July 26, 2004

The thin line between error and mere variation, part 4: Do I misspeak?

Much of my recent research has to do with syntactic variation in English (I really must edit my website to reflect this) -- sometimes on details of constructions that are for the most part uncontroversial, sometimes on phenomena that are very widespread but are condemned by some usage manuals, sometimes on relatively infrequent and largely disregarded phenomena. I seem to have specialized in variation that isn't tied in any obvious way to the standard extralinguistic factors (geographical region, class, age, sex, race/ethnicity), although a few of the variables are associated with informal style or with speech as opposed to writing. (To get the flavor of this research, check out the handout for my "Seeds of Variation and Change" paper at the 2002 NWAV conference.)

Now, I'm used to having people, especially non-linguists, respond to some of my data through the lens of rules they've been taught. Being Blinded By the Rules, I call it. It seems that once you've had a generalization about grammar, however spurious, made explicit for you, you can no longer judge language like a normal person; a little learning is a dangerous thing. You may deny that you use some variant -- possessive antecedents for pronouns, split infinitives, stranded prepositions, certain types of "dangling modifiers" -- when in fact you use it with some frequency. You may make tortured attempts to avoid this variant. You will certainly discredit reports that other people use a variant that you don't -- say, Isis ("The problem is is that I don't speak that way"), GenXso ("I'm so not going to talk about this"), or themself ("Everybody should get themself a research project"). You'll be inclined to treat these usages as errors, not as real linguistic variants, that is, parts of somebody's grammar (maybe your own).

But for some time now I've been getting Blinded By the Rules responses from other linguists.

Sometimes, most annoyingly, on variants that I use myself, like "a morphological rule or rules". I'm told that if I'd only think about these examples, I'd see that they were ungrammatical. This is a covert reference to some hypothesis about the structure of English, in the face of which colleagues just deny other people's judgments (and practice). It's just like a prescriptivist confronting someone who says "Me and Kim did it" with the instruction to just think about the sentence; you don't say "Me did it", do you? (You don't say "a morphological rules", do you?) When variants can be associated with recognizable social categories, most linguists know enough not to treat variation as error -- that would be insulting -- but when the variants aren't socially anchored, linguists tend to revert to behaving like ordinary people, except of course that the linguists are in possession of theoretical hypotheses that probably wouldn't have occurred to non-specialists.

If I'd stuck to looking at the details of constructions that are for the most part uncontroversial, there wouldn't be much of a problem. I could continue looking at stranded to (infinitival to stranded by Verb Phrase Ellipsis, as in "They told me to jump, but I didn't want to"; see my "Stranded to and phonological phrasing in English", Linguistics 20.3-57 (1982)) and the Quasi-Serial Verb construction ("I'll go see who's at the door"; see Pullum's "Constraints on intransitive quasi-serial verb constructions in Modern English", OSU WPL 39.218-39 (1990)), and I'd get little grief from my colleagues. (I do, in fact, continue to look at these.) But when I branch out from there, I start to be confronted by collegial disbelief.

This is easiest to respond to for variants that are condemned by usage manuals, like nominative coordinate object pronouns ("Just talk to Kim and I") or themself (the topic of a 2003 Stanford honors thesis by Joel Wallenberg). Every linguist knows that things get in the usage manuals only because a lot of people are inclined to say/write them, so that if a lot of usage manuals condemn it, there must be a "dialect" (in some sense of this word) in which this variant is grammatical, even if this dialect is not easily associated with extralinguistic factors (as these two are not).

After that, we wade into dark waters. Two sample cases, one involving a variant that I don't have (Wh-that, as in "I wonder how many people that were at the party"; see my 2002 article "I wonder what kind of construction that this example illustrates", in Beaver et al., The Construction of Meaning, 219-48), another involving a variant I do (GoToGo, as in "She's going to San Francisco and talk about firewalls", the topic of a 2004 Stanford qualifying paper by Laura Staum). Note that some sort of error was almost surely involved in the historical origin of each of these constructions, so there's room here for the exercise of the etymological fallacy, which would class them as errors rather than variants, but that's not my point here.

