Language Log

More risky RNR

Ben Zimmer ran across this puzzling coordination on the website for the University of Chicago's Workshop for the Anthropology of Latin America and the Caribbean, at the very beginning of an announcement for a workshop (on responding to the movie Apocalypto) this Wednesday:


This would appear to be a reduced coordination, of the sort known in the syntax literature as Right Node Raising (RNR).  The unreduced version would be:


RNR is versatile, but not versatile enough to treat a yes-no question and a WH question as equivalent structures.


RNR (last discussed here back in August) allows for coordinations of  the form


(where Z is a constituent, but X and Y are not both constituents) to have the right constituent Z "factored out", giving


So, for the coordination of VPs


we get the RNR version

(where bring liquids to and place liquids on are not constituents), which is the VP in


This is an acceptable reduced coordination.  (It's the acceptable version of the damaged RNR "Do not bring to, or place liquids on, this machine" in the August posting.)

On to academics engaging with popular culture.  The unreduced version seems to have the form


(omitting the if so, which is important for the meaning but not crucial for the structure).  So X is the inverted auxiliary do; Y is the fronted WH word how in combination with the inverted auxiliary should (a combination that is certainly not a syntactic constituent); and Z is apparently a base-form clause, academics engage with popular culture.  If so, then the badness of the RNR version


is mysterious.

But there might be an explanation.  The English Subject-Auxiliary Inversion (SAI) construction has been the focus of quite a lot of literature over many years, and one of the issues surrounding it is the question of what the structure of inverted clauses is.  There are three possibilities (assuming that constituent structures are continuous):


Theoretical considerations have figured very prominently in the literature on the structure of SAI clauses.  Within some theoretical frameworks, only structure (1) -- the one I assumed at first in my discussion of the popular culture example -- is possible.  In other frameworks, what the correct structure is is an empirical question, and many writers have opted for structure (3) (which is, in fact, my choice).  The evidence in favor of structure (2) -- with Aux + Subj as a constituent of some novel type -- isn't zero, but it's not very compelling, so the matter pretty much comes down to a choice between (1) and (3).

And now RNR is relevant.  With structure (1), we have no obvious explanation for the problem in our original example.  But with structure (3), we have an immediate explanation: the "factored" material, in bold face, does not constitute a single syntactic constituent; instead, it's just two constituents in sequence, and the conditions for RNR are not satisfied.

Some other, invented, examples that seem to me as bad as the original:


(It hadn't occurred to me when I started writing up this one odd example that it might bear on the analysis of SAI.  This is an example of how "little" Language Log postings can turn into something more substantial.)

Posted by Arnold Zwicky at November 6, 2007 03:39 PM