We need to select questions that meet a certain level of difficulty from the SQuAD database to accurately assess reading comprehension from difficult concepts. These criteria include:

• Syntactic variation: word structure of a sentence in the text is different from how it is phrased in the question (eg. Question: What Shakespeare scholar is currently on the faculty? Phrase from the text: Current faculty include...)
• Multiple sentence reasoning: multiple sentences required from text to answer a question, not just a single sentence

The SQuAD paper analyzes other question types, but for purposes of our research we only need to meet these criteria