A variety-aware retrieval-augmented generation framework for terminology support in Chinese medical interpreting

Standard machine translation and large language models frequently treat Chinese medical terms as a single unified target, overlooking critical regional variation across Mainland China, Hong Kong, and Taiwan. This study frames regional terminology assistance as a variety-aware selection problem in which terminology linked to a knowledge base (KB) remains distinguishable from provisional wording. VACMI-LLM is a retrieval-augmented generation (RAG) framework that combines a concept-centred three-region medical KB, threshold-gated retrieval, and structured output with explicit KB-support labels. Evaluation used a constructed 700-expression retrieval stress set and a separate 300-case raw-query pipeline set. At an exploratory threshold of 0.90, the system achieved 95.2% Recall@3 for KB-supported expressions and returned no KB concept for 78.5% of KB-external expressions. In E2, reference-row accuracy for KB-supported expressions was similar between VACMI-LLM and ungated RAG, whereas KB-support labelling for external expressions was 96.4% versus 15.9%. Eight external professionals rated five fixed displays highly for terminology accuracy, regional appropriateness, and usefulness for interpreter preparation. These findings show how region-labelled knowledge and explicit support status can make terminology assistance more traceable under controlled conditions.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-13
DOI
https://doi.org/10.1038/s41598-026-73832-w
Primary Topic
Natural Language Processing Techniques
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A variety-aware retrieval-augmented generation framework for terminology support in Chinese medical interpreting

Andrew K. F. Cheung, Yingqi Huang, Ruiyang Ma, Xianru Wang
2 links
Scientific Reports
Natural Language Processing Techniques
article

A variety-aware retrieval-augmented generation framework for terminology support in Chinese medical interpreting

Andrew K. F. Cheung, Yingqi Huang, Ruiyang Ma, Xianru Wang
article en

Abstract

Standard machine translation and large language models frequently treat Chinese medical terms as a single unified target, overlooking critical regional variation across Mainland China, Hong Kong, and Taiwan. This study frames regional terminology assistance as a variety-aware selection problem in which terminology linked to a knowledge base (KB) remains distinguishable from provisional wording. VACMI-LLM is a retrieval-augmented generation (RAG) framework that combines a concept-centred three-region medical KB, threshold-gated retrieval, and structured output with explicit KB-support labels. Evaluation used a constructed 700-expression retrieval stress set and a separate 300-case raw-query pipeline set. At an exploratory threshold of 0.90, the system achieved 95.2% Recall@3 for KB-supported expressions and returned no KB concept for 78.5% of KB-external expressions. In E2, reference-row accuracy for KB-supported expressions was similar between VACMI-LLM and ungated RAG, whereas KB-support labelling for external expressions was 96.4% versus 15.9%. Eight external professionals rated five fixed displays highly for terminology accuracy, regional appropriateness, and usefulness for interpreter preparation. These findings show how region-labelled knowledge and explicit support status can make terminology assistance more traceable under controlled conditions.

Scientific Reports
Hong Kong Polytechnic University (HK), Southeast University (CN)
Natural Language Processing Techniques
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon