As noted in the introduction, this particular problem surfaced at MIT when doctoral student John Ross brought it to the fore. I'll illustrate the problem with this example which Howard Lasnik provides in his paper, "When Can You Repair an Island by Destroying It?":
(1) *That he'll hire someone is possible, but I won't divulge who that he'll hire is possible.
Notice that if you want to make (1) grammatical you can delete the words after "who" and you get (2).
(2) That he'll hire someone is possible, but I won't divulge who.
Those words which are deleted are called "the island". You can read Lasnik's paper to get an idea of some traditional generative approaches to problems such as the above. Other papers which you could read which would offer other arguments using generative theory, by such generativists as Lasnik, Danny Fox, Jeroen van Craenenbroeck, Marcel den Dikken and Jason Merchant, are listed at the end of this article.
All of the examples in this article were borrowed from articles written by the above-named generativists, and I will credit the examples as I go along. I would like to start with Lasnik's "When Can You Repair an Island by Destroying It?" since these examples strike me as particularly interesting. The first example pair - (1) and (2) above, repeated here and modified slightly - comes from this work:
(1) *That The Donald will hire someone is possible, but I won't divulge who that he'll hire is possible.
(2) That The Donald will hire someone is possible, but I won't divulge who.
So the question everyone is trying to answer is Why is it that ending the sentence with "who" leaves (2) grammatical, but leaving on the words "that he'll hire is possible" messes things up?
I'll start with a very simple approach. In the second part of (1): *That The Donald will hire someone is possible, but I won't divulge who that he'll hire is possible.
Let me do one thing in order to simplify matters. Let's take that fronted complementizer in the first part of (1) and let's revert it to a form which is grammatically equivalent:
(2): *It is possible that The Donald will hire someone , but I won't divulge who that he'll hire is possible.
In the second part of (2), "who" is seen as a pronoun and "that" as a relative pronoun, since it "relates" something to "who". One test to show that "that" is a relative pronoun is to show that, in a relative clause construction, the "who that" sequence can be expanded to "who IT IS THAT":
(3a) but I won't divulge who IT IS THAT he'll hire.
The "IT IS THAT" structure can exist partially:
(3b) but I won't divulge who IT IS he'll hire.
or not at all:
(3c) but I won't divulge who he'll hire.
but it can't exist with the relative pronoun alone:
(3d) *but I won't divulge who THAT he'll hire.
(3a-c) are, of course, grammatical. Why isn't (3d) grammatical? Simply because "hire" doesn't need two direct objects, "who" and "that", right next to each other. One will suffice; hence (3c). And this would be a very good time to point out that the "who" in (3a-c) is actually "whom", since "hire" obviously takes an object:
(3a) but I won't divulge who(m) IT IS THAT he'll hire.
(3b) but I won't divulge who(m) IT IS he'll hire.
(3c) but I won't divulge who(m) he'll hire.
(Note: the case of "who(m)" is clearly not determined by "divulge". In the following sentence, we see that "who", not "who'm" must follow "divulge":
but I won't divulge who(*m) IT IS THAT hired Fred.
So what, then, is the object of DIVULGE? Well, actually, in (3a) it's everything that follows DIVULGE; it is the person represented by
who(m) IT IS THAT he'll hire.)
Now, suppose we employ the allowable versions of IT IS THAT and then add "is possible" to the structures in (3a-c):
(3e) but I won't divulge who(m) IT IS THAT he'll hire is possible.
(3f) but I won't divulge who(m) IT IS he'll hire is possible.
(3e) but I won't divulge who(m) he'll hire is possible.
Why aren't any of these grammatical? It's clear now, isn't it? "Who(m)" can't be both the object of "hire" and the subject of "is." And, mark you, "is" does need a subject. So we are checkmated. We must give IS it's own subject, and that is EXPLETIVE IT.
but I won't divulge who(m) he'll hire IT is possible.
