As the title says, in ‘prefer for John to stay’, is ‘for’ a complementizer and the following is a CP, or a preposition?
there is some confusion in the other answers to this question. let me be clear: on any understanding of the term "complementizer," the word for is indeed a complementizer in the context you give. for heads a non-finite CP. allow me to give a sense of the sheer enormity of evidence pointing in this direction.
the main thrust of the evidence is in establishing a clear parallelism between for-clauses and other sorts of clauses headed by complementizers, especially that-clauses. the parallelism serves to show that for-clauses can be described as standard CPs, as shown below. note that infinitival to is a non-finite auxiliary verb, another fact for which there is a preponderance of evidence (see Bob Levine’s paper “auxiliaries – to’s company”).
as mentioned, the CP headed by that is finite (tensed) while the CP headed by for is non-finite (not tensed). we can take this to be a selectional difference between the two Cs; that selects for a finite TP complement, while for selects for a non-finite TP complement. see the appendix at the end of this answer for more details on selection.
now, on to the evidence for the parallelism. i’ll mainly cite arguments from bresnan 1972 and huddleston & pullum 2002, with points 1-5 coming from bresnan and points 6-8 coming from huddleston & pullum (with some overlap). point 9 is surely not original to me, but i don’t know of a source for it off the top of my head.
1. for fails to co-occur with other complementizers like that and whether.
bresnan notes that two members of the set of complementizers cannot appear in the same clause together. this suggests they serve the same role in the clause (i.e. they are all complementizers). the example below illustrates with that and for.
a. *Marie asked that for us to leave her alone.
b. *For that he refuses would not be surprising.
2. for occupies the same position in the clause as other complementizers (i.e. complementizers are clause-initial).
this is noted by bresnan 1972 p32-33, as well as huddleston & pullum 2002 p1183
a. it is important [CLAUSE that detailed records be kept].
b. it is important [CLAUSE for detailed records to be kept]
(huddleston & pullum p1183)
3. for occurs in the same paradigms as that, just with non-finite clauses instead of finite clauses.
The following examples are from Bresnan p37, where she notes that for-clauses can have the same sorts of syntactic roles that that-clauses can. similar observations can be found in huddleston & pullum 2002 p1182.
(3) Object Complementation
a. I know [OBJECT that he is wise].
b. I prefer [OBJECT for you to speak English].
(4) Subject Complementation
a. [SUBJECTThat he was alone was obvious from the report.
b. [SUBJECT For you to leave right now] would be inconvenient.
(5) Nominal Complementation
a. The idea [COMP-N that nobody will survive] is apalling.
b. The command [COMP-N for all troops to move out] was given Friday.
(6) Copular Complementation
a. The problem is [COMP-COPULA that you are arrogant].
b. The command is [COMP-COPULA for all troops to move out].
4. Non-finite clauses headed by for can be extraposed in just the same was as finite clauses headed by that.
bresnan observes that, just like finite that-clauses, for-clauses can be extraposed to sentence-final position. when this happens, an expletive it occupies the clauses' original position in the main sentence. the example below is bresnan's, with extraposition taking place from subject position. thus, expletive it occupies subject position in the main sentence.
a. It should not be as important as you make it sound [that they agree with us].
b. It should not be as important as you make it sound [for them to agree with us].
5. for, just like that, is sometimes optional when it heads a complement clause, but is always obligatory when it heads a subject clause.
a. I don’t think [ (that) you should have said it].
b. I want [ (for) you to help me].
a. [ *(That) he was alone] was obvious from the report.
b. [ *(For) you to leave right now] would be inconvenient.
