Generating Semantic Sentences

Given the growing complexity of tasks and data in NLP there are limited unsupervised learning methods to tackle the problems. In this paper we will look into novel approach to one of the problems as representing sentences in latent space using Recurrent Neural Network (RNN) and Variational Auto Encoder. Combining VAE-LSTM approach we will repharase (and generate similar and meaningful) sentences from given sentence. An our assumption same architecture can also be applied to language modeling problem. We will give results on using single layer encoder as well as 2 layer encoder for our VAE.