1[To appear in Linguistics and Philosophy ]
Fragments and ellipsis
Jason Merchant
University of Chicago
October 2003
ABSTRACT: Fragmentary utterances such as ‘short’ answers and
subsentential XPs without linguistic antecedents are proposed to have
fully sentential syntactic structures, subject to ellipsis. Ellipsis in these
cases is preceded by A'-movement of the fragment to a clause-
peripheral position; the combination of movement and ellipsis
accounts for a wide range of connectivity and anti-connectivity
effects in these structures. Fragment answers furthermore shed light
on the nature of islands, and contrast with sluicing in triggering island
effects; this is shown to follow from an articulated syntax and the PF
theory of islands. Fragments without linguistic antecedents are argued
to be compatible with an ellipsis analysis, and do not support direct
interpretation approaches to these phenomena.
The situations in (1)-(3) and Ben’s utterances that close them pose a fundamental challenge for
standard linguistic theories of the form-meaning relation.
(1) Abby and Ben are at a party. Abby asks Ben about who their mutual friend Beth is
bringing as a date by uttering: “Who is Beth bringing?” Ben answers:
“Alex.”
(2) Abby and Ben are at a party. Abby sees an unfamiliar man with Beth, a mutual friend of
theirs, and turns to Ben with a puzzled look on her face. Ben says:
“Some guy she met at the park.”
(3) Abby and Ben are arguing about the origin of products in a new store on their block, with
Ben maintaining that the store carries only German products. To settle their debate, they
walk into the store together. Ben picks up a lamp at random, upends it, examines the label
(which reads Lampenwelt GmbH, Stuttgart), holds the lamp out towards Abby, and
proudly proclaims to her:
“From Germany! See, I told you!”
In each of the above three situations, a linguistic utterance is preceded by some discourse
context, which may (as in (1)) but need not (as in (2) and (3)) include linguistic material. The
linguistic utterance in each case has two properties, which, taken together, give us cause for
alarm. First, the phonetic signal that each of the above utterances gives rise to corresponds to the
lexical content of a DP (as in (1) and (2)) or a PP (as in (3))—in short, to a node which is
syntactically non-sentential. Second, (1)-(3) have the same conversational function as (4)-(6)
respectively; they can be used to advance the purposes of the discourse in the same way as fully
sentential utterances—that is, they can have the same propositional content and assertoric force
as utterances of what are uncontroversially fully sentential syntactic structures.

2(4) Beth is bringing Alex.
(5) He’s some guy she met at the park.
(6) It’s from Germany.
Call linguistic expressions such as those found in (1)-(3) fragments. The question that
arises with respects to such fragments, then, is the following: how can we account for the
semantically propositional character of what appear to be syntactically less than sentential
structures? This question arises, of course, on the usual assumption that syntactically sentential
objects like TPs give rise to semantic objects of propositional type which can be used to make
assertions, but that syntactic objects like DPs and PPs do not. In other words, does the
propositional content of (1)-(3) come from (i) the usual mechanisms (an interpreted syntactic
structure, i.e., an LF structure), or (ii) a novel method of generating and interpreting such
fragments?
The answer to this question has obvious and fundamental consequences for the
architecture of the grammar of human languages. If possibility (i) is correct, the grammar must
include a new kind of ellipsis operation, one with properties that appear to be quite distinct from
the kinds of ellipses that are, at this point, fairly well understood, such as VP-ellipsis and
sluicing. If possibility (ii) is right, then we must allow non-sentential syntactic objects either to
be able to be denote propositions, or we must allow the non-propositional semantic objects they
give rise to to be able to be used make assertions (further, under some assumptions, we may also
need to propose new ways of building syntactic structures).
Of these two possibilities, it seems clear that the first is a more conservative one, in
preserving the usual mapping of syntax and semantics and beyond but in placing the burden on
the syntax, in particular on the non-pronunciation of certain syntactic structure. Call the first
approach the ellipsis approach; the ellipsis approach has been pursued since the earliest attempts
in generative grammar to account for these phenomena (Hankamer 1979, Morgan 1973, 1989, and
more recently Stanley 2000, Reich 2002, 2003, Brunetti 2003, Ludlow to appear).
The second approach is more radical in that it requires a revision of the systems of form-
meaning mappings. Call the second approach the direct interpretation approach; this approach
has been proposed in various forms by a number of researchers (van Riemsdijk 1978, Hausser
and Zaefferer 1978, Yanofsky 1978, Carston 2002, Ginzburg and Sag 2000, Jackendoff 2002, and
especially Barton 1990, Stainton 1995, 1997, 1998, to appear, the latter of whom introduces data
on which the examples in (2) and (3) are modeled). These modern researchers follow a venerable
tradition in linguistic theorizing in trying to account for the properties of fragments without
relying on ellipsis; in fact, theorists in the early twentieth century were already upbraiding their
predecessors for a perceived overreliance on ‘ellipsis’ as an explicator. Bühler 1934:155, for
example, called ellipsis the ‘alte crux der Sprachwissenschaftler’ (‘the linguists’ old crutch’),
while Sütterlin 1907:9 maintained that ‘nach unserer heutigen Betrachtungsweise [liegt] eine
wirkliche Auslassung viel seltener vor als nach der Auffassung der früheren Zeit’ (‘on
contemporary views, a true omission occurs much less frequently than was earlier believed’).
Jespersen 1924:306 had these choice words for skeptics of his version of the direct interpretation