(bresnan p.40. note the notation is intended to show that that and for are obligatory here)
6. for-clauses can have expletive subjects
huddleston & pullum’s basic argument is that the for-phrase is a clause, just like any other, save that it is non-finite. they emphasize that the subject of the for-clause acts just as we would expect the subject of any clause to. the subject can be an expletive there, for instance, something that is typically reserved for subject position.
a. It is essential for there to be no misunderstanding on this point.
b. It is essential that there should be no misunderstanding on this point
(the a example is from huddleston & pullum p1183. i added the b example for comparison)
now it might be worth noting that expletive there can raise to object, even being able to raise to object of preposition, for instance in the following examples.
a. You can always count on there to be rain.
b. I can almost always rely on there to be a group of students playing Ultimate Frisbee every day at 3:30 in the afternoon
(b example from culicover & jackendoff 2005 p210)
however, there is a huge range of properties with respect to which for-phrases diverge from the on cases in (11), such that a similar raising analysis of there is not possible with for. just to give one example, on plus the subject of the infinitival clause forms a constituent, which can be fronted. this is not possible with for:
a. [On whom]i can you always count __i to do a good job?
b. *[For whom]i do you want __i to win?
this points towards an analysis of the on cases as involving a true PP, such that the preposition plus infinitival subject form a constituent. this is not a possible analysis with for clauses.
7. for-clauses can have idiom subjects
huddleston & pullum (p1183) further point out that the subject of a for-clause can be part of an idiom, something that’s also true of the subject of a that-clause.
a. he called for close tabs to be kept on the new recruits
b. he hoped that close tabs would be kept on the new recruits
(a example from huddleston & pullum p1183. i added the b example for comparison)
8. placement of adverbial particles
huddleston & pullum (p1183) also note that the placement of adverbial particles like both in for-clauses mirrors the placement of such particles in that-clauses. this suggests that the subject of a for-clause occupies the same syntactic position as the subject of a that-clause.
a. it’s necessary for both your parents to sign the form.
b. it’s necessary for the form to be signed by both your parents.
c. It’s necessary for your parents both to sign the form.
(huddleston & pullum 1183)
compare with the placement of both in a that-clause:
a. it’s likely that both your parents will sign the form.
b. it’s likely that the form will be signed by both your parents.
c. It’s likely that your parents both will sign the form.
9. extraction of the subject of a for-clause yields a complementizer-trace effect, just as extraction of the subject of a finite clause would
because many people are not familiar with the complementizer-trace phenomenon, i will explain a bit more here. one of the most-discussed phenomena in syntactic theory is the “comp-trace” effect, wherein the subject of an embedded clause cannot be extracted when that clause is headed by an overt complementizer. so for instance the canonical illustration would be:
a. Whoi did you say __i won?
b. *Whoi did you say that __i won?
notice that the complementizer can be overt with object extraction, just not with subject extraction.
Whoi did you say (that) you like __i?
this is sometimes called the “that-trace” effect, but “comp-trace” is more accurate, as it applies with all complementizers, for example with if and whether, as well. since if and whether aren’t in free variation with null realizations, extraction of the subject of an if or whether clause is blocked.
a. *Whoi did you ask if __i won?
a. ?Whoi did you ask if I like __i?
a. *Whoi do you wonder whether __i won?
a. ?Whoi do you wonder whether I like __i?
For-clauses participate exactly as we would expect if they were structurally parallel to other clauses headed by complementizers. When for is overt, extraction of the subject is blocked. When for is null, extraction is allowed.
a. Whoi do you want __i to win?
b. *Whoi do you want for __i to win?
For is optional when the object is extracted, just as with the other complementizers.
Whoi do you want (for) John to like __i?
thus for acts just as a complementizer would be expected to with respect to the comp-trace effect.
on two differences between for clauses and other CPs
there are sure to be more reasons than the ones cited above. these are just the ones that came to my mind without too much digging.
now, it might be worth noting two ways in which for-clauses differ from other sorts of CPs in english. first, for-clauses are non-finite. this is no reason to think that for is not a complementizer. there is no reason to stipulate that complementizers head finite clauses only, especially given that non-finite complementizers are by no means exclusive to english; they appear in a variety of languages, as a google scholar search for “non-finite complementizer” easily reveals.
