Language Log

QE2

Bob Ladd follows up on my posting about the sentence


with some suggestions about further factors that might be playing a role in interpreting (or failing to interpret) the coordination in this example, beyond the structural considerations I posted about: pragmatic and performance factors.


To recap in some detail: the QE example has a VP with a coordination in it:


The question is: what are the conjuncts?  The second conjunct is clearly the VP deliver a speech (hereafter, (1)).  What's the first conjunct?  There are three possibilities:


Possibility (2) is a dead loss: (1) be coordinated with the "highest" of the three alternatives (against the default tendency Low Attachment, to associate expressions with the structurally lowest available alternative); and the two conjuncts would not be structurally parallel ((2) is a finite VP, while (1) is a base-form VP, thus running against the default preference Internal Parallelism, favoring internally parallel conjuncts); and, fatally, though the subject of the sentence (Queen Elizabeth) is interpretable with the finite VP (2) as its predicate, it doesn't fit with the base-form VP (1), against what is at the very least a strong default preference -- often treated as a rigid requirement -- that the "factor" in coordination (the subject Queen Elizabeth in this case) should distribute equally over each conjunct (Distributivity): the whole thing fails on the bizarreness of Queen Elizabeth deliver a speech as a finite clause (parallel to Queen Elizabeth will be opening the British Parliament). 

I remind you that all three of these principles -- Low Attachment, Internal Parallelism, Distributivity -- are preferences, not laws of nature.  (Even Distributivity is violable in certain circumstances; see below.)

There are still two candidates left.  Low Attachment recommends (4) + (1), opening the British Parliament and deliver a speech, and this is the interpretation I got at first.  But this loses points on Internal Parallelism ((4) is a present-participial VP while (1) is in the base form) and is very bad news on the Distributivity front: (4) is fine as a complement of be, but (1) is unacceptable in that context in standard English (o.k.: be opening the British Parliament, bad: be deliver a speech).

On to (3) + (1).  This loses points by not having Low Attachment (o.k., Low Attachment is just a preference, and there are tons of perfectly fine examples where it's violated), and also by not having Internal Parallelism, since (3) is progressive aspect while (1) is plain, or unmarked, aspect (though again, Internal Parallelism is just a preference, not a requirement).  Meanwhile, (3) is fine on Distributivity, with (3) and (1) both serving as base-form complements of the modal will, and this was the intended interpretation of the QE sentence.

Now Bob Ladd introduces two further factors.  The first of these is the event structure denoted by the coordinated VPs.


This is a general fact about coordinated expressions, especially in "reduced" coordinations: they will ordinarily be understood not merely as denoting two independent things, but as denoting two CONNECTED things -- connected either by temporal sequence, as in


or by association as parts of a larger entity, as in


The tighter the syntactic association of the conjoined expressions -- the more "reduced" the coordination is -- the stronger the implicature of sequence or association in a coherent whole is.  The sentence


is bizarre (but entirely grammatical), even though it's possible that the speaker met both Magic Johnson (the basketball player) and Philip Johnson (the architect) at this party.  It's bizarre because it suggests that the two men constituted some sort of unit.

So Ladd's first point is that (3) + (1) isn't so bad, because opening Parliament and delivering a speech together constitute a single event.  (My "improved variant" has the sequence interpretation, clearly marked by then, so it's golden on almost all fronts.)

Ladd's second point:

Again, this is a familiar effect: in a variety of contexts (the details are very complex) longer-before-shorter is not as good as shorter-before-longer.  This "law of increasing members"-- the term is not my invention, but a genuine technical term, translated from German, and entertaining because of the mild raciness of "increasing members"; no actual "law" is involved, however -- shows up all over the place: salt and pepper is a bit better than pepper and salt, I gave up this fruitless quest for truth is a lot better than I gave this fruitless quest for truth up, and so on.  (There's an enormous literature on manifestations of the tendency.)

I agree with Ladd that the QE example is improved when the second conjunct is longer; there's more time to process the second conjunct before the sentence comes to an end.

And now for still another possible take on things.  On the Low Attachment parsing, the QE example has the present-participial VP (4), opening the British Parliament, conjoined with the base-form VP (1), deliver a speech.  Although (as I pointed out in the earlier posting) this is not a GoToGo example (as in I'm going home and take a nap), it has one significant point in common with GoToGo: the coordination of a present-participial VP with a base-form VP, in which the former fits the larger syntactic context (be takes a present-participial complement in the progressive construction) and the latter does not.  Distributivity fails in both.

The fact is that Distributivity ALWAYS fails for GoToGo; present-participial going (simultaneously representing prospective (be) going (to) and also motional go in construction with a goal adverbial) is a feature of the construction, and so is a base-form VP as the second conjunct.  This is just the way the construction works, for those of us who have  it.  (One more time: for a fair number of people, GoToGo is NOT an inadvertent error, but a regular, though non-standard, part of their linguistic system.)

Here's the point: Distributivity is not some sort of natural law or logical necessity.  It's just a way parts of people's linguistic systems can work.  It makes sense, because it's iconic of the semantic parallelism of the conjuncts: each conjunct is treated the same way formally.

But Distributivity isn't the only possible game in town.  For instance, individual idioms and "small constructions" (like GoToGo) can work any way they want to.  People learn them as special cases, overriding more general conditions like Distributivity.

And there are alternatives to Distributivity -- in particular, a Single Marking scheme, in which one conjunct gets the marking appropriate to the context, and the other conjuncts appear in some default form.  This is exactly as "logical" and "natural" as Distributivity.  In fact, Distributivity can be seen as wasteful and redundant: why mark every conjunct for some feature, when one would do?  Compare this situation to the marking of negation within clauses: many languages have negation distributed to all eligible constituents -- this is "multiple negation" or "negative concord", as in non-standard English I didn't see nobody, and quite generally in Romance and Slavic languages -- but others mark negation only once in the clause, as in standard English I didn't see anybody and I saw nobody.

(I chose the words "wasteful and redundant" deliberately, to counter those who would say that Distributivity is "only logical".  Note that multiple negation in varieties of English is widely, though absurdly, labeled "illogical" -- exactly the opposite value judgment from an insistence on Distributivity in coordination.)

In fact, Single Marking is well attested in (various corners within) the world's languages.  (Confession: I know this is so, but I haven't had the time to do many days of library research surveying the matter, and I might never have it.  But I recall allusions to such systems in the ancient Indo-European languages, and I'm familiar with VP chaining in various languages, in which VPs occur in sequence -- without overt conjunctions -- with the first marked for the appropriate categories in the syntactic context and the rest with much reduced marking.  And the phenomena of agreement "with the nearest" and government "by the nearest" -- not as inadvertent errors, but as parts of a linguistic system -- are clearly related.)

Now back to the QE example.  Maybe this is just Single Marking in the (preferred) Low Attachment configuration: the first conjunct (opening the British Parliament) is marked appropriately (in the present participle form) for the syntactic context, while the second conjunct (deliver a speech) defaults to the base form.  It would be like GoToGo, but on a much larger scale.

Why do I suggest this?  Because the GoToGo Crew (a loose association of linguists who've thought about the construction) occasionally come up with things that look like Single Marking in current English VP coordination, with base-form VPs in non-initial conjuncts.  Here are two from Joel Wallenberg:


I have a few more of these squirreled away in places I can't at the moment locate.  They might, of course, be inadvertent errors.  But they might be indications that Single Marking is around, as a minor constructional option, for some speakers in modern English.

Posted by Arnold Zwicky at December 6, 2007 01:39 PM