Noah A . Smith . Ellipsis Happens , and Deletion Is How NOAH

I offer an overview of recent literature on ellipsis phenomena in natural language, including the most salient empirical arguments surrounding the classification of ellipsis as a phenomenon of the Phonological Form or of the Logical Form. I show that the majority of evidence supports the view of ellipsis as PF-deletion rather than LF-copying. I then give a theoretical argument that supports this hypothesis under the Minimalist Program. Introduction The phenomenon of missing phrasal constituents (known as “ellipsis”) has been noted from a variety of angles in recent years. It remains difficult to classify, as it appears to involve phonology (due to its similarity to deaccenting), syntax (by virtue of its distribution), semantics (evidenced by its apparent licensing conditions), and pragmatics (because of the cognitive load it imposes). This discussion shows how empirical and theoretical evidence lead to the conclusion that ellipsis is a phenomenon of overt syntax involving deletion. The first section gives an overview of evidence for the nature of ellipsis in current literature. The next section outlines an argument showing how a PF deletion account is necessitated under the Minimalist approach to the generative grammar. 1 Empirical Investigation In recent literature, various approaches have been taken to explain ellipsis phenomena. Such phenomena include VP ellipsis (1), empty IPs (2, 3), and pseudogapping (4). (1) George likes winning votes as much as Al does. (2) George tried to win the election, and Al also tried. (3) Somebody won the election, but we don’t know who. (4) George will select Dick, and Al will Joe. ELLIPSIS HAPPENS, AND DELETION IS HOW 177 Arguments over the derivational conditions underlying these phenomena, assuming that they can be accounted for derivationally, are essentially of two types. Some, e.g., Sag (1976), have argued for a PF deletion account of such phenomena (specifically, VP ellipsis). Under this account, such constructions would have a completed VP in the lower clause, and that VP would be eliminated at PF: (5) PF: George likes winning votes as much as Al does like winning votes. (6) LF: George likes winning votes as much as Al does like winning votes. Others who have explored the problem, e.g., Williams (1977), suggest that ellipsis constructions are generated with empty terminal nodes which are then filled in through a copy operation in the logical component. Under this account, (1) would have the following representations given the NP interpretation of “winning votes”: (7) PF: George likes winning votes as much as Al does [VP [V !] [NP [Adj !] [N ! ] ] ]. (8) LF: George likes winning votes as much as Al does like winning votes. One important fact about the phenomenon in general is that ambiguities in the antecedent phrase do not multiply in the elided phrase. In (1), the phrase “winning votes” could be interpreted as either an NP roughly equivalent to “votes which win (elections)” or the IP “PRO winning votes.” Yet whichever interpretation is given to the antecedent (George) VP, the elided VP will have the same interpretation. This holds as well in data involving pronominals. The PF-deletion approach, which is currently favored, presents a unique problem. Deletion of a phrase requires a relation between the deleted elements and some antecedent. The nature of this relation is clearly semantic, and therefore it is not apparent how information pertaining to the relation is available to the phonological component. 1.1 A Similar Phenomenon: Deaccenting Tancredi (1992) attempts to overcome this difficulty in the VP ellipsis case by drawing a parallel between a phonetically empty VP and a phonetically deaccented VP. VP deletion, he argues, is merely a strong form of deaccenting. He notes the similarity in the distributions 1 There are non-derivational accounts given for these phenomena, but the most recent evidence favors the derivational approach. This discussion shall consider one instance where an ellipsis site is argued to be a deep anaphor.