Peningkatan Akurasi Long Short-Term Memory (LSTM) Menggunakan Fasttext dan Glove untuk Penerjemah Bahasa Batak-Indonesia

Abstract

Automatic translation of regional languages into Indonesian is a complex challenge in the field of natural language processing (NLP). This research aims to improve the accuracy of the Long Short Term Memory (LSTM) model in translating Batak language to Indonesian by leveraging two popular word embedding techniques, FastText and Glove. LSTM is chosen for its capability in handling long and complex sequences of data. FastText is used to capture richer word representations by considering sub-word information, while Glove is employed to obtain more global word representations based on the context in large corpora. Experiments are conducted by combining embeddings from FastText and Glove into the LSTM architecture. Results show that this combination can improve translation accuracy compared to the individual use of embeddings. Evaluation is performed using the BLEU metric to measure translation quality. This research contributes to the development of more effective regional language translation systems and demonstrates the potential of combining embedding techniques in LSTM models for other NLP applications.