BURMESE (MYANMAR) – ENGLISH NAMED ENTITY TRANSLITERATION

Abstract

This system presents a novel approach to Burmese (Myanmar) -English Named Entity Transliteration System leveraging Transformer models, focusing on character, sub- syllable, and syllable segmentation based on a meticulously prepared dictionary containing both foreign and native Myanmar-English entries. Transliterating named entities accurately between Myanmar and English poses significant challenges due to script differences, linguistic nuances, and varying entity structures. The proposed system addresses these challenges by incorporating advanced segmentation techniques and a comprehensive dictionary. The core of the approach lies in the segmentation of Myanmar named entities into character-level, sub-syllable, and syllable units, utilizing linguistic knowledge and domain-specific dictionaries. Linguistic rules are employed to segment Myanmar text into meaningful units, capturing the rich morphology and orthographic complexities of the Myanmar script. This segmentation process is crucial for accurately aligning Myanmar entities with their English transliterations. The system is built upon the Transformer architecture, a state-of-the-art deep learning model renowned for its sequence-to-sequence capabilities and attention mechanisms. The Transformer model is trained on a large corpus derived from our prepared Myanmar-English dictionary, learning the intricate mappings and transliteration patterns between the two languages. The performance of the system is evaluated using a benchmark dataset comprising diverse Myanmar named entities and their corresponding English transliterations. The experimental results demonstrate the efficacy of the approach, achieving superior transliteration accuracy compared to baseline methods. Extensive analyses are also conducted to investigate the impact of different segmentation strategies, dictionary sizes, and model configurations on transliteration quality. In conclusion, the Myanmar-English Named Entity Transliteration System based on character, sub-syllable, and syllable segmentation, coupled with a meticulously prepared dictionary, represents a significant advancement in cross-lingual natural language processing. The system offers a reliable and efficient solution for transliterating Myanmar named entities into English with exceptional accuracy and scalability, paving the way for enhanced multilingual communication and data interoperability.