Neural Machine Translation

The objective of this project is to convert Hindi sentence to English sentence for non-native speakers using Neural Machine Translation. The dataset is preprocessed by removing punctuations and all duplicate values as well as all quotes, numbers and white spaces that are present in text data. Then English and Hindi vocabulary is prepared by splitting sentences into words. Encoder Decoder architecture is used to build the model wherein decoder is an LSTM model with embedding layer, sequence to sequence layer, attention mechanism and dense layer. The model is appropriately trained with the train data and saved the model weights, which are loaded while inferencing the test data for making the prediction. The predicted English translation will be obtained from decoded sequence function.

Salient Features:
– The state of the art architecture for language translation
– Adopting proven NLP technique
– Model inferencing with model weights
– Can easily be extended to other native language to be translated to English or vice versa
– Enhanced customer experience
– Applicable to wide variety of use case such as eCommerce, eLearning, Tourism etc.

Tech Stacks:
Deep Learning Frameworks, Python

Industry:
Retail, Education