In the Language Log there was a recent article about a syntactic form the author didn't understand. My analysis is in one of the comments, and I also wanted to preserve it here:
This sentence is fully grammatical and semantically valid in my native dialect. Let's isolate the most important elements.
"In a recent survey, more people liked Pepsi than they did Coke."
I take this sentence to be equivalent to:
"The percentage of people in a recent survey who expressed a preference for Pepsi exceeded the percentage of people in that survey who expressed a preference for Coke."
The less interesting (to me) change in the translation is from "like" to "prefers." I see that the sentence takes for granted an assumed model of how the survey works.
The assumed model of the survey in this example is that of a taste test where the person indicates a preference between the two drinks. This is a little different than the original example where the assumed model is that the survey gives the person a chance to characterize each group of ads, and that one or more of the responses counts as negative. In both cases, if the hearer were to take a different model of the way the survey works, the derived meaning changes to match the model. If the models of the speaker and hearer do not match, a misunderstanding is likely.
A more interesting change to me involves the verb "did." Did what? In this example, the predicate is "to express a preference for
" and the phrase "...did Coke" uses a structure where the predicate is the same and "Coke" fills the valence . In the original example the predicate is "found to be negative." This works almost the same, except it adds the extra twist that "Mr. Kerry's" is a shortening of "Mr. Kerry's ads" in context.
That leaves the frustrating "they." I we treat "they" like a pronoun with a rigid referent then "...they did Coke" would have "they" refer to the subset of people on the survey who liked Coke. The problem is that if we believe that the rigid reference is drawn from the context, the only compatible choice in the context is "More people" which refers to the people who liked Pepsi --- exactly the opposite of the people who liked Coke.
My theory is that the same structure which allows the word "Coke" to replace the word "Pepsi" in the predicate (or "Mr. Bush's ads" with "Mr. Kerry's [ads]") also affects the interpretation of "they." The syntax of the predicate form requires a survey subset as a subject, and the semantic of the fully qualified predicate selects which subset.