Leveraging Sentence-oriented Augmentation and Transformer-Based Architecture for Vietnamese-Bahnaric Translation

TL;DR

This work tackles the challenge of translating Vietnamese to Bahnar in a low-resource setting by applying two data-augmentation strategies within a transformer-based NMT framework. It demonstrates that multi-task learning data augmentation (MTL-DA) and sentence boundary augmentation substantially improve BLEU scores, with token+swap and high-coverage boundary methods delivering the strongest gains (around 41 BLEU on a sizable Bahnar corpus). The methods do not require complex preprocessing or additional data beyond existing parallel corpora, making them practical for resource-constrained language pairs and valuable for Bahnar language preservation. The results underscore the potential of targeted augmentation to mitigate data scarcity and segmentation issues in low-resource translation tasks, and suggest directions for broader application to dialects and other under-resourced languages.

Abstract

The Bahnar people, an ethnic minority in Vietnam with a rich ancestral heritage, possess a language of immense cultural and historical significance. The government places a strong emphasis on preserving and promoting the Bahnaric language by making it accessible online and encouraging communication across generations. Recent advancements in artificial intelligence, such as Neural Machine Translation (NMT), have brought about a transformation in translation by improving accuracy and fluency. This, in turn, contributes to the revival of the language through educational efforts, communication, and documentation. Specifically, NMT is pivotal in enhancing accessibility for Bahnaric speakers, making information and content more readily available. Nevertheless, the translation of Vietnamese into Bahnaric faces practical challenges due to resource constraints, especially given the limited resources available for the Bahnaric language. To address this, we employ state-of-the-art techniques in NMT along with two augmentation strategies for domain-specific Vietnamese-Bahnaric translation task. Importantly, both approaches are flexible and can be used with various neural machine translation models. Additionally, they do not require complex data preprocessing steps, the training of additional systems, or the acquisition of extra data beyond the existing training parallel corpora.

Figures (3)

• Figure 1: The commonly used methods of DA for NMT
• Figure 2: Taxonomy of DA methods
• Figure 3: General pipeline of augmenting, training and evaluating process