Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Efficient Terminology Integration for LLM-based Translation in Specialized Domains

Sejoon Kim, Mingi Sung, Jeonghwan Lee, Hyunkuk Lim, Jorge Froilan Gimenez Perez

TL;DR

This paper introduces a methodology that efficiently trains models with a smaller amount of data while preserving the accuracy of terminology translation, through a systematic process of term extraction and glossary creation using the Trie Tree algorithm.

Abstract

Traditional machine translation methods typically involve training models directly on large parallel corpora, with limited emphasis on specialized terminology. However, In specialized fields such as patent, finance, or biomedical domains, terminology is crucial for translation, with many terms that needs to be translated following agreed-upon conventions. In this paper we introduce a methodology that efficiently trains models with a smaller amount of data while preserving the accuracy of terminology translation. We achieve this through a systematic process of term extraction and glossary creation using the Trie Tree algorithm, followed by data reconstruction to teach the LLM how to integrate these specialized terms. This methodology enhances the model's ability to handle specialized terminology and ensures high-quality translations, particularly in fields where term consistency is crucial. Our approach has demonstrated exceptional performance, achieving the highest translation score among participants in the WMT patent task to date, showcasing its effectiveness and broad applicability in specialized translation domains where general methods often fall short.

Efficient Terminology Integration for LLM-based Translation in Specialized Domains

TL;DR

This paper introduces a methodology that efficiently trains models with a smaller amount of data while preserving the accuracy of terminology translation, through a systematic process of term extraction and glossary creation using the Trie Tree algorithm.

Abstract

Traditional machine translation methods typically involve training models directly on large parallel corpora, with limited emphasis on specialized terminology. However, In specialized fields such as patent, finance, or biomedical domains, terminology is crucial for translation, with many terms that needs to be translated following agreed-upon conventions. In this paper we introduce a methodology that efficiently trains models with a smaller amount of data while preserving the accuracy of terminology translation. We achieve this through a systematic process of term extraction and glossary creation using the Trie Tree algorithm, followed by data reconstruction to teach the LLM how to integrate these specialized terms. This methodology enhances the model's ability to handle specialized terminology and ensures high-quality translations, particularly in fields where term consistency is crucial. Our approach has demonstrated exceptional performance, achieving the highest translation score among participants in the WMT patent task to date, showcasing its effectiveness and broad applicability in specialized translation domains where general methods often fall short.

Paper Structure

This paper contains 18 sections, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Construction of the Terminology Glossary: Terminology Aligner
  4. Term Identification in the Source Text: Trie-Tree Algorithm
  5. Application of Terms in Translation: LLM Prompting and sLLM Fine-Tuning
  6. Experimental Results and Application
  7. Discussion
  8. Advantages of Our Methodology Over Traditional Approaches
  9. Focused Learning on Domain-Specific Terms
  10. Efficient Data Utilization through Terminology Extraction
  11. Enhanced Translation Consistency and Accuracy
  12. Improved Model Generalization
  13. Customizability for Different Domains
  14. Conclusion
  15. Appendix
  16. ...and 3 more sections

Figures (4)

  • Figure 1: Training method in terminology-based LLM translation
  • Figure 2: Instructions for Term Extraction
  • Figure 3: Overall process of term extraction to translation
  • Figure 4: Instructions for Term Extraction