Low-resource Machine Translation for Code-switched Kazakh-Russian Language Pair

2025-03-25

AI Summary

This study addresses the low-resource Kazakh–Russian code-switching machine translation (CSMT) task by proposing a fully unsupervised end-to-end modeling approach. Confronted with the dual challenges of scarce annotated parallel corpora and pervasive code-mixing, we first construct the first publicly available Kazakh–Russian code-switching parallel corpus and design a synthetic data generation framework integrating back-translation and large language models (LLMs). Our method combines unsupervised neural machine translation (UNMT) with domain-adaptive fine-tuning. The key contribution lies in generating high-quality, code-mixed training data without human annotation—thereby overcoming both low-resource and code-mixing bottlenecks. Experimental results show that our model achieves 16.48 BLEU on standard test sets, approaching the performance of commercial systems, and significantly outperforms them in human evaluation.

Technology Category

• Natural Language Processing: Code Generation / Program Synthesis from Natural Language
• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data

Abstract

Machine translation for low resource language pairs is a challenging task. This task could become extremely difficult once a speaker uses code switching. We propose a method to build a machine translation model for code-switched Kazakh-Russian language pair with no labeled data. Our method is basing on generation of synthetic data. Additionally, we present the first codeswitching Kazakh-Russian parallel corpus and the evaluation results, which include a model achieving 16.48 BLEU almost reaching an existing commercial system and beating it by human evaluation.

Problem

• Develop machine translation for low-resource Kazakh-Russian code-switching
• Create synthetic data to train without labeled examples
• Build first parallel corpus and evaluate against commercial systems

Innovation

• Synthetic data generation for low-resource translation
• First Kazakh-Russian codeswitching parallel corpus
• BLEU-competitive model without labeled data