Advancements in Natural Language Understanding- Driven Machine Translation: Focus on English and the Low Resource Dialectal Lusoga

Abstract

This review explores recent advancements in Natural Language Understanding-driven Machine Translation (NLU-MT) with a focus on English and the low-resource dialectal Lusoga. A Low-resource language, such as Lusoga, faces significant challenges in Machine Translation (MT) due to the scarcity of high-quality parallel corpora, the complex morphology inherent in Bantu languages, and the dialectal variations within Lusoga itself, particularly between Lutenga and Lupakoyo. This paper examines the role of NLU-based MT systems in overcoming these challenges by shifting from word-for-word mapping to meaning-based translations, enabling better handling of these dialectal differences. We highlight the success of leveraging linguistic similarities between Lusoga and related languages, such as Luganda, to improve translation performance through multilingual transfer learning techniques. Key advancements include the use of transformer-based architectures such as Multilingual Bidirectional and Auto-Regressive Transformer (mBART) and Multilingual Text-To-Text Transfer Transformer (mT5), specifically selected for their effectiveness in NLU-driven contexts, which have shown promise in enhancing translation accuracy for African low-resource languages. However, the review also identifies ongoing obstacles, including historical low demand and the lack of well-developed corpora, which hinder scalability. The paper concludes by emphasizing the potential of hybrid approaches that combine community-driven corpus-building initiatives with improved model architectures to drive further progress in low-resource MT. Ultimately, NLU-MT is positioned as a crucial tool not only for bridging communication gaps but also for preserving linguistic diversity and cultural heritage.