This is my PhD thesis on the use of different semantic models in question answering tasks.
It begins with a description of my question answering framework and its architecture.
Then I describe large-scale experiments on Yahoo Answers datasets that I carried out during my internship at Yahoo Labs in Barcelona. I used a combination of learning-to-rank algorithms and a large number of hand-crafted features, and show how distributional semantic models can replace most of them.
Finally I describe how I built an agent that plays "Who Wants to Be a Millionaire?" using machine learning and natural language processing techniques.