Deep Learning Approach in Machine Translation for Indian context: A Survey and Discussion

Abstract

Due to availability of abundant data and increase speed in processing by infrastructure, the complex problems in several domains like Speech Processing, Image Processing, Natural Language Processing, Machine Learning etc. got appealing solutions which simplifies life of human being. Due to which crucial problem like Machine Translation (MT) shifted its development from statistical machine translation to Neural Machine Translation (NMT). A new statistical environment is been adapted for faster and efficient solution to MT. Various new approaches of neural network added which covers Artificial Neural Network, Convolutional Neural Network, Feed Forward Neural Network, Encoder Decoder Architecture. Also use of Long Short Term Memory (LSTM), Attention Factor is included to speed up processing. Languages with homogeneous structure like English and French, gives more promising results, but fails to give such results for languages having heterogeneous structures. In Indian context due to scarcity of data and strong morphological structure of Indian languages it is difficult to deal with translation with traditional MT approaches. More focus is to be given on various underline structures of languages like part of speech (POS), morphological structure of languages and relatedness to similar domain need to be addressed to improve performance. In this paper approaches of various researches for Machine Translation specifically in Indian domain are discussed. It can be said that for languages belonging to same domain have fluent translations and accuracy as compared with languages in heterogeneous domain. Accuracy can further be improved by adding Attention parameter.