Can we just leave well enough alone? No, we can't. "IT is possible" is a clause which can qualify an event, and can be followed by a complementizer THAT. Well, THAT can't just sit out there at the end of the sentence. That's no way to complementize. It needs to be inserted before the complmentary clause:
but I won't divulge who(m) IT is possible THAT he'll hire.
But wait. What about this, where WHO is the subject of HIRE, not the object:
but I won't divulge who(*m) IT IS THAT will hire Fred.
Let's tack "is possible" onto the end of that one:
but I won't divulge who(*m) IT IS THAT hired Fred is possible.
Can't WHO be both the subject of HIRED and the subject of IS? I believe it can. That means that we're not checkmated here the same way we were with
(3e) but I won't divulge who(m) he'll hire is possible.
Consider: since there's no problem dropping a relative clause:
but I won't divulge who(*m)... is possible.
we are left with a person (WHO) being POSSIBLE. But is WHO what is possible, or is it an event which is possible? Clearly, it is the event of someone's being hired which is possible, and so we come once again to the same conclusion: the subject of IS POSSIBLE is not WHO, but EXPLETIVE IT.
but I won't divulge who(*m) IT IS POSSIBLE THAT hired Fred .
But wait!!! That's not right either. Why is it we can say:
but I won't divulge who(m) IT is possible THAT he'll hire.
but not the sentence just above it? Have we been checkmated again? No, not at all. We have just run into the Anti-Complementizer Effect. As I have shown in other writings, a complementizer cannot be followed by a sequence which cannot stand alone as a complete sentence. Thus,
THAT he'll hire is fine: He'll hire is a stand-alone.
THAT hired Fred is no good. Hired Fred cannot stand alone.
So how do we make this grammatical?
but I won't divulge who(*m) IT IS POSSIBLE THAT hired Fred .
Simple. We do this:
but I won't divulge who(*m) it is possible THAT IT WAS THAT hired Fred .
The first THAT is a comlementizer.
The complementizer is followed by IT WAS, which is a stand-alone.
The second THAT is a relative pronoun, which relates to who(*m).
What "is possible" is not a "who" but "an act of hiring"; in other words, a "what". The problem with (1) is that the two parts of the sentence are talking about two different entities or concepts, two concepts which are - for lack of a better term - "topically non-parallel". The first part talks about the possibility of an act of hiring; the second tries to talk about "who is being hired" - but the addition of "is possible" makes that part of the sentence ungrammatical. To make the second part of the sentence talk about "the act of hiring", then we need to place the clause "it is possible" just before "that he'll hire":
(3) It is possible that The Donald will hire someone, but I won't divulge who it is possible that he'll hire.
Doing so changes "that" from a relative pronoun to a complementizer, the same as in the first part of the sentence.
Anyone who has ever applied for a job or watched "The Apprentice" knows that applying for a job and getting it are not the same. There is only one way to repair this island, and that is to make sure that both parts of the sentence are talking about the same thing:
(3) That The Donald will hire someone is possible, but I won't divulge who it is possible that he'll hire.
That seems pretty good to me. But how do we know that both parts of the sentence are talking about the same entity, i.e. "the person that it is possible that The Donald will hire."? Perhaps we could see things better if we changed
(4) That The Donald will hire someone is possible...
to this
(5) It is possible that The Donald will hire someone...
Those two complex structures mean exactly the same thing, though structured differently. I call them "grammatically equivalent". And so if we substitute (5) for (4) we get (6):
(6) It is possible that The Donald will hire someone, but I won't divulge who it is possible that he'll hire.
Now that we have substituted (5) for the grammatically equivalent (4), the two parts of Sentence (6) are clearly topically parallel: both clauses refer to a person who IT IS POSSIBLE that The Donald will hire. The two parts of Sentence (3) are also topically parallel:
(3) That The Donald will hire someone is possible, but I won't divulge who it is possible that he'll hire.
and you will notice that "That The Donald will hire" in the first part of (3) is structurally the same as "that he'll hire" in the second part. That said, it is perfectly OK to leave off the last part of this sentence:
(6) It is possible that The Donald will hire someone, but I won't divulge who (it is possible that he'll hire.)
