November 16, 2007

Opening Parliament and deliver a speech

Back on 11/5/07, I caught this on BBC News on NPR:

Queen Elizabeth will be opening the British Parliament ... and deliver a speech.

I balked.  It sounded to me like a GoToGo sentence --

I'm going to the library and study for my test.

in that it seemed to have a present participial VP (opening the British Parliament, going to the library) paired in coordination with a base-form VP (deliver a speech, study for my test), when only the first structure would be licensed by the relevant head V, (apparently) be.  But the cases differ: the details aren't quite the same (GoToGo mostly has going to, or at least some present-participial motion verb plus infinitival to, while the QE example seems to have no such restriction on its first conjunct); and for me (and some others), the GoToGo examples are just fine, while the QE example struck me as bizarre.

Then I saw that the QE example had a acceptable parsing, though not one that was easy to discern.

This is the parsing in which the VP is

[ will ]  [ [ be opening the British Parliament ] and [ deliver a speech ] ]

that is, as a reduced variant of

[ will [ be opening the British Parliament ] ] and [ will [ deliver a speech ] ]

from which the shared head will can be "factored out".  The result is a coordination of two base-form VPs (be opening the British Parliament, deliver a speech), so what's the problem?

Well, two problems.  The first, and more subtle, is that though both VPs in this analysis are in the base form, and so are in some sense parallel, they are different internally: be opening the British Parliament is a progressive VP, deliver a speech an unmarked-aspect VP.  Now though it would be lunatic to require in general that conjuncts be internally parallel (see my extended attack on this idea here), sometimes the internal composition of the conjuncts does make a difference.  The trick is to figure out when and how.

I'm a linguist, so I play with the variables and see what happens.  First, I note that other progressive+plain examples are also dubious:

I must be going soon and deliver a speech.

I have been going to Paris and gone to Vienna as well.

but that the reverse ordering is much less jarring:

I will deliver a speech in Antwerp and be going to Brussels soon after.

I have gone to Vienna and been going to Paris as well.

I've jiggled the context a bit to improve these, but the important points are (a) that it's incredibly hard to improve the progressive+plain examples much by jiggling the context, and (b) the reordered versions are hugely better than the originals.  Similarly, compare the original QE example with this (improved) variant:

Queen Elizabeth will open the British Parliament ... and then be delivering a speech.

So, yes, ceteris paribus, we'd often prefer parallel conjuncts to be parallel internally.  BUT there's something else going on.

The extra thing is a processing strategy, an analogue to what's known in the trade as Low Attachment: when a modifier follows a modified XP ending in an XP, the default is to interpret the modifier as attached to the structurally lower of the two XPs.  So, for example, confronted with a NP of the form

NP1  [ P NP2 ]  [ P NP3 ]

without any information about the content of these expressions, most people will take [ P NP3 ] as modifying NP2 rather than NP1, that is as having the structure

NP1  [ P [ NP2 [ P NP3 ] ]

rather than

[ NP1  [ P NP2 ] ]  [ P NP3 ]

Advice manuals warn you against high-attachment structures, telling you that modifiers MUST be next to the things they modify, and they supply what they take to be dire examples, like this one from Richard Lederer:

An ethnically diverse crowd of about 50 gathered at the Falkirk Mansion in San Rafael yesterday for a speakout against hate crimes organized by the Marin County Human Rights Roundtable.

(Lederer understands the sentence to be saying that it was it hate crimes, rather than the speakout against them, that the Rountable organized.)

But in the real world, low attachment is not a rule but a (default) preference, and context and real-world knowledge often favor high attachment (as indeed they should in Lederer's example).  Compare the low-attachment

an inventory of errors that were printed in the NYT

with the indisputably high-attachment

an inventory of errors that was larger than any previously published

I have a pile of high-attachment examples, and so do other people.  Most such examples pass by without notice, because their interpretation is clear in context.  There is no grammatical rule here.

Back to the original case, involving coordination (rather than modification).  The be going to segment in the QE example can be parsed as be followed by a coordination:

[ will ]  [ be [ [ going to ... ] and [ deliver a speech ] ] ]

(a kind of low attachment for deliver a speech), or it can be parsed with be going to as the beginning of the first conjunct:

[ will ]  [ [ be going to ... ] and [ deliver a speech ] ]

(a kind of high attachment for deliver a speech).

The first is the analysis I gave to the QE sentence at first, without reflection.  For whatever reason (psycholinguists, to your labs!), we very much prefer low attachment here -- but it just won't fly, because of the non-parallel conjuncts.  High attachment hard to get, low attachment not parallel.

The reason reordering the conjuncts improves things so much is that in the reordered coordinations there is no choice in the parsing of the be going to... segment, since it's in the second, and final, conjunct, which has to be taken, as a whole, to be in coordination with deliver a speech.  Any residual problems with the reordered coordinations presumably have to do with the semantics and pragmatics of progressive aspect and unmarked aspect.

Posted by Arnold Zwicky at November 16, 2007 04:57 PM