A Deep Learning Based Translation Performance Analysis of International Languages

Machine Translation (MT) is a specialized domain within the broader field of Natural Language Processing (NLP) that utilizes machine learning algorithms to translate speech or text from one language to another language without human participation. The goal of such system construction is to find relatively high accuracy, cost effective and low error in the outcome. Numerous MT systems have been developed and deployed for widespread use with many individuals, particularly learners encountering difficulties while utilizing them. In recent times Recurrent Neural Networks (RNNs) and Long Short Term Memory (LSTM) techniques have shown promising results for machine translation as compared to other existing approaches of Neural Machine Translation (NMT). The current work compares performances of NMT and RNN methods in different international datasets (FRENCH-ENGLISH, RUSSIAN-ENGLISH, and ITALIANENGLISH). Their translation errors have been calculated by using Sparse Categorical Crossentropy loss function. We have observed the effect of size of the data in specified settings and found that there is no significant change in loss function after 50k training data.