Can we parse the hell out of this construction?

Question

• John only had the Ferrari for three months, but while he had it, he drove the hell out of it.
• I'm pretty sure I aced the hell out of that test.
• That last stuff you got us, we smoked the hell out of it.

This works like most other intensifiers: "X Y'd the hell out of Z" means roughly "X Y'd Z intensely/rapidly/immediately/determinedly", or, for that matter, "X fucking Y'd Z". But how does it work like that? ("The hell out of" is clearly not any kind of syntactic constituent that can head a phrase, unlike the paraphrases…)

Notice that this is different from:

1. "The hell" in wh phrases (What the hell is that?), which has been studied quite a bit (and which has a relevant question on this site: What "the hell" is this?).
2. "The hell" in non-sentential utterances (The hell you say!), which I think I've seen covered in passing a few times.
3. "The hell" as a specifier on a particle or preposition (We got the hell out), which I've only seen a tentative blog post on, but seems relatively simple.¹

Presumably there are connections of some kind (and the last one can rarely be ambiguous with this one²), but this clearly isn't the same construction as any of those, and, in particular, there's a major syntax-semantics mismatch here that doesn't exist in any of them.

Syntactically, "he drove the hell out of his car" seems to be identical in form to a transitive resultative, like "he drove his car into town"—"the hell" is a normal NP complement/direct object, and "out of his car" is a normal PP complement.³ Wh-movement, topicalization, passivization, pronouns, etc. all work the way you'd expect.⁴

But semantically, it's very different. The verb subcategorizes for a patient as the direct object, but instead it gets "the hell". The preposition demands a location but instead gets the verb's patient, and wants to describe a path but instead has its semantic content bleached out or ignored.^{5, 6}

A few more details:

• I think "the hell" can be replaced by any the-hell-NP that works in the particle-specifier case (#3 above), so at least there's no mystery there.⁷
• The verb seems to have to be one that normally takes a patient as an (obligatory or optional) direct object and has no other obligatory complements; there may be more restrictions to it than that.⁸
• I have no idea what cross-language parallels there are, if any, or any dialect variation,⁹ or even how much variation there is in acceptability among speakers beyond the handful I quizzed on a tiny corpus.

But my question is how this can be analyzed.

• In mainstream Chomskyan grammar, I have no idea how you'd even begin to move things into the right position to get the LF and PF right. Or, for MP instead of GB/P&P, I assume the movement happens before case marking, and there's a whole chain of leapfrogging jumps, but I have no idea what constraints could trigger any of that to happen, or what the original form is. (There's no little-n/abstract-N akin to Larson's account of verbs, and, even if there were, the apparent movement here is entirely out of the NP/DP/whatever.)
• In Jackendoff's Parallel Architecture, I can imagine the basic idea a lot more easily, along similar lines to the way he handles, e.g., light verbs. But I'm not sure what kind of constraints in the lexical entry could restrict unification properly without something¹⁰ that isn't necessary for any of his worked examples.
• Similarly, in most Construction Grammar variants, how to staple the verb and patient to the construction is obvious, but how to license only the right set of entries… I have no idea (unless it's by listing the cartesian product of every the-hell-able verb and every valid the-hell phrase as individual constructions, which seems implausible).
• If it's been studied in some other formalism, that's fine, but I'd appreciate a pointer to an introduction, along with the study of this construction.
• Failing that, if there are parallels with other, better-studied constructions, that would be helpful.
• … or, if nothing else, at least a solid description without any attempt at analysis.

