In the first stage of this 2-stage model we retrieve relevant passages from a text collection, usually using a search engines of the type we saw in the previous section. In the second stage, a neural reading comprehension algorithm passes over each passage and finds spans that are likely to answer the question.