Handle questions formed as .. you know .. questions!

[SeanTater]

The LAT detection module should be able to recognize more of these question forms:

• Who (the LAT should probably be "person")
• What (the LAT should be connected to "what" in the dependency graph)
• How (much|many|any adjective) is/are/was/were ___
  • Much and many need a number
  • The others vary. I hate the idea of hard coding many of them.
  • Rather than using LAT for these, can we use the similarity of dependency linkage?
  • We need worked examples to see if this is practical
• When
  • Need to make sure it's actually a question word (not linking phrases)
  • This may be available in CoreNLP already but if not we may be able to tag answers as being "when" using a regex

[SeanTater]

These are not found in the Jeopardy questions. They're actually from the TREC QA dataset, which is available in another project: https://github.com/brmson/yodaqa/blob/master/data/trec/trecnew-curated.tsv

[SeanTater]

"Which" and "what" are reasonably easy, I adjusted ClueType to match.