---

_{1. Whatever analysis you like for "right" as a left modifier on Prt/P/PP/whatever, it should straightforwardly apply to "the hell", especially given "right the hell out/*the hell right out". If I'm missing something that makes it tricky, let me know and I'll ask a separate question on that…}

_{2. Most blatantly, "I flunked history 301" and "I flunked out of history 301" are near-synonyms. You can the-hell-out-of the former, and you can the-hell-intensify the particle from flunked-out, and either way you get "I flunked the hell out of history 301". I think the first one may actually be invalid here, because this transitive sense of flunked doesn't take a patient (the other does, of course, so "the teacher flunked the hell out of John"), but there may well be cases that are perfectly acceptable as either.}

_{3. I know there's some dispute about "out of it" in the first place—PP with a compound P, PP with a particle or adverb specifier, nested PP… I don't think it makes a difference; whatever analysis you give normal uses works the same here.}

_{4. That isn't quite true, but I think the exceptions are all caused by register or other pragmatic violations. For example, pied-piping never sounds good with blatantly colloquial constructions except in, e.g., ironic butler-speak, and the same is true here: "Out of whom did the young master beat the hell?"}

_{5. But notice that the PP sounds weird with a specifier, unlike the particle-specifier case: "We got out of there" can be double-intensified to "We got right the hell out of there" just fine, but double-intensifying "We smoked it" to "?We smoked the hell right out of it" is questionable. (In fact, from a quick survey, it sounds British to Americans, and American to Brits, for whatever that's worth…), and you definitely can't use a the-hell specifier in a the-hell-out-of VP, as in "*We smoked the hell the hell out of it" or "We smoked the fuck the hell out of it".}

_{6. Of course "out of the car" is a path, but that sense doesn't seem to appear in the meaning at all—John is literally driving the car, not causing something to take a path out of the car, even metaphorically.}

_{7. For example: He drove the hell/*hell/the nine hells/*nine hells/the heck/the H-E-double-hockey-sticks/the fuck/the fudge/the shit/the sugar/*the shoot/the flaming hell/the all-time fuck of all fucks/every last fuck/*all the shit/*the bastard out of it, which is the same distribution in We got the hell/*hell/etc. off his land.}

_{8. Some verbs are tricky. Besides the "flunk out" case above, there's a whole class of verbs beat/kick/whip/etc. that have an open idiom "beat the tar/stuffing/etc." out of. And you can stretch that idiom to some extent to other contact verbs, even positive ones like kiss/screw/etc. So, is "I beat the hell out of him" a the-hell-out-of of "I beat him", or the "beat-the-N-out-of-NP" idiom with "hell" as the N?}

_{9. For example, some British speakers can use a much wider range of words as intensifiers the way Americans use "fucking". They can at least arsing drive a manual twatting transmission, but can they drive the twat or arse out of a car the way Americans drive the fuck out of it?}

_{9. A second-order lambda that …???}

Answer 1

After reading a bit more Jackendoff and Culicover, I think I understand how this works in Parallel Architecture. But (a) I still have a remaining sub-question at the end, and (b) I'd still love to know the MP or GB analysis. So, I'm answering my own question, but leaving it open for someone else to do (a) or (b) or anything else that seems better.

tl;dr: PA can handle all of the selection issues in terms of whether the words can be coerced at the Conceptual Structure (semantics) level, so the construction doesn't need to specify them.

---

Oversimplifying, PA is basically a Construction Grammar framework, but one where constructions are explicitly made up of parallel Phonology, Syntax, and Conceptual Structure (and maybe some other stuff, but not relevant here) trees, with links between the components.

So, this construction might look something like this:

• P(honology): […]₁ […]₂ [out of]₃ […]₄
• S(yntax): [_VP V₁ NP₂ [_PP P₃ NP₄]]
• C(onceptual)S(tructure): F₁(PATIENT₄, MANNER₂)

With the particular example "drive the hell out of the car", you get something like:

• CS: DRIVE(PATIENT:the car, MANNER:intensely-as-hell)

(I'm ignoring the subject and the tense here and just dealing with the VP.)

So far, this is just like the CS you get from "drive the car quickly", except with a different MANNER.

---

But you can't just use any NP that can be coerced into expressing a MANNER. It has to be definite, and its head has to be one of those all-purpose-nounish-swear-words. I don't think all such NPs express the same fixed MANNER; the different shades of "drive the hell out of the car" vs. "drive the everloving fuck out of the car" must come from the semantics of "hell" vs. "everloving fuck".

