July 31, 2004

couldan't, shouldan't, wouldan't

I followed Mark's link to Nathan's Notebook and found the following interesting clipping from a text transcript of Jon Stewart's appearance on Larry King Live on June 25.

Stewart has just mentioned that he is "not a pacifist" -- "As a matter of fact, I like bombing countries." Larry King is surprised, and Stewart clarifies:

Well, just purely for the knowledge of geography. It's just fascinating to learn about these countries. ... I didn't know Kabul was the capital of Afghanistan until we started bombing it. ... If we would haven't gone to war there, I certainly wouldn't have known that.

Would haven't? Totally ungrammatical, I think, and take comfort in the fact that at the top of the transcript page it says plainly:

THIS IS A RUSH TRANSCRIPT. THIS COPY MAY NOT BE IN ITS FINAL FORM AND MAY BE UPDATED.

I've never heard anyone actually utter one of these types of examples; if I didn't have a paper to get back to, maybe I'd hunt down some audio/video of the Jon Stewart appearance on Larry King to hear what it sounds like. It really is so strikingly ungrammatical to me that I can't even clearly imagine how it would sound.

But then I Googled {"would haven't"} and got 5,930 hits. (Of course, Google helpfully asks Did you mean: "would have", which would have gotten me 15,300,000 hits.) Here are some more interesting results (taking reductions of HAVE into account, which as we all know are often spelled in various ways):

modal-HAVE-n't
modal-HAVE-not
modal-n't-HAVE
modal-not-HAVE
would haven't
would ofn't
would'ven't
wouldan't
5,930      
would have not
would of not
would've not
woulda not
71,900      
wouldn't have
wouldn't of
wouldn't've
wouldn'ta
1,690,000      
would not have
would not of
would not've
would nota
2,960,000      
could haven't
could ofn't
could'ven't
couldan't
5,980      
could have not
could of not
could've not
coulda not
15,400      
couldn't have
couldn't of
couldn't've
couldn'ta
973,000      
could not have
could not of
could not've
could nota
1,520,000      
should haven't
should ofn't
should'ven't
shouldan't
5,920      
should have not
should of not
should've not
shoulda not
20,700      
shouldn't have
shouldn't of
shouldn't've
shouldn'ta
843,000      
should not have
should not of
should not've
should nota
1,270,000      
TOTALS
17,907      
111,468      
3,655,900      
4,647,310      

Although I find all of the modal-HAVE-n't examples in the left-hand column ungrammatical, I'll assume that these are not errors of some sort and that some speakers find them grammatical. But I have vague memories of reading somewhere (probably something by (Pullum &) Zwicky?) that n't can only be enclitic to the highest verb in a verbal projection (in this case, the modal) affixed to finite auxiliaries (including modals) [as Zwicky & Pullum (1983:507) point out; see update below]. So what gives? Here's my hypothesis:

  1. Some speakers have reanalyzed modal + reduced HAVE as a single verb finite auxiliary/modal.
  2. The modal-HAVE-n't examples written with unreduced HAVE are not really pronounced with [hæv] -- they're pronounced the reduced way, with [ǝv] or just [ǝ].
  3. Speakers who find the modal-HAVE-n't forms grammatical are among those who have reanalyzed modal + reduced HAVE as a single verb, which is what allows n't to be enclitic affixed in this context.

I wouldn't be surprised to find that there is work out there somewhere showing that (1) is true, or at least plausible. (Remember, I'm a phonologist, I don't read much of this stuff anymore. For all I know, the whole topic I'm talking about here has already been addressed somewhere.)

The empirical claim in (2) needs verification, and in fact one might surmise that the separated Google results directly contradict it:

modal-HAVE-n't
(unreduced HAVE)
modal-HAVE-n't
(reduced HAVE)
would haven't
5,930      
would ofn't
would'ven't
wouldan't
29
could haven't
5,980      
could ofn't
could'ven't
couldan't
20
should haven't
5,920      
should ofn't
should'ven't
shouldan't
28
TOTALS
17,830      
77

On average, then, I found about 232 instances of unreduced HAVE for every one instance of reduced HAVE (all three spellings combined) in the relevant examples. That's pretty striking. But here's what I think: a person writing down one of these examples really doesn't have much choice. Consider the options. Even if this person often writes e.g. woulda or would of, adding enclitic affixing n't to one of these is extremely odd. Enclitic The affix n't doesn't really fit well on would've either, because the result is a form with two apostrophes. So the person is left with would haven't -- not perfectly consistent with the reduced pronunciation, but the best orthographic alternative under the circumstances.

I realize this isn't quite sufficient evidence to reach the conclusion in (3), but that's my story and I'm sticking to it until I hear a better alternative. (Arnold? Geoff? Bueller?)

Interestingly, the single hit for could'ven't that I came across is an example sentence in a handout from a talk given by David Lightfoot. The sentence is starred:

34.a.Kim visited NY and Jim could've VPe.
b.Kim visited NY and Jim couldn't VPe.
c.Kim visited NY and Jim couldn't've VPe.
d.I'd've visited NY.
e.*Jim could'ven't seen it.

The reason? Apparently:

32.E.Syntactic rules can affect affixed words, but cannot affect clitic groups.
F.Clitics can attach to material already containing clitics, but affixes cannot.

(As far as I can tell, there are no principles A-D anywhere in the handout. Maybe this is an example of auto-numbering in Microsoft Word gone haywire.)

Update, 8/1/04: Nope, just a case of citation without full representation in the body of the handout. Arnold Zwicky writes to tell me that (32E,F) are the last of six criteria distinguishing (inflectional) affixes from clitics in Zwicky & Pullum (1983), "Cliticization vs. inflection: English n't", Language 59.3, pp. 502-513 (cited in Lightfoot's references).

My defense is in three parts:
  1. It was late. I was tired. Really.
  2. It didn't occur to me to look in the references, since the only citations in the body of Lightfoot's handout are (a) Lasnik & Saito 1984, (b) Aoun, Hornstein, Lightfoot & Weinberg 1987, (c) "Syntactic structures (1957)", (d) Gibson & Wexler 1994, (e) Dresher 1999, and (f) Clark 1992.
  3. I at least remembered having read something by Zwicky (& Pullum) about this subject.
Now that Arnold has called me out, I am re-reading Zwicky & Pullum 1983. If you care to join me, you can download a smaller (low-res) file here or a larger (high-res) file here (courtesy of JSTOR). I'm also very much looking forward to Arnold posting a reply to this post.

Assuming that n't is an affix but that 've and 'd are clitics [as shown by Zwicky & Pullum 1983; see update above], the contrast between (34c,d) and (34e) follows from (32F).

(32E) is necessary to explain the following contrast. (33a) is grammatical because couldn't is an affixed word and thus licensed, by (32E), to invert with the subject Kim. (33b) is ungrammatical because could've is a clitic group and so is not licensed to invert.

33.a.Couldn't Kim see that?
b.*Could've Kim seen that?

So, now I wonder a few things.

  1. What would Lightfoot [rather, Zwicky & Pullum; see update above] say about the grammaticality for some speakers of modal-HAVE-n't examples like (34e)?
  2. How do speakers who find the modal-HAVE-n't examples grammatical judge the form in (33b)? (My hypothesis predicts that (33b) should be grammatical for them.)
  3. What am I still doing awake?

Good night.

[ Comments? ]

Posted by Eric Bakovic at July 31, 2004 02:58 AM