And since we've already established the fact that the first clause in (6) is grammatically equivalent to
(4) That The Donald will hire someone is possible...
then now we know why this is good:
(2) That he'll hire someone is possible, but I won't divulge who (it is possible that he'll hire.) .
and why this is not:
(1) *That he'll hire someone is possible, but I won't divulge who that he'll hire is possible.
That's the first part of the answer. Now I would like to expand on this a bit more.
First of all, for those who are looking to understand what the workings of the language tell us about the workings of the brain, I think it's entirely possible that Lasnik's problem is telling us something of interest: the brain recognizes that two clauses in a grammatical sentence don't necessarily have to be structurally parallel, as long as one of the clauses is grammatically equivalent to one which is parallel to the other. Thus, both (3) and (6) are grammatically equivalent, even though the first clauses of (3) and (6) are inversions of one another. There are two more permutations which are equivalent as well:
(7) That The Donald will hire someone is possible; that I will divulge who it is possible that he'll hire is unlikely/impossible/not going to happen. (both clauses with fronted complementizer)
(8) It is possible that The Donald will hire someone; that I will divulge who it is possible that he'll hire is unlikely/impossible/not going to happen. (second clause with fronted complementizer)
The brain recognizes all of these permutations - (3, 6, 7 and 8) - as grammatically equivalent. So I think we learned quite a bit from this first example by Lasnik. Now let’s look at another one which Lasnik presents:
(9) *We left before they started playing party games. What did you leave before they did?
“Did” can only stand for “started playing.” This gives us:
(10) *We left before they started playing party games. What did you leave before they started playing?
The “trip-up” here is that the second clause question, the one which parallels
"We left before…"
should not be "What did you leave…"
but "When did you leave…"
But we can ask “when you left” in an oblique way: by asking what was happening at that particular time since, as far as we can gather, the schedule of events at the party in question corresponded to the time of the evening. And so instead of (10), we should have:
(11) We left before they started playing party games. What were they doing when you left?
which has the implied meaning of
(12) We left before they started playing party games. When did you leave?
Both sentences in both (11) and (12) discuss time, one obliquely and the other directly.
Is there anything we can use here to tell us about how the brain works? Possibly. The brain recognizes that events progress along a schedule which corresponds to the passage of time. And so if the speaker recognizes this, then a discussion of events is really a discussion of time. But to ask what time something happened means that we must keep the linguistic focus on specific events (games, etc.) To state that someone left the party is getting away from the focus on time, and the mind recognizes the irrelevance. Now let’s consider this third Lasnik example:
(13) *They want to hire someone who speaks a Balkan language, but I don't know which Balkan language they want to hire someone who speaks.
(14) They want to hire someone who speaks a Balkan language, but I don't know which (Balkan language).
In approaching this problem, it might be best if we acknowledge right away that that the second clause can try to refer not just to a language but to a "someone". It can try to say "I don't know which speaker" or it can say "I don't know which language." Thus:
(14) They want to hire someone who speaks a Balkan language, but I don't know which (Balkan language) / (which speaker of a Balkan language).
The problem here is that “Balkan language” is not parallel to “someone”. In other words, the “someone” they want to hire is a “speaker”, not a Balkan language. And so if we substitute “speaker” after “which”, we are going in an altogether different direction from (13):
(15) They want to hire someone who speaks a Balkan language, but I don't know which speaker of a Balkan language they want to hire.
On the other hand, if we want to leave “Balkan language” after “which”, then we can do this:
(16) They want to hire someone who speaks a Balkan language, but I don't know which Balkan language they want to hire a speaker of.
or: (17) They want to hire someone who speaks a Balkan language, but I don't know of which Balkan language they want to hire a speaker.
