A Rule-Based Machine Translation Framework for Low-Resource Language Pairs

Translation is not just about linguistic accuracy; it requires cultural sensitivity to preserve idiomatic expressions, cultural references, and syntactic structures unique to each language. Translating from Hindi, one of India's most widely spoken languages, to Tulu, a Dravidian language with a rich cultural and oral tradition, presents specific challenges. Tulu's colloquial phrases, cultural nuances, and emotional depth often get lost in translation, as existing tools like Google Translator, Galaxy.ai, and Lang-Lation fail to provide contextually accurate or grammatically precise translations. These tools highlight the limitations of current techniques, which often lack the ability to handle low-resource language pairs effectively. This paper identifies key challenges such as handling orthographic variations, capturing contextual meaning, and adapting cultural nuances while maintaining grammatical accuracy. To address these challenges, we propose a Rule-Based Machine Translation (RBMT) system tailored specifically for Hindi-to-Tulu translation. The system focuses on linguistic fidelity, cultural adaptation, and robust handling of syntactic and morphological differences. By leveraging curated linguistic rules, a bilingual dictionary, and advanced preprocessing techniques, this research aims to bridge the linguistic gap between Hindi and Tulu while preserving their cultural essence. The findings not only demonstrate the system's ability to produce more accurate translations but also emphasize the need for customized approaches to support low-resource language pairs in machine translation.