Coordination and Ellipsis, Part 2
Posted by Neal on August 24, 2008
OK, after giving the background, I’m finally ready to show how Beavers & Sag 2004 handle multiple-level coordination. First we’ll do one of the kinds typified by Be pompous, obese, and eat cactus that most of the entries in the “Multiple-level coordination” category are about, and which they specifically intend their analysis to cover. Then we can move on to the one from The Dark Knight: These people can’t be bought, bullied, reasoned, or negotiated with.
We’ll start building the tree diagram for Be pompous, obese, and eat cactus from the bottom up, starting by coordinating be obese and eat cactus. In the diagram on the right, you can see that and attaches to the imperative eat cactus, which is CRD:NO, to form the phrase and eat cactus, which is CRD:YES. Since this phrase is CRD:YES, it is eligible to be the second chunk in a coordinate structure, if we can find a similar-category phrase that is CRD:NO. Be obese is another imperative, and it’s CRD:NO, so the two can form the coordinate phrase Be obese and eat cactus, which can be either CRD:YES or CRD:NO, as circumstances require. Notice that no ellipsis has gone on; everything in the individual imperatives (be obese, eat cactus) appears in the coordination of the two.
Now let’s put together be obese and eat cactus with Be pompous. At this point we can ignore the structure underneath that top Imper node in the tree we just looked at. It doesn’t matter at this point how it got put together; all that matters is that we have the phrase be obese and eat cactus to coordinate with Be pompous. I’ll do that in the new tree below. 
We can’t just put these two phrases together and call the result a coordinate structure. For that to work, the first one has to be CRD:NO, which it is, and the second one has to be CRD:YES. The way we left it at the lower level of the coordination, it could be either, so at this upper level, it can be CRD:YES. At this point, we have two phrases in a coordinate structure, and each of them starts with be. So the second be can undergo ellipsis, which I’ve indicated by putting it in parentheses. It doesn’t matter that the be wasn’t the same thing that underwent ellipsis at the lower level (i.e., nothing).
Moving on to the Dark Knight coordination, I’ll simplify it by just doing bullied, reasoned, or negotiated with. At the lower level, we’ll coordinate reasoned with or negotiated with. This time it’s an or that gets together with a CRD:NO phrase (negotiated with) to form a CRD:YES one. This is coordinated with reasoned with. We now have a coordinate structure where each element ends with the same thing: with. Therefore, the first with can undergo ellipsis, indicated with parentheses.
Now let’s finish off with the top level of the coordination, with the fresh tree below. 
Beavers and Sag’s analysis won’t handle every kind of coordination. For example, ellipsis won’t cover sentences like Doug and Adam are my two sons, since it’s not equivalent to Doug is my two sons and Adam is my two sons. But it does unify several kinds of coordination in a single analysis, and it’s a good sign that it can also handle a previously unnoticed (as far as I know) kind of coordination, with no changes necessary.
Literal-Minded 
Glen said
These non-parallel coordinations aren’t acceptable in my personal grammar, and I think you’ve said they aren’t allowed in yours, either. And I imagine other structural models of English grammar, unlike Beavers & Sag’s, would rule them out. But I’m curious to know if there’s a *tweak* to Beavers & Sag’s — say, one additional constraint — that would do the same thing. In other words, could you use Beavers & Sag’s model to describe both people who find these coordinations acceptable and those who don’t?
(To be clear: I know that parallel constructions would fit just fine in Beavers & Sag’s model, but I’m wondering if there’s a modification of their model that would cleanly rule out the non-parallel ones.)
Neal said
Well, once I was sufficiently sensitized to deviations from parallel structure to notice multiple-level coordinations, I judged them unacceptable. OTOH, Be pompous, obese, and eat cactus went for years unremarked, sounding perfectly fine to me. Both the Ridger and Ingeborg Norden find Women who are nursing, pregnant, or may become pregnant ungrammatical, but the Ridger seems to be OK with bought, bullied, reasoned, or negotiated with, saying in a comment that putting another or before reasoned makes the sentence sound stupid. My current opinion is that multiple-level coordinations are generated (that is, are not accidents) in many people’s grammar, though not everyone’s. For those who find them ungrammatical, the tweak to B&S’s rule would be that the missing material for each level of the coordination does have to be the same.
The Ridger said
I’m not so much “OK with” it as feeling that if you want to coordinate those four things, that’s the only way to do it. Maybe I just find deleting the extra preposition more acceptable than essentially inserting a verb?
Multiple-level coordination « Arnold Zwicky's Blog said
[…] If you think about MLCs in descriptive terms, then you notice other things, which have been brought out in Whitman’s postings and in papers on coordination, in particular a paper by Beavers and Sag that Whitman discusses in some detail (notably, here and here). […]
Multiple-Level Coordination with And and But « Literal-Minded said
[…] previous posts, I’ve summarized the analysis of Beavers & Sag (2004) for this kind of coordination, and […]