So:

• P: […]₁ […]₂ [out of]₃ […]₄
• S: [_VP V₁ [_NP … N₅ …; definite]₃ [_PP P₃ NP₄]]
• CS: F₁(PATIENT₄, MANNER:all-purpose-nounish-swear₅)

Is it plausible that we have a class like "all-purpose-nounish-swear" that includes "hell" and "fuck" and "fudge" but not "shoot"? Definitely.

---

The construction doesn't constrain the verb at all. And that's fine—if you try it with a verb that can't take a PATIENT and a MANNER, and there's nothing in the discourse that will lead you to coerce it appropriately, it fails semantically. For example, you can't normally play the hell out of the book, but if you come up with the right context where you can play the book, or play the book quickly (maybe the book is a Fighting Fantasy gamebook), you can play the hell out of it. You can even give the hell out of the book with the right context (there's a textbook donation drive, and you're giving lots of copies of the same one), even though give normally wants a BENEFICIARY and THEME, not a PATIENT. But you can't sleep the hell out of the book, for the same reason you can't sleep the book quickly, or just sleep the book, because I don't think there's any context that would let you coerce SLEEP to take a PATIENT.

So, my question about how to specify a construction that only applies to verbs with the right argument structure is simple: you don't bother.

---

Similarly, it doesn't constrain the MANNER at all, even though it seems to always be an intensifier. I think that's also fine. It seems reasonable that every word in that all-purpose-nounish-swear category includes, as part of its lexical entry, some vague measure of intensity (and all of them are positive intensity). Or maybe something else (a measure of rudeness?) can be coerced to intensity. Otherwise, how could we know that "drive the fuck out of the car" is more intense than "drive the heck out of the car"?

Could some of them also include something else coercible to a manner, that competes with their intensity? Maybe. But actually, I'm not sure there isn't actually such a shading to the meaning. If you compare "drive the shit" vs. "drive the hell", I think the former does seem to connote poor driving as well as intense, and the latter, rebellious as well as intense. If I'm wrong, then presumably we need some kind of operator that forcibly coerces through INTENSITY rather than just on the best path to MANNER: CS: F₁(PATIENT₄, MANNER:INTENSITY(all-purpose-nounish-swear₅))

---

Finally, consider:

1. drive the squirrel out of the car (CS something like FORCE(EVENT:MOVE(AGENT:squirrel, GOAL:out of the car)))
2. ?drive the stink out of the car (interpreted like 1, but odd because a stink isn't animate)
3. drive the hell out of the car
4. drive the cruk out of the car (interpreted like 3, inferring that "cruk" is an all-purpose-nouny-swear)

So, what drives people to interpret 2 and 4 the way they do (assuming I've guessed right)?

I think this gets into performance details: the the-hell construction is pretty low-activation, so it usually doesn't compete with the normal one, but hearing a general-purpose-nouny-swear primes it (and, presumably, the what-the-hell-question, indifferent-what-the-hell, the-hell-you-say, and any other constructions). When you hear a non-word like "cruk", well, it sounds like a made-up swear word. In fact, many sci-fi swear words (most infamously "frak") do. And in the production direction, trying to express "really intense" also primes the various hell constructions (or just primes the words, which prime the constructions).

But what about this made-up swear word from Hitchhikers Guide:

• drive the Belgium out of the car

For me, that sounds like a perfectly good hell-intensifier. But however you define "sounds like a swear word", surely "Belgium" isn't even close. It sounds French/Latin rather than Anglo-Saxon, it's multisyllabic, it has a soft coda, and, worst of all, it's already a real word.

My hope is that people who haven't seen the TV version of Hitchhiker's Guide and/or read the US version of the book don't interpret this like "cruk", and instead reject it for similar reasons to 2, so there's no issue here. But if not, then there's a problem that I don't have an answer for. (And possibly the very thing that I felt was missing when I first read Simpler Syntax but could never nail down…)