Study of Neural Machine Translation With Long Short-Term Memory Techniques

Abstract

The growing demand for having a conversation amongst people who come from different areas, across the globe, resulting from globalization, has led to the development of systems like machine translations. There are techniques like statistical models, Bayesian models, etc. that were used earlier for machine translations. However, with growing expectations towards better accuracies, neural networks aided systems for translations termed as neural machine translations (NMT) have come up. Models have been proposed by several organizations like Google NMT (G-NMT) that are widely accepted and implemented. Several machine translations are also based on RNN models. This work studies neural machine translations with respect to long short-term memory (LSTM) network and compares them on the basis of several widely accepted accuracy metrics like BLEU score, precision, recall, and F1 score. Further, a combination of two LSTM models is implemented for better accuracy. This work analyzes the various LSTM models on the basis of these metrics.