## Zubin Mehta violates the CSC

From the one-column interview of Zubin Mehta in the NYT Sunday Review on the 31st:

We [Parsees] have a Sunday dish called dhansak. That’s three or four different kinds of lentils, heavily spiced, eaten with brown fried rice with chicken in it. It’s something you eat and go to sleep.

Note the relative clause modifying something: (thatyou eat and go to sleep , with two VP conjuncts. The first conjunct is missing an object for eat; the standard generative-grammar story here is that the object of eat has been “extracted”, and surfaces as the head of the relative:

something-i you eat ____i

(where the index indicates reference and the underline after eat indicates the position of a “gap” representing the extracted material).

But the second conjunct has no gap — it’s just an ordinary intransitive VP — so the two conjuncts aren’t parallel in structure, and many people have found their coordination (though comprehensible) awkward at best, or even ungrammatical. The latter view was enshrined in the Coordinate Structure Constraint (CSC) in Haj Ross’s Ph.D. dissertation.

From John Lawler’s website, in a section on “Ross Constraints”, abstracted from the discussion in J. R. Ross (1967), Constraints on Variables in Syntax:

The Coordinate Structure Constraint:

Coordinate Structures are those joined with a Coordinating Conjunction, like and, or, but, etc. An element from one conjunct cannot be moved out of that structure:

• Bill cooked supper. What did Bill cook?
• Bill washed the dishes. What did Bill wash?
• Bill cooked supper and washed the dishes.
• *What-i did Bill cook ____i and wash the dishes?
• *What-i did Bill cook supper and wash ____i

(The “processual” metaphors in the terms move(ment), extract(ion), etc. are built into the formalisms of much generative grammar, but can be dispensed with in other formalisms.)

Lawler notes a series of cases in which, despite the way the examples look, the CSC does not apply. Things like the Mehta example are not on this list.

However, there is a considerable literature now in which it’s claimed that the CSC is not a condition on grammatical acceptability, but a consequence of sentence-processing strategies: on ths view, the problem with the “violations” is that the structures are hard to process, taking more work and extra time for comprehension. (Of course, such proposals need to be fleshed out to make them testable.)

On the wider topic of parallelism in coordination, see my Language Log postings here, on “failures of parallelism”, and here, on “coordination of unlikes”. Parallelism is not nearly as simple a matter as many people have thought.