A Deep Learning Approach for Word Segmentation in Javanese Letter Manuscript Transliteration

Traditionally written using the Javanese Letter or 'Aksara Jawa' script, the Javanese language encompasses a rich corpus of manuscripts that record diverse subjects such as history, culture, and traditional practices. With over 121,668 Javanese manuscript titles identified, only a fraction has been transliterated into alphabetical writing, underscoring the urgent need for efficient language preservation methods. This study evaluates deep learning models for word segmentation in Javanese manuscript transliteration. Results from experiments conducted on an unseen dataset reveal that the Bidirectional Long Short-Term Memory (BiLSTM) model outperforms other architectures consistently across all metrics. With an accuracy of 98.62% and superior scores in f1-score, precision, and recall, the BiLSTM model demonstrates robustness in capturing the intricate linguistic patterns and textual structures inherent in Javanese manuscripts.