Comparing LSTM Algorithm with Word Embedding: FastText and Word2Vec in Bahasa Batak-English Translation

The Long-Short Term Memory (LSTM) algorithm is an advancement of the Recurrent Neural Network (RNN) method inspired by the thought patterns of humans. LSTMsand RNNs are often employed in machine translators to address sequential issues in text and provide accurate translations. In this problem, LSTM cannot stand alone. it requires word embeddings for low-dimensional vector representation of words, where similar or closely related words are placed closer to each other. With word embeddings, the semantic and contextual meanings of words can be better represented, allowing the model to understand and work with the text more effectively. In this paper, we propose using FastText and Word2Vec word embeddings to enhance the performance of the LSTM algorithm and compare the LSTM's performance with these word embeddings. Our results indicate that FastText embeddings improve the quality of Batak-English translations more effectively than Word2Vec. Our research suggests that FastText is more accurate and efficient in sentence translation.