A Conceptual Framework for Malay-English Code-Switched Neural Machine Translation

This paper presents a conceptual framework for addressing the challenges in translating Malay-English code-switched texts using neural machine translation. The framework comprises of three phases: language identification, code-switching type identification, and segment translation. In the language identification phase, a trained model attaches language tags (M for Malay, E for English, M-E for ambiguous) to each word. Code-switching types, including intra-sentential (AM and AE) and inter-sentential (EM and EE), are identified in the code-switching type identification phase. The segment translation phase utilizes an RNN model trained on a Malay-English code-switched parallel corpus and a homonyms dictionary with POS tagging. Our framework addresses linguistic characteristics, informal language usage, structural differences, and ambiguity. It contributes to the advancement of machine translation in code-switching contexts. Despite the conceptual nature of the framework without concrete results, our thorough analysis of code-switching types and associated challenges lays a foundation for future model enhancements. By providing a comprehensive solution, it enables more accurate and effective communication in code-switched language scenarios. Further research can build upon this framework to enhance code-switched translation models.