Statistics for Wasserstein Autoencoders with Mixture of Gaussian Priors for Stylized Text Generation