An Implementation of Stochastic Resonance based Machine Translation Model for Low Resource Languages using Novel Attention Models

In this paper we have proposed a new model in which the stochastic resonance technique is used to include white noise in the cell state and hidden state vectors of the LSTM model. A novel attention calculation technique is also being used in which cell state vectors are also used for attention score calculation. Earlier only hidden state or short-term memory vectors were used or attention score calculation, but in this novel approach we have used both cell state and hidden state vectors for attention score calculation. The encoder-decoder model which we have used produces better results than earlier models on the low-resource Indian languages.