In questions where a wh-element refers to the object, we can observe SAI (Subject-Auxilliary inversion).

[Who did [you see]]?

As far as I'm aware, C-head has a [+Q] feature and it's occupied by a null bound morpheme that needs a verb to attach to. Since V-to-C is impossible (for the previous step, V-to-I is illicit as well) in English, only auxiliaries have the power to raise to C.

But why is it so that wh-elements pertaining to the subject of the clause do not trigger SAI? The C-head still has the [+Q] feature and do-support is available in English if there is no other auxilliary present.

[Who [read this book]]?

  • 5
    A wh-question requires a wh-expression at the beginning and a verb right after it. When the subject is a wh-expression, these requirements are already met, so no further adjustment is required.
    – Greg Lee
    Apr 12, 2018 at 8:52
  • Right. it matches the parse signal and that's enough. The other peculiarity of a subject wh-expression is that it can't deleted, unlike most fronted relative pronouns. That's also because it's the subject and the clause is tensed; tensed clauses require actual subjects -- no Equi.
    – jlawler
    Aug 8, 2018 at 2:45
  • I don't know if that can help you, but if you are interested by comparison with other languages, here an example in Riffian where such inversion is possible : i-cci-t utares = he-ate-it man (the man ate it) >>> wi t-icc-in = who it-eat-PART (who ate it)
    – amegnunsen
    Oct 7, 2018 at 10:16

2 Answers 2


Now, as I'm reading my question, it seems that there's simply no "space" to do such an inversion. "Who" is in [spec, CP] and auxiliary after inversion occupies the C-head after head-to-head movement. Sure, according to Rizzi's proposal there are other functional projections above CP, but it'd seem too convoluted to even resort to that.

So, in conclusion, for SAI to take place, either the subject (wh-element) would have to stay in its EPP position, or the auxiliary to move to an unnatural (for him) [spec, CP] position. Given it's a head, such movement would be illicit.

[Who [read this book]]?

enter image description here

  • 1
    If you ignore the 'dead wood' in that tree, you will see that the subject doesn't move at all. Also, if there was real Subject-Aux inversion with a Q-subject then you would have "Did who read this book?", and that is mixing a question of noun with a question of truth.
    – amI
    Dec 14, 2017 at 20:53
  • What does 'dead wood' mean?
    – TheTobruk
    Dec 15, 2017 at 18:10
  • 2
    I meant those branches that have no leaves. I doubt that the brain stores anything that isn't needed, and X-bar theory may go too far in trying to fit every possibility onto a single tree pattern.
    – amI
    Dec 18, 2017 at 20:32
  • 1
    there is in fact strong evidence that the subject does move in subject wh questions, despite the surface order being preserved. the standard analysis of SAI is that it involves raising of A. since the subject also moves in subject wh-questions, the original order is restored. it remains a puzzle why you don't get "who did read this book", plausibly a phonological phenomenon. Jun 12, 2020 at 11:51
  • The usual scientific term is for theoretically required but practically irrelevant pieces of an analysis is "epicycles"; "dead wood" refers to unproductive faculty members.
    – jlawler
    Feb 7, 2021 at 19:45

The question and the answer offered by the OP overlook the key insight concerning inversion and subjects in English. This insight is expressed in the comments produced by Greg Lee, John Lawler, and aml. The reason subject-auxiliary inversion does not occur in matrix clauses when the subject is questioned is that the canonical position of the subject is to the left of the finite verb, whereas the canonical positions of all other sentence participants are to the right of the finite verb. Thus, inversion in matrix questions is necessary when something other than the subject appears in the initial position to the left of the finite verb, since inversion then helps identify that participant as a non-subject.

This explanation is consistent with an approach to syntax that grants the online production and processing of sentences a decisive role. Syntactic structures are produced and processed in time from earlier to later. If the left-most wh-expression in a question in English is not the subject, it must be marked as such by inversion, and this inversion then aids the processing of the sentence.

The explanation is independent of the particular theoretical apparatus that one prefers. The Government and Binding (GB) (or Minimalist Program (MP)) apparatus assumed in the question and the OP's answer is at best a tangential matter. One need not acknowledge various X-bar theoretic constructs (e.g. EPP, CP, V-to-C, etc.) for the explanation to work. I would even argue that such constructs result in opaque and obtuse linguistic reasoning.

  • but inversion doesn't happen in embedded questions, whether the subject or anything else moves. "i wonder who he likes"/"i wonder who likes him." your explanation in terms of processing, the idea that inversion is necessary to identify fronted constituents as non-subjects, just doesn't work. May 12, 2020 at 2:35
  • @one-off-post, It works for matrix interrogative clauses. The order is (X)SV in embedded interrogative clauses. The distinct patterns distinguish between direct vs. indirect questions, which is important for processing the content of any question in terms of the relevant speech act. What will he say? is certainly processed distinctly from what he will say. May 12, 2020 at 3:13
  • inversion or lack thereof is not tied to the type of speech act - some dialects have inversion in embedded clauses, some lack inversion in matrix clauses. and anyway, why would this pragmatic difference negate the processing advantage you proposed? it applies equally to embedded clauses. the problem remains. op's question doesn't have some common sense answer like the one you suggested. May 12, 2020 at 4:10
  • @one-off-post, Hmmm. I do not understand what you are asking. Yes, the processing of language applies to all clauses, matrix and embedded. I am claiming that, for instance, the clause Will Fred stay? expresses a basic speech act, namely a question, whereas whether Fred stays does not, i.e. it is not question in the sense of a speech act. If you disagree, please illustrate your point with an example, so I can better understand what you mean. May 12, 2020 at 7:36
  • you said inversion is necessary to identify when something in initial position is not the subject, because initial position is the canonical position of the subject. i understand this to predict: if a non-subject appears at the beginning of its clause, this must be marked by inversion. the prediction is not borne out, because when non-subjects appear in fronted position in embedded questions, inversion does not occur. for example, "i wonder who he likes?" where in the embedded clause, the clause-initial constituent is "who," the object, not the subject. May 12, 2020 at 10:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.