Coordination and Ellipsis, Part 1
Posted by Neal on August 23, 2008
So before I show how Beavers & Sag’s analysis of coordination works for the Dark Knight coordination, it’s worth showing how it handles ordinary coordination. They start off with an assumption about the structure of coordinated phrases. I’ll illustrate with the example the good, the bad, and the ugly. Here’s one structure you might assign to this phrase:
This structure takes the entire coordinated phrase to consist of four chunks, one being the conjunction and, and the other three being the three coordinated noun phrases. If there were N items in the coordination, there would be N+1 branches coming off the top node (or more, if you had more conjunctions, as in the good and the bad and the ugly). This is not the structure Beavers & Sag (and other syntacticians) believe that coordinate structures have. Their view is that a coordinate structure has only two branches at the top level, regardless of how many items are actually coordinated. The left branch is the first item (the good), and the right branch leads to the rest (the bad and teh ugly). This branch can then split off into two more branches, the first representing the first of the rest (the bad), and the second branch leading to the rest of the rest (and the ugly). This branching (or more specifically, binary branching) can go as deep as necessary. Under this view, the structure for our example would be:
At this point, we need to have some way of making sure that a coordinate structure actually has a conjunction in it. Otherwise, the rule that says you can string phrases of like category together to get a coordinate structure like we’ve done here would also approve stuff like *the good, the bad, the ugly. This would actually be OK in languages that don’t require conjunctions in their coordinate structures, but English isn’t one of those languages. (Yes, in some situations you can leave out the conjunction, a phenomenon known as asyndeton, but that doesn’t happen in normal, everyday-register English, which is what this analysis covers.) The way Beavers and Sag enforce this requirement is to say that in a coordinate structure, the second element needs to have a label that says, in essence, “I’m the second element in a coordinate structure!” I’ll use the subscript CRD:YES to do that. They also stipulate that all phrases by default have a label that says, “I’m not the second element in a coordinate structure!”, which I’ll note with a CRD:NO subscript.
How does the second element get a CRD:YES label? There are only two ways. The first way is to be glued to a conjunction, which takes something that’s CRD:NO and with it forms a phrase that’s CRD:YES. Thus, and the ugly is CRD:YES, and can combine with the bad, which is CRD:NO, to form the coordinate NP the bad and the ugly.
But how does the chunk the bad and the ugly get its CRD:YES label, since it’s not attached to an and? Now we come to the second way of being labeled CRD:YES. Unlike other phrases, a phrase consisting of a CRD:NO and a CRD:YES does not automatically have a CRD:NO label. It could be CRD:NO or CRD:YES, depending on how it’s used, the same way that the can be singular or plural, depending on whether it appears with a singular or a plural noun. So when we use the bad, and the ugly as the second element in a coordinate structure, it will resolve to CRD:YES.
The next part of Beavers and Sag’s account is where the ellipsis comes in. Let’s say the good, the bad, and the ugly are part of a larger phrase: The good, the bad, and the ugly came to my party. Under their analysis, this is really the following coordination:
Their rule regarding ellipsis in coordinate structures is that if the same material appears on the right side of the first element and the second element, it can be omitted from the first one. So put together the bad came to my party with and the ugly came to my party and you end up with the bad and the ugly came to my party at the next level up. Combine that with the good came to my party, and you end up with the end result, the good, the bad, and the ugly came to my party.
Their schema has a mirror-image treatment for material that appears on the left side of both elements: When that happens, you can omit the identical material from the second one. So if you were to combine I love the bad with and I love the ugly, you could omit the I love from the second chunk and end up with I love the bad and the ugly. Combine that with I love the good, and you could end up with I love the good, the bad, and the ugly. There can even be coordinations where material from both sides undergoes ellipsis, such as Veeblefester cooked, ate, and threw up a pizza, where Veeblefester is omitted from the parts containing ate and threw up, and a pizza is omitted from the parts containing cooked and ate.
So how does this analysis handle multiple-level coordinations of the Be pompous, obese, and eat cactus variety, as well as the bought, bullied, reasoned, or negotiated with example from The Dark Knight? This is where their analysis gets really clever. I’ll save those details for the next post, but the key is this: Although in our example, it was the same material undergoing ellipsis at both levels of the tree (i.e. came to my party), it doesn’t have to happen that way.