From English-Centric to Effective Bilingual: LLMs with Custom Tokenizers for Underrepresented Languages

TL;DR

This paper introduces a cost-effective, model-agnostic method for developing bilingual large language models that support underrepresented languages using custom tokenizers, vocabulary expansion, and new evaluation metrics.

Contribution

It presents a novel approach combining vocabulary expansion and training techniques to improve underrepresented language support in LLMs while reducing computational costs.

Findings

Improved language performance for Ukrainian, Arabic, and Georgian

Reduced computational costs compared to traditional methods

Vocabulary size significantly affects generated text quality

Abstract

In this paper, we propose a model-agnostic cost-effective approach to developing bilingual base large language models (LLMs) to support English and any target language. The method includes vocabulary expansion, initialization of new embeddings, model training and evaluation. We performed our experiments with three languages, each using a non-Latin script - Ukrainian, Arabic, and Georgian. Our approach demonstrates improved language performance while reducing computational costs. It mitigates the disproportionate penalization of underrepresented languages, promoting fairness and minimizing adverse phenomena such as code-switching and broken grammar. Additionally, we introduce new metrics to evaluate language quality, revealing that vocabulary size significantly impacts the quality of generated text.