Are “do you think” embedded questions exceptions?

Question

In most embedded questions, the "Wh" portion of the question stays grouped together with the rest of the converted question (e.g. "What time is it?" becomes "Could you tell me what time it is?"), the rules seem different when "do you think" is used (e.g. "what time is it" becomes "What time do you think it is?").
Is this splitting of the embedded question with "do you think" a unique exception, or is there something I'm missing? This would be great to clarify!

Answer 1

There's something you missed. The two embedded questions are not parallel; one is a single clause, and the other (the one with think) is complex -- i.e, there are two clauses here instead of one.

Let's just look at the embedded question parts and see where they come from:

1. .. what time it is?

   • comes from What time is it? -- by Embedded Question Formation (with Subject Inversion)
   • which comes from It is [indef] time -- by Wh-Question Formation (with what for [indef])
   • which comes from The time is [indef] -- by Extraposition (leaving a Dummy It subject).

2. .. what time you think it is?

   • comes from What time do you think it is?
   • which comes from You think (that) it is [indef] time.
   • which comes from You think (that) the time is [indef].

Notice that in (2), the order is always it is, never is it. That's because the main clause that the wh-question is formed from is You think ..., rather than It is ... That's why the ordinary (not embedded) wh-question would need Do-Support, which of course is not the case in embedded questions.

So nothing has been "split". Syntax doesn't work that way. The rule has been applied to the main clause (with verb think) of a complex sentence, while the wh-word has been moved from the object complement clause (with verb is) of think.