Deep Latent Variable Models for Natural Language Processing

This dissertation is organized into a main text, consisting of a background section and chapters summarizing and discussing the contributions, and an appendix collecting the the four papers included in this thesis. The main body of this dissertation consists of four parts. The first part, Chapter 2, establishes the context underlying the remaining of the dissertation. In the second part, Chapter 3, we provide an overview of the three main latent variable models studied in the included papers. In the third part, Chapters 4 and 5, we revisit the contributions made in variational inference and optimization (Papers B and D). In the fourth and last part, Chapters 6 and 7, we explore the task of medical question answering based on two methods (Papers C and B).