Optimizing Language Augmentation for Multilingual Large Language Models: A Case Study on Korean

TL;DR

This paper presents strategies to improve multilingual large language models for less-resourced languages, demonstrated through Korean, by expanding vocabularies, using bilingual pretraining, and instruction-tuning, resulting in superior performance.

Contribution

The study introduces a comprehensive approach combining vocabulary expansion, bilingual pretraining, and instruction-tuning to enhance LLM performance for less-resourced languages like Korean.

Findings

Bllossom outperformed previous Korean models in qualitative assessments.

Vocabulary expansion improved language expressiveness.

Instruction-tuning with high-quality data enhanced task performance.

Abstract

Large language models (LLMs) use pretraining to predict the subsequent word; however, their expansion requires significant computing resources. Numerous big tech companies and research institutes have developed multilingual LLMs (MLLMs) to meet current demands, overlooking less-resourced languages (LRLs). This study proposed three strategies to enhance the performance of LRLs based on the publicly available MLLMs. First, the MLLM vocabularies of LRLs were expanded to enhance expressiveness. Second, bilingual data were used for pretraining to align the high- and less-resourced languages. Third, a high-quality small-scale instruction dataset was constructed and instruction-tuning was performed to augment the LRL. The experiments employed the Llama2 model and Korean was used as the LRL, which was quantitatively evaluated against other developed LLMs across eight tasks. Furthermore, a qualitative assessment was performed based on human evaluation and GPT4. Experimental results showed that our proposed Bllossom model exhibited superior performance in qualitative analyses compared to previously proposed Korean monolingual models.