I noticed that in English, it is incorrect to say "Who did eat the apples?" but it is correct to say "who ate the apples?" It would be very helpful if you can give me some clues based on syntax. Thanks very much!
I don't know how much background in syntax you (or later readers) have, so I'm going to start with some basics (the complementizer layer, do-support, and object Wh-questions) before going into subject Wh-questions. Also, fair warning: syntax is not really my field, so things may be incomplete or incorrect, and anyone is invited to add or correct things.
The complementizer layer for yes/no questions
This is based on Carnie (2013, §7.3).
A sentence like You have seen Louis is analyzed with have in T(ense) and you moved from Spec,VP to Spec,TP, the default for subjects:
To account for the word order in the corresponding yes-no question, Have you seen Louis?, it is argued that have moves from T to a higher layer C:
The reason behind this is that the C head has a [+Q] ("question") feature, which needs to be phonologically realized, i.e. it must contain something / cannot be null. Support for this comes from languages like Irish, which have an overt [+Q] complementizer (rather than T-to-C movement), Ar:
(1) Ar thit Seán?
Q fall John
"Did John fall?"
Further support comes from the fact that T-to-C is blocked when there is an overt complementizer, which is the case in embedded questions in English:
(2) a. Fabio asked if Claus had run a marathon. (overt C; no T-to-C)
b. *Fabio asked if had Claus run a marathon. (overt C and T-to-C)
c. *Fabio asked had if Claus run a marathon. (T-to-C and overt C)
This is based on Carnie (2013, §10.2–3).
This poses a problem for sentences without auxiliary verb. The T head cannot move into C to give it phonological content, because it is null itself! As a last resort, a dummy do is inserted in T so that it can move to C:
(3) a. You eat apples.
b. Do you eat apples?
This is based on Carnie (2013, §12.1).
We now come to Wh-questions. We first need to look at questions where the Wh-element is the object. In this case, the object moves from its position in the VP layer to Spec,CP:
Recall that above we argued that T-to-C was triggered by a [+Q] feature on the C head. We now say that there is a [+Wh] feature on the C head as well. The Wh-phrase whom moves to Spec,CP to check this [+Wh] feature.
(The existence of the [+Wh] feature is again supported by Irish, which has at least three complementizers in main clauses: go [-Q, -Wh]; an [+Q, -Wh]; aL [+Q, +Wh].)
Note that do-support is still needed because whom cannot check [+Q].
What you observe is called the T-to-C asymmetry (Koopman 1983): we do have T-to-C in object Wh-questions (and hence require do-support), but we do not have T-to-C in subject Wh-questions:
(4) a. What did Mary buy?
b. *What Mary bought?
c. *Who did buy the book?
d. Who bought the book?
(Example (4c) is possible when did is focused, but this is not a case of do-support because it is not required for [+Q] feature checking. It also occurs in Mary did buy the book!)
Why this happens is a bit of an open question, so good job on noticing the asymmetry.
Below are two possible solutions. There may be more, this is not really my field.
Pesetsky & Torrego (2001): nominative DPs have a tense feature
Pesetsky & Torrego (2001) argue that in main interrogative clauses, C bears an uninterpretable T feature uT. This means that something must be done to delete this feature. This is done by moving T to C. This is as before, but the terminology has been updated: we now talk about a uT feature that must be deleted instead of a [+Q] feature that must be checked.
Next, they propose that nominative DPs have a uT feature on the D head. This is the key proposal of their paper.
Within this framework, when a feature is deleted it remains somehow available, and it can itself delete features of the same type. So once the subject has moved into Spec,TP, it bears a deleted feature
uT. By moving it further into Spec,CP, it can delete uT on C. When this is done, there is no need any more to also move T to C.
A question that may arise is: why in this situation must Spec,TP move to Spec,CP to delete uT on C; why can T not move to C instead? This would give Did who buy the book?, which is clearly ungrammatical. The answer here is twofold. First, Spec,TP really must move to Spec,CP because we still have the [+Wh] feature to check (or, rather, the uWh feature to delete). Second, by assumption, we only allow as many movements as are necessary to delete all uninterpretable features. After Wh-movement there is no uT feature left on C, so there is no need for T-to-C.
Of course one needs to explain why a nominative DP has a uT feature. Pesetsky and Torrego argue that this is not entirely unexpected because the reverse is true: T holds uD features like person and number for which it needs a subject DP.
Adger (2003): uclause-type on T
Adger (2003, §9.3–4) gives a slightly different account for do-support, from which the lack of do-support in subject Wh-questions follows.
In his theory, T has a uninterpreted uclause-type feature that can be deleted by C[+Q]. This explains T-to-C movement in yes/no questions and object Wh-questions. As above, the requirement of T-to-C leads to do-support.
In subject Wh-questions, there is at some point a [+Wh] DP in Spec,TP. Adger suggests that the [+Wh] feature in this position deletes uclause-type on T. Afterwards, the subject goes on to Spec,CP to delete uWh over there.
The difference is that Pesetsky and Torrego have a uT feature on C that is deleted by T-to-C or Wh-movement into Spec,CP, whereas Adger has a uclause-type feature on T that is deleted by T-to-C or Spec,TP.
- Adger, David. 2003. Core Syntax: A Minimalist Approach. Oxford University Press.
- Carnie, Andrew. 2013. Syntax: A Generative Introduction. 3rd ed. Wiley-Blackwell.
- Koopman, Hilda. 1983. ECP effects in main clauses. Linguistic Inquiry 14: 346-350.
- Pesetsky, David, and Esther Torrego. 2001. T-to-C Movement: Causes and Consequences. In Michael Kenstowicz (ed.), Ken Hale: A Life in Language. MIT Press.