second, the subject of a for-clause is marked with accusative case rather than nominative case (as would normally be the case in english). historically, the reason is that for-clauses are derived from prepositional phrases, which in english assign accusative case (or “objective” case of “oblique” case or whatever term one prefers). syntactically, the lack of nominative case has been related to non-finiteness. it is widely believed that nominative case is associated with tense. since for-clauses are non-finite, lacking tense, they can’t assign nominative case.
as with any phenomenon in syntax, really careful investigation reveals many interesting puzzles. if we look really closely, and examine a wide range of complex syntactic properties, we'll certainly find some further ways in which for-clauses differ from that-clauses. but our first response should not be to say that for is not a complementizer, or that it is some sort of exceptional construction. we should construct theories that capture whatever differences we find, while also capturing the deep parallelisms that exist between for-clauses and other clause types.
Bresnan, J. W. (1972). Theory of complementation in English syntax. MIT.
Culicover, P. W., Culicover, P. W., Jackendoff, R. S., & Jackendoff, R. (2005). Simpler syntax. Oxford University Press.
Huddleston, R. D. & Pullum, G. P. (2002). The Cambridge grammar of the English language. Cambridge University Press.
Levine, R. D. (2012). Auxiliaries: To's company. Journal of Linguistics, 48(1), 187-203.
appendix: some basic background on selection
a commenter insists that the fact that for-clauses are non-finite disqualifies them from being CPs. however, as i pointed out, non-finite CPs are well-attested crosslinguistically, so there is no comparative reason for skepticism towards the analysis of for-phrases as non-finite CPs. in addition, as shown in great detail above, all evidence points towards for-phrases being structurally parallel to other English CPs, including that-phrases. there's no reason to think that CPs must be finite. thus, the idea that for-phrases are non-finite CPs is perfectly plausible.
that being the case, it's not entirely clear what the commenter has in mind. however, i think some light can be shed on the objection with the following comment, posted on a different answer:
"In the phrase structure tree, the template is predetermined. Words are inserted into positions in the template. In this case, inserting for into the C position incorrectly predicts that the finite complement should be possible, since it is possible for the other Cs"
here the assumption seems to be that phrase structure representations are templates into which words are freely inserted. the given trees then make the wrong prediction, because they predict C can be filled by any C, T can be filled with any T, etc. Under this assumption, there's an incorrect prediction, because in fact T can't be just any T: with that in C, T has to be filled by a finite T, and with for in C, T has to be filled by a non-finite T.
to clarify, i do not assume that phrase structure representations are templates into which words are freely inserted. instead, it is known that syntactic representations are constrained by selection relations. we will say that for selects (or "subcategorizes") a non-finite TP as its complement, while that selects (or "subcategorizes") a finite TP as its complement.
it might be helpful to compare this instance of selection to other instances. a straightforward case occurs with determiners. (i will assume for accessibility's sake that N is the head of nominal phrases, not D. the argument works the exact same if one assumes D is the head). notice that the two noun phrases below are considered to have the same structure.
a. that dog
b. those dogs
in one case, we get a singular determiner that, and in the other, a plural determiner those. further, we can't switch the determiners, getting "those dog" or "that dogs." this is straightforwardly described as an instance of selection. we will say that singular nouns select singular determiners, not plural determiners. we will say that plural nouns select plural determiners, not singular determiners. notice that we don't want to say that singular nouns and plural nouns are different categories (they are both nouns), nor do we want to say singular determiners are a different category from plural determiners (they are both determiners).
we will use the same apparatus to describe the fact that for must have a non-finite clause as its complement, and that that must have a finite clause as its complement.
I personally would say that it is not a complementizer.
For instance, if we compare the sentences:
(1) Mark prefers for John to stay
(2) John prefers to stay
I personally want to think of (1) and (2) as having the complement clause to be about the same, because they are both "(for John) to stay" but (2) has the John implied because it is correferential with the subject of "prefer." Because the "for" part is not necessary to make it a relative, and "to" is already a complementizer that gives the clause the complement status, you can safely say that "for" is not likely to be a complementizer.