Wh-that is not at all frequent -- from ten years of fortuitous collection, I have only 50 examples -- and it occurs mostly in speech, where it could easily be overlooked. Colleagues I showed examples to were at first inclined to dismiss them all as production errors, blend slips, in fact; these colleagues would point out that if you just thought about the examples, you'd see they have an extraneous that (extraneous from the point of view of the usual formulations of the rules for embedded Wh constructions). Quite possibly some of the examples were production errors. But in other cases, a single speaker produced a number of examples, never self-corrected, and (when I was able to question them) judged them entirely acceptable, typically not understanding why I was asking. So I argued that, for at least some of the speakers, we were looking at a genuine variant, not an error.

GoToGo isn't very frequent, either, but it's been around a long time (David Denison, at Manchester, has a big file of examples going back to ca. 1930), and for those of us who have it (maybe 20% or so of modern American English speakers, including me and Ivan Sag, but not my daughter or Tom Wasow, just to choose people from my immediate milieu) it's unremarkable. When challenged on it, by colleagues who said, helpfully, just think about, you'll see it's ungrammatical (a failure of parallelism in coordination), my first response was, but how else would you say it? (The answer is: "She's going to go to San Francisco and talk on firewalls", with two occurrences of go rather than the one in my version.)

In both cases, my colleagues' response can be unpacked into two parts: (1) this isn't English, because it's not grammatical for me (the colleague); and (2) it can't be English, because it violates a well-known constraint (on embedded Wh constructions, on parallelism of form in coordination). Part (1) isn't a problem for me; everybody's entitled to their own grammar, after all, and if Wh-that and GoToGo aren't your things, that's ok with me. Part (2) is the sticking point; keep your goddamned hands off my grammar. If my grammar doesn't obey the rules you cite, then either you've formulated the rules wrong (this is, I think, the case for certain instances of "dangling modifiers", as in "As a parent, my concern is with the children"), or there are special constructions to which your rules do not apply (this is what I claim for Wh-that and GoToGo). If the special constructions were associated with, say, the Pittsburgh area, or working-class speakers in Northern cities, or African American men, linguists would readily admit the reality of the variants, however far from their own grammars they might be. But if it's just some random collection of speakers sprinkled across the geographical and social landscape, then even linguists are inclined to think they're looking at error rather than variation. Sigh.

Now for an even tricker case. The story begins with a posting of mine to ADS-L on 7/9/04: "For fans of the type of nonconstituent coordination referred to in the generative literature as Right Node Raising (RNR), here's an extraordinary example from the 7/8/04 Palo Alto Daily News (p. 12), in a letter from Charles Browning, M.D. of Palo Alto, on health care costs: A 2004 Institute of Medicine report... documents that the uninsured, unable to afford health insurance, have less access to, and receive inferior, care."

I should say at the outset that, for me, RNR tends to be formal, self-conscious, and mostly restricted to writing, though there are certain instances of it that are unproblematic for me in that context ("They gave books to, and in return took money from, numerous customers"). The Browning sentence was notable enough for me to label it an "extraordinary coordination", something I wouldn't expect to come across in speech rather than writing. Other ADS-Lers agreed with me, but others had their doubts. Larry Horn fired back wryly that same day (7/9/04) with a flagrantly zeugmatic example, suggesting that the Browning sentence was similarly flawed: "And then there are the unconcerned uninsured, who have less access to, but don't really, care." After that, things tended to decline into a parade of My Favorite Zeugma (Flanders and Swann and all that).