So what have we learned here? One thing we all learn at some point is that a speaker and his language are inextricably connected. Is this something that the brain just "knows"? More likely we see people speaking various languages and we make this connection. Thus, finding out who the speaker is tells you what the language(s) is (are). And so to understand how to formulate a sentence of this complexity we must first understand at least that much about how the world works.
Now, suppose you just want to find a “speaker of a Balkan language”, but you don’t care which of the Balkan languages he speaks. Sentence (15) might be perfect for that purpose. But that’s a different thing than trying to find a speaker of a specific Balkan language, where (16) or (17) might be more precise. Note that (15) demands and provides a parallel between “someone” and “speaker”, something which (13) above does not do. Note also that if it is a specific language which is sought, (16) and (17) provide a way of making that distinction, while still preserving the fact that it is always a speaker, not a language, that is to be hired.
Well, the Lasnik examples certainly do provide a lot of interesting ideas for our consideration. Let's see if we can generalize a bit:
A. The brain - and thus the language - makes very fine distinctions. This, of course, shouldn't come as a surprise to anyone. Still, "what is" and "what possibly is" are very different things and the grammar must respect that. Verbal systems do it, and so there's no reason that sentence structures can't as well.
B. The brain perceives a correlation between the succession of events and the passage of time, or between one thing (a speaker) and a characteristic of that thing (language spoken). The language can use these correlations to ask questions obliquely, but the correlations must be relevant.
C. Language can use a variety of structures to mean the same thing, and the brain can perceive which are valid variations of one another and which are not.
There is one more point I should make, and you've probably noticed it as well. Notice that even if we take the perfectly grammatical Sentence (6):
(6) It is possible that The Donald will hire someone, but I won't divulge who it is possible that he'll hire.
there's something about it that doesn't quite ring true from a semantic point of view. And that is this: the first clause could be interpreted as merely speculating on the possibility of The Donald's hiring of someone, whereas the second clause could be interpreted as assuming that there is someone under consideration. In other words, the grammatical version of the very bad (1), even when cleaned up, has a "made up" quality to it, as though someone simply invented the sentence as a kind of puzzle. To be more precise, one might instead say:
It is possible that The Donald will hire someone, but I won't divulge whether he will or not and, if he does, who it might be.
This "made up" quality also seems particularly true of
(9) *We left before they started playing party games. What did you leave before they did?
Would anyone really say that? I found this "invented" quality present in any number of the sample sentences that I looked at from the various scholarly papers. Clearly, all of the unsluiced sentences such as (1), (9) and (13) are poorly conceived, and there is no doubt that people are constantly going around uttering poorly conceived sentences. But the thing that nearly all of the sample sentences have in common is that they feature mismatching of one form or another.
who will be hired versus who will possibly be hired
the place that was left versus the time someone left
the language versus the speaker of that language
And so, whether these "island problems" are made up or not, mismatching is what we have to look for in order to solve most of them. Once you understand that point, the solutions are relatively easy. All that you need in order to "repair the island", as I showed with the above examples, is a bit of insight into how the world works: time corresponds to events, speakers correspond to the languages they speak, and to be hired is not the same as to be considered for hiring. With this approach, we can solve them all. Let's do it.
Now let's move along to some examples provided by...
: "Successive Cyclic Movement and Island Repair: The Difference Between Sluicing and VP Ellipsis" also by Howard Lasnik along with Danny Fox.
"Ellipsis and EPP Repair" by Jeroen van Craenenbroeck and Marcel den Dikken. Another star in this field is Jason Merchant, who wrote "Variable Island Repair Under Ellipsis."
that the "that" in the first clause (a complementizer) and the "that" in the second clause are not the same since, for one thing, the second "that", is preceded by a relative pronoun and it needs to be preceded instead by an embedded clause (...it is possible...) which modifies the meaning of the second clause, just as - if you reverted the inversion in the first clause - you would have exactly that:
(5) It is possible that The Donald will hire someone...
And so the "that he'll hire" in the first clause of (1) is not the same as the "that he'll hire" in the second.
