Neural Machine Translation for Low-Resource Languages: Experiments on Makassar-Indonesian

In today’s globalized world, many indigenous languages are at risk of extinction due to the dominance of widely spoken languages and their declining use. Neural machine translation (NMT) offers a promising approach to documenting and preserving low-resource languages by using advanced technology to support their continued use and development. This study aims to develop an NMT-based translation system for the Indonesian and Makassar languages and assess its performance. Experiments were conducted using two base models, Transformer and LSTM, with various training steps through OpenNMT tools. The dataset comprised a parallel corpus of 9,165 sentence pairs. Results presented that the Transformer-based model outperformed the LSTM-based model, achieving BLEU scores of 56.45 for Indonesian to Makassar translation and 34.78 for Makassar to Indonesian. These findings demonstrate the potential of NMT to assist in the documentation and preservation of low-resource languages.