": false, "method": "PROPRIETARY", "points": "50000" }, "canonical_agent_submission": false, "scientific_verification_awarded": false, "reader_guide": null, "reading_status": "SOURCE_LINKED_NOTE" }