And then (still on 7/9/04), from a distinguished colleague I'll refer to as K, there came this e-mail message: "Hey, Larry, with his wit, has made me see that the example is actually completely ungrammatical." K went on to explain that it couldn't be grammatical, because it violated a constraint on reduced coordination, namely that the shared element had to be assignable to a single category, but that "care" in "have less access to care" was a NP while "care" in "receive inferior care" was only a Nom (depending on your religious beliefs, you can substitute, for NP and Nom, either N" and N' or, oh alas, DP and NP). There is actually a very subtle point here, having to do with K's implicit claim that "care" in "have less access to care" is a NP but not a Nom, but let's not get hung up on that; what so annoyed me about K's response was that it simply denied my grammaticality judgments and cited, in defense, a hypothesis about the analysis of RNR -- the (1)-(2) punch above. K and I have failed to come to any sort of understanding about this.

It gets worse. On 7/11/04 I wrote to K: "by the way, what the hell licenses "a morphological principle or principles" 'a morphological principle or morphological principles'?" This was a matter of some concern to me because I'd just used this very phrase, quite unreflectingly, in a Language Log posting, and then noticed the odd way in which modifiers were distributed across the conjuncts. True to form, K replied, still on 7/11/04: "I don't regard "a morphological principle or principles" as grammatical. I think it sneaks by under the radar like "if you have or would apply"...." That is, it's an error; if I'd only thought about its form, I would have seen that it violates a condition on reduced coordination. Later that day I defended myself: "It's perfectly fine for me, and parallel examples (which I find acceptable as well) are easy to find. In fact, "a specific/particular person or persons" and "another person or persons" are abundant in legal and administrative contexts."

During this exchange, K resurrects another coordination puzzle, from three months before: "If you have or would apply..." On ADS-L on 4/11/04, I reported that a grad student had sent me the example (from a friend in e-mail) "I could (and have) watched people play that game for hours." Here there are two coordinated auxiliary verbs, requiring two different verb forms (base and past participle, respectively), which should (according to some hypotheses about reduced coordination) result in an irresolvable conflict, and ungrammaticality -- but, instead, the form required by the nearer conjunct (the second one) determines the verb form. Similar examples can be found in more elevated contexts:

(Ex 1) There are plenty of venues at which Mr. Chirac could, and has, demonstrated his rapport with Mr. Schroeder. (New York Times editorial, “Playing Politics with D-Day”, 1/19/04, p. A20)

(Ex 2) To the Editor: The United States government's attempts to manipulate the world price of crude oil by increasing gasoline taxes or by releasing oil from the Strategic Petroleum Reserve have and always will fail. (Richard J. Stegemeier, "retired chairman and chief executive of Unocal Corporation", New York Times letter, 5/27/04, p. A30)

As I reported on ADS-L: "The grad student suggested I do a Google search on "could +and have" -- which yielded over 8k hits (without going into Google groups)! Some of these are irrelevant, of course, but, still, the number is huge. Actually, "would +and have" netted over 17k hits. Even "might +and have" got about 2k. So there's a hell of a lot of determination by the nearest going on in unguarded writing. Agreement with the nearest is ridiculously easy to collect from unscripted [English] speech, though most linguists seem to treat it as a performance error. [It's well-known that agreement with the nearest is grammaticalized in certain contexts in some languages.] I'd imagine that government by the nearest is also common in speech, given how incredibly frequent it is in writing. I'm tempted to suggest that government by the nearest conjunct is in fact the rule for vernacular English -- which would explain why it's so hard to teach people to avoid this construction in formal writing."

But K insists on treating government by the nearest as an error, not a variant. After all, it's not in K's grammar (or mine, for that matter), and it violates a generalization about how reduced coordination works.

One brightish note: when I sent K a note on 5/15/04 about "I have had my car washed and hair cut" (from a postcard I'd just written to a friend), which has yet another sort of odd distribution of a modifier across conjuncts, K merely confessed puzzlement about its analysis -- undoubtedly because K found it grammatical (as I do).

Still, I sometimes despair about getting colleagues to take my research program on syntactic variation seriously -- as research on syntax, not on speech errors.

zwicky at-sign csli period stanford period edu

Posted by Arnold Zwicky at July 26, 2004 09:13 PM