That "for" is a complementizer. If it were a preposition, it would take an object which could be pronominalized with "it" or "that", but *"John won't stay though I'd prefer for it". On the other hand, when it is a complementizer, pronominalization of the entire nominalized construction may be possible: "John won't stay, though I'd prefer that."
Generally, the structure depends on the governing verb -- here "prefer". Compare this with "wait". "John won't stay, so don't wait for that" as compared with *"John won't stay, so don't wait that". Here, the preposition interpretation is at least possible.
The word for in the sentence I prefer for John to stay is neither clearly a preposition nor clearly a subordinator (aka complementizer), but rather it has a unique status. To help capture this unique status, one could coin a new term. One could call it an adpositional subordinator because it behaves as a preposition with respect to its nominal complement and like a subordinator with respect to the to-infinitive complement.
Let’s consider first some evidence suggesting that for in I prefer for John to stay is not a pure subordinator. Pure subordinators can take a finite clause as their complement, yet for cannot do this, e.g.
(1) *I’d prefer for John stay(s).
(2) *I’d prefer for that John stay(s).
If for were a subordinator like other subordinators, these sentences would be good. Since they are clearly bad, we can indeed conclude that for is not a pure subordinator.
Let’s consider next some evidence demonstrating that for in I prefer for John to stay is not a pure preposition. Most of the time, pure prepositions take a nominal complement alone; they do not take a to-infinitive complement as well, e.g.
(3) *On Tuesday to go home, we left early.
(4) *With Tom to leave early, I have spoken.
(5) *In two hours to go home, we are leaving.
The prepositions and their complements are fronted in these examples in order to avoid the confounding factor of the to-phrases being interpreted as rationale phrases. If the prepositions on, with, and in could take a to-infinitive complement like for can, these sentences would likely be acceptable.
The insights so far demonstrate that for should indeed be granted a special status in the grammar. It has special properties that cannot be captured by the simple designation subordinator or preposition alone. This situation is a problem for standard phrase structure approaches because such approaches are not well-equipped to address lexical idiosyncrasy. Thus, the approaches that BillJ, Tsutsu, Jlawler, Greg Lee, and matan-matika envision above cannot address the status of for in I’d prefer for John to stay in a convincing way, for they lack an appropriate category that captures its distribution. It is not the head of CP, IP, TP, etc. because these categories are all compatible with finite clauses, whereas for does not license such clauses, as demonstrated above with examples (1-2). The category PP is also not appropriate because the P of PP does not take a to-infinitive complement. Indeed, what is a phrase structure grammarian to do?
A dependency grammar (DG) approach is not faced with these difficulties because it conceives of sentence structure in terms of directed links, i.e. dependencies, between words. There is hence no need to force for into the one or the other preconceived category, for it is possible to produce a structural analysis without assigning for to any specific category at all. The DG analysis of the example sentence that I prefer is next:
This analysis is particularly strong insofar as both John and to stay are immediate dependents of for. The status of for as subordinator-like is accommodated in that the whole constituent for John to stay conveys the meaning of a proposition, i.e. of a complete clause, and the status of for as preposition-like is also accommodated insofar as it takes the nominal John as its immediate dependent similar to how a normal preposition takes a nominal as its immediate dependent.
Another important strength of the analysis is that the string John to stay is not a constituent, that is, it is not a complete subtree. This matches the fact that most tests for constituents fail to identify John to stay as a constituent, such as pronominalization, e.g. * I prefer for that/it. Note that the phrase structure analyses that BillJ, Tsutsu, matan-matika and others advocate for will likely construe John to stay as a constituent, contrary to what most tests for constituents actually reveal.
In sum, the question is good because it helps reveal a problem facing phrase structure syntax in general. The phrase structure template of syntactic categories (CP, TP, IP, PP, etc.) is hardly capable of addressing lexical idiosyncrasy. Dependency syntax presents a plausible alternative in this regard, one that has no problem with lexical idiosyncrasy.