@inproceedings{cho-etal-2026-vocmoe,
title = "{V}oc{M}o{E}: Vocabulary-Guided Mixture-of-Experts for {K}orean Medical Language Models",
author = "Cho, Sangyeon and
Han, Jaeho and
Jeon, Mingyu and
Kim, Junyeong",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.607/",
pages = "12796--12815",
ISBN = "979-8-89176-462-0",
abstract = "In Korean medical LLMs, domain-trained models underperform General LLMs of comparable parameter scale. We attribute this *paradoxical failure* to a *cascading two-bottleneck* structure: tokenizer over-fragmentation (*Fragmentation bottleneck, B1*) compounded by a learning-stage representation-alignment failure (*Alignment bottleneck, B2*). Using the Korean medical domain as a case study, we diagnose this paradox and propose **VocMoE**, which is designed to address both bottlenecks: DVE selects domain vocabulary from a medical BERT via an information-theoretic score, and MoDE uses the DVE output (the partition of token ids) as the vocabulary-conditioned, deterministic routing key of the MoE FFN. In this view, the vocabulary partition itself serves as a routing prior, shifting routing from hidden-state learning to vocabulary identity. MediKo-9B-A4B reaches simultaneous SOTA on KorMedMCQA (**57.5**, +14.8 pt), KMMLU Medical, and HAE-RAE, while MediKo-4B-A1B (active 1.69B) surpasses Qwen3-8B (8B) and KULLM3 (10.7B) on KorMedMCQA, updating the active-parameter efficiency frontier."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="cho-etal-2026-vocmoe">
<titleInfo>
<title>VocMoE: Vocabulary-Guided Mixture-of-Experts for Korean Medical Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Sangyeon</namePart>
<namePart type="family">Cho</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jaeho</namePart>
<namePart type="family">Han</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mingyu</namePart>
<namePart type="family">Jeon</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Junyeong</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: EMNLP 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>In Korean medical LLMs, domain-trained models underperform General LLMs of comparable parameter scale. We attribute this *paradoxical failure* to a *cascading two-bottleneck* structure: tokenizer over-fragmentation (*Fragmentation bottleneck, B1*) compounded by a learning-stage representation-alignment failure (*Alignment bottleneck, B2*). Using the Korean medical domain as a case study, we diagnose this paradox and propose **VocMoE**, which is designed to address both bottlenecks: DVE selects domain vocabulary from a medical BERT via an information-theoretic score, and MoDE uses the DVE output (the partition of token ids) as the vocabulary-conditioned, deterministic routing key of the MoE FFN. In this view, the vocabulary partition itself serves as a routing prior, shifting routing from hidden-state learning to vocabulary identity. MediKo-9B-A4B reaches simultaneous SOTA on KorMedMCQA (**57.5**, +14.8 pt), KMMLU Medical, and HAE-RAE, while MediKo-4B-A1B (active 1.69B) surpasses Qwen3-8B (8B) and KULLM3 (10.7B) on KorMedMCQA, updating the active-parameter efficiency frontier.</abstract>
<identifier type="citekey">cho-etal-2026-vocmoe</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.607/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>12796</start>
<end>12815</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T VocMoE: Vocabulary-Guided Mixture-of-Experts for Korean Medical Language Models
%A Cho, Sangyeon
%A Han, Jaeho
%A Jeon, Mingyu
%A Kim, Junyeong
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F cho-etal-2026-vocmoe
%X In Korean medical LLMs, domain-trained models underperform General LLMs of comparable parameter scale. We attribute this *paradoxical failure* to a *cascading two-bottleneck* structure: tokenizer over-fragmentation (*Fragmentation bottleneck, B1*) compounded by a learning-stage representation-alignment failure (*Alignment bottleneck, B2*). Using the Korean medical domain as a case study, we diagnose this paradox and propose **VocMoE**, which is designed to address both bottlenecks: DVE selects domain vocabulary from a medical BERT via an information-theoretic score, and MoDE uses the DVE output (the partition of token ids) as the vocabulary-conditioned, deterministic routing key of the MoE FFN. In this view, the vocabulary partition itself serves as a routing prior, shifting routing from hidden-state learning to vocabulary identity. MediKo-9B-A4B reaches simultaneous SOTA on KorMedMCQA (**57.5**, +14.8 pt), KMMLU Medical, and HAE-RAE, while MediKo-4B-A1B (active 1.69B) surpasses Qwen3-8B (8B) and KULLM3 (10.7B) on KorMedMCQA, updating the active-parameter efficiency frontier.
%U https://aclanthology.org/2026.findings-emnlp.607/
%P 12796-12815
Markdown (Informal)
[VocMoE: Vocabulary-Guided Mixture-of-Experts for Korean Medical Language Models](https://aclanthology.org/2026.findings-emnlp.607/) (Cho et al., Findings 2026)
ACL