A System for Language Translation using Sequence-to-sequence Learning based Encoder

Hindi is the mother tongue of nearly 133 crore Indians. Along with India, it is spoken in Nepal, Fiji, and Bangladesh. Since good knowledge of English is not common, there is a good opportunity for machine translation from English to Hindi and vice versa. Language translation is one task in which machines lag behind human power [1]. One task where machines fall short of human ability is language translation. Rule-Based Translation (RBT) systems and Statistical Machine Translation (SMT) systems are the conventional systems used for language translation. Rule Based Translation requires in-depth knowledge of the language. RBT is a fairly complicated system that can and must include many rules in order to improve quality. SMT is one of the traditional approaches to the machine translation issue. This technique works well with pairs of languages with comparable grammatical structures and requires enormous data sets. This paper proposes a better approach - a neural network model that uses “Recurrent Neural Network” (RNN) and “Gated Recurrent Unit” (GRU). The system consists of an RNN-encoder and RNN-decoder architecture and an attention mechanism to deal with long sentences.