The Weekly Newsletter of MIT Linguistics

Ling-Lunch 11/10 — David Erschler

Speaker: David Erschler (University of Massachusetts Amherst)
Tittle: Predicting embedded gapping
Date/Time: Thursday, November 10/12:30pm-1:50pm
Location: 32-D461

I show that in a number of languages gapping can occur in embedded clauses. I argue that this provides support for a movement plus deletion analysis of gapping. The ability of gapping to be embedded in a given language depends on the height of the ellipsis-licensing feature and the availability of a landing site for moved constituents sufficiently high in the embedded clause.

Gapping is a construction, discovered and named by Ross (1970), where the finite verb is missing from the clause, (1). It is fairly common cross-linguistically.

(1) Some will eat beans and others will eat rice. Johnson (2009)

Famously, English and a number of other languages disallow gapping to occur in embedded clauses, Hankamer (1979) and the subsequent literature:

(2) *Some ate mussels, and she claims that others ate shrimp. Johnson (2009)

However, the ban on embedding is not universal: adding to the recent work on Persian by Farudi (2013), I show that embedded gapping occurs in several languages including Russian, Georgian, and Ossetic.

A number of accounts, starting from Jayaseelan (1990), assume that the material that survives gapping moves out of the constituent to be deleted. Some of these analyses, e.g. Aelbrecht (2007), Gengel (2013), and Farudi (2013), use the feature-based approach to ellipsis licensing.

I use the basic insight of earlier “move and delete” proposals and argue that, in languages that allow embedded gapping, it results from movement of surviving constituents and deletion of the XP that they moved from. A necessary condition for this to occur in embedded clauses is that landing sites are available for such movement. The size of the deleted constituent may vary cross-linguistically. The feature E that triggers deletion is hosted either by some head H, which is a priori either &, i.e. the head of the conjunction phrase, or a head within the clause where the gapping occurs. To trigger deletion, the feature must agree with the head whose complement is to be deleted.

If the licensing feature is located on &, it fails to agree with material in the embedded clause, for locality reasons. On the other hand, if the feature is located within the clause where ellipsis is to occur, gapping is predicted to be possible. I show that this analysis makes a number of correct predictions about languages with embedded gapping.

A wider implication of the findings is that an approach that operates in terms of licensing features, their location, and the size of constituents to be deleted is superior to taxonomic approaches to ellipsis.