LSTM Model Using Adam’s Optimizer for Indonesian – Bugis Bidirectional Translation System

The purpose of this research is to develop a machine translation of Bugis to Indonesian and vice versa in order to preserve the Bugis language. This research utilizes a recent dataset consisting of 30,000 Bugis-Indonesian sentence pairs from the online Bible. This research conducts scraping to compile the corpus which is then followed by manual and automatic pre-processing. The method chosen is Neural Machine Translation (NMT) while for training and testing models Long Short-Term Memory (LSTM) is used. The performance of the model is evaluated by Bilingual Evaluation Understudy (BLEU) score to measure the translation accuracy at various epochs. In addition, this study also compared the use of Adam's optimizer with non-optimizer. The results showed that the use of Adam's optimizer significantly improved the performance of the model where at epoch 2000 the model achieved the highest BLEU score of 0.996261 indicating highly accurate translation quality. In contrast, the model without the optimizer showed lower performance. Other results also found that the translation from Bugis to Indonesian was more accurate than from Indonesian to Bugis. This is due to the more balanced word count difference in the Bugis to Indonesian translation, which makes it easier for the model to match words. In conclusion, the use of NMT with Adam optimizer effectively improves the accuracy of two-way translation from Bugis-Indonesian.