Wanna is a colloquial contraction of want to. On its own, that's not a very interesting fact. English is full of contractions; the use of contractions like wanna and gonna and can't is characteristic of ordinary speech, while in formal speech speakers tend to avoid them. This is all quite familiar to any native English speaker. But wanna is actually very interesting, because it provides evidence for something predicted by syntactic theory, and something that ought to surprise you: there are invisible words in our speech.
The following is a digression which you should feel free to skip if you want.
Linguists put a lot of effort into making sure their theories are philosophically sound. Moreso, in my experience, than folks in other sciences. They worry a lot about explicitly making sure their ideas meet the standards for scientific theories. For instance, they labor to make each theory as minimal as possible — that is, to make sure that there's no extraneous theoretical machinery. You start with the smallest hypothesis possible, and you only add to it under duress. If you work that way, it sometimes means developing seemingly strange ideas about how things work within a language in order to avoid adding any new rules or mechanisms to the theory.
Of course, that's part of the scientific method whether you're studying languages or tree frogs. You shouldn't propose anything that you don't have to; it's a leap forward when you can find a way to eliminate something unnecessary from your theory. But linguists spend a lot more time talking about how to make their theories "pure" than a lot of scientists.
I have my own idea about why this is. Linguistics, as a type of scientific inquiry, is relatively young. And linguistics is a social science; it's often treated as a branch of anthropology. As much as any linguist would like to say that linguistics is driven by empiricism, it's also partially driven by the trends that you see in any social science. I don't mean to disparage the social sciences — in fact, I think it's the only possible approach to answering some very important questions. But there's a difference between linguistics and, say, chemistry. There was never a historical shift between "structuralist chemistry" and "generative chemistry". Paradigm shifts in the natural sciences are driven by new discoveries; while knowledge and methods improve over time, the basic approach of chemistry doesn't. New discoveries happen in linguistics as well, but shifts in the field tend to be heavily motivated by changes in how linguists look at the field, and new ideas about how to go about conducting linguistic inquiry.
Part of the problem is that it's simply harder to do an experiment in linguistics. In the natural sciences, in many cases you can simply set up certain conditions and see what happens. Linguistics has to be grounded in observations. The big questions often can't be answered through straightforward experimentation. For example, there's an eternal question of exactly how much grammatical machinery is encoded in the brain at birth. And there's no way to answer it directly via experiment, since doing so would have to involve experimenting with depriving a child of linguistic stimulation or deliberately exposing it to bizarre artificial languages to determine exactly which possibilities are ruled out by the brain's language hardware. In other cases, it would be simply impossible. Even if it were ethical to dissect someone's brain to watch sentences form, there's no way to do so.
So linguists use indirect approaches instead. In syntax, the major one is attempting to create sentences that capture some particular phenomenon to be examined. If you come up with good choices, you can use a native speaker's judgment to decide what constitutes a grammatical sentence and what constitutes an ungrammatical one, and thereby tease out the underlying ways in which sentences are formed from words. It's slow and difficult, and it's often hard and occasionally impossible to create sentences that capture the phenomenon under investigation. But it's all there is. In the remainder of this writeup, I'm going to walk through one such analysis and show how it demonstrates a key feature of syntactic theory.
The concept of invisible words troubled me when I started studying syntax. I'm an empiricist at heart; I don't like to take things on faith. I want evidence. You see, invisible words — traces, as they're called in linguistics — are a very nice little theoretical tool, but it's hard to prove they actually represent something real about language. Traces are little blank spots left when a constituent — a word or phrase that acts as a syntactical unit — moves from one spot in a sentence to another. Movement seems like an odd concept in and of itself; the idea is that when your brain first puts together a sentence, it takes a different form than it appears on the surface. Sometime between when it starts out and when it actually leaves your lips, stuff gets moved around. That is, there's an underlying structure for a sentence that's different from what comes out in the end.
Movement makes sense, though. You can see it easily. Take a sentence like Maria will bake a chocolate cake. Then turn it into a wh-question asking what Maria baked — a wh-question is one that uses a question word like what or where. You end up with What will Maria bake? Notice that the direct object of bake is moved to the front of the sentence (the modal verb will moves as well, but that's not what's important here.) In English, a verb's direct object normally follows it, which might make you predict something like Maria will bake what? But that's not what you get. The wh-word moves sometime in between when it's used as the object of bake and when you speak it.
However, the reasoning behind traces is a lot less convincing. Traces show up because they're useful from a theoretical perspective. They allow us to maintain certain theoretical assumptions even when they appear to be violated. This sort of reasoning bugs me. Like I said, I'm an empiricist. Elegant theories that involve only a few basic rules are nice and all, but when the notion of traces was first presented to me, I was unconvinced. It felt a lot like the facts were being modified to coincide with notions of what makes a theory elegant, when obviously it's the theory that should change to fit the facts.
