> Can you explain the difficulty in engineering a search apparatus over a corpus of text data?

My guess: "If we ever imply that's possible, people might start asking questions about all the other work we've ripped off, so the official answer is that it's impossible".