S
/ \
/ \
/ \
NP VP
/ \ / \
The N' jumps PP
/ \ / \
quick N' over NP
/ \ / \
brown fox the N'
/ \
lazy dog
This picture is a syntactic tree; the details aren't really important; it's a bit simplified to skip irrelevant parts. (Sorry about the quality of the ascii art.) The labels on the nodes in the tree aren't important either. There are two key things things to understand. Each node on the tree, and everything below it, represents a constituent — that is, each is a type of syntactic unit. Syntactic operations, like movement or replacement, apply to nodes on a syntactic tree. Either of the NP (short for "noun phrase") nodes, for instance, can be replaced with it. The other important thing to see here is that the tree is binary-branching: that is, each node, except for the nodes representing individual words, has exactly two daughters.
A common notion in modern syntactic theory is that all syntactic trees are binary-branching; this idea comes from the way trees are imagined to be built in the mind. It starts with individual words, which are the lowest nodes on the tree; one word is merged with another to form a higher-level node, which can then be merged with other nodes, eventually encompassing an entire sentence. So, from a theoretical perspective, it's a bit easier to imagine that merging always takes place between exactly two nodes. The merge operation is simpler this way; from one perspective, that makes the theory itself smaller, and naturally the theory that involves the least underlying machinery in order to match the relevant facts is the best.
This is when traces come into play. If you pluck a random word out of the tree and move it somewhere else in the sentence, you have a problem. You see, suddenly that word's mother node only has one daughter — it's no longer binary-branching. Take the question What did the quick brown fox jump over? The object of the preposition over is now at the front of the sentence. Over no longer has a sister in the tree; it can't simply form the end of the tree, because over, on its own, isn't allowed to merge with jumped — jumped is looking for a PP ("prepositional phrase"), while over is just a bare preposition. You can't merge this bare preposition with the verb (trust me on this one, because the reasons are complicated.) And we can't assume that there is a PP with only one daughter, over, because our tree would no longer be binary-branching.
So instead, we posit that when something moves from one spot in the sentence to another, it leaves a trace behind. A trace is basically an invisible word; it's not pronounced, but according to syntactic theory it's still present on some underlying level. This is necessary to maintain our theoretical assumption that trees are binary-branching. But doesn't this seem like a weak-ass justification for its claim? And conveniently, since the word is invisible, there's no way to prove that it's not there. Which means the theory is unfalsifiable. The philosophers of science have pointed out that falsifiability is a vital aspect to any theory — after all, if you can't find any way to test a theory, if you can't imagine any piece of evidence that could prove it wrong, then how can you believe it?
You can understand my skepticism.
This is where we come back to wanna. You see, wanna provides a neat bit of evidence that traces are real. Take the following two sentences:
[1] (a) I want to invite Carlotta to the party.
(b) I want Carlotta to come to the party.
Now turn these sentences into questions:
[2] (a) Who do you want to invite to the party?
(b) Who do you want to come to the party?
Now, of course, these two sentences look the same, on the surface. They're word for word identical, except for the words invite and come. But we know they're not the same underneath, because the corresponding statements have a different structure. So let's put those traces in the proper spots and see what happens. I've marked them with <angle brackets>, which is typical notation. (You can substitute whom if you prefer; I use who because, like most English speakers, I usually don't maintain the historical distinction between who and whom.)
[3] (a) Who do you want to invite <who> to the party?
(b) Who do you want <who> to come to the party?
Those invisible words are in two different places in the two questions. And lo and behold, something interesting happens when you try contracting want to:
[4] (a) Who do you wanna invite to the party?
(b) *Who do you wanna come to the party?
Whoops. If you're a typical English speaker, (4a) sounds just fine. Perfectly natural. But (4b) sounds, well, odd. It's ungrammatical (which is indicated by prefixing a sentence with an asterisk.) This isn't just my intuition, either — examination of people's spontaneous speech confirms that people simply don't do wanna-contraction there. Why? Because there's an invisible word in the way.
So it turns out that an analysis of the conditions that permit wanna to contract demonstrates something really cool: these things called traces, predicted by linguistic theory for reasons that are provocative but not convincing on their own, are real. The sentences you speak have invisible words in them. Furthermore, an introductory syntax course will show that, in fact, every sentence contains traces. Through analyses like the one above, we can show that there are words and phrases moving around from place to place in every sentence.
A person looking at the basic claim might well be skeptical. Because of a highly abstract assumption in syntactic theory — an assumption about the branches that form syntactic trees — we have to make certain predictions about how sentences work. It's gratifying and, if you're a linguistics nerd, unendingly cool to see those predictions borne out by evidence. By examining lots of little pieces of data like this one, it's possible to tear apart the basic workings of language to see what's going on underneath.
Reference
Adger, David, 2003. Core Syntax: A Minimalist Approach.
Plus lots of lectures on this issue and related ones. | 
