Concrete Minimalism, Branching Structure and Linear Order

Peter W. Culicover
The Ohio State University

	A cornerstone of modern generative grammar is the notion of phrase
structure. Traditionally, phenomena such as movement, deletion, and scope,
and to a lesser extent coordination, have been taken as evidence about the
organization of phrase structure, and about the constituents that make it
up.  While early work on phrase structure suggested that multiple branching
structures are possible, a simplifying assumption of later work was that all
branching structure is binary.  This assumption has made it possible to
develop a range of analyses in which hierarchical structure is the
foundation for the basic relationships between parts of a sentence.
	In this paper I will consider the question of whether it is possible to
dispense with certain aspects of branching structure in syntactic
representations.  In particular, I will be concerned with the branching
structure that is central to the Antisymmetry Thesis (AT) due to Kayne 1995,
namely the structural relationship between heads and maximal projections
that serve as arguments and adjuncts to the heads.  The AT accounts for
linear order in terms of asymmetric c-command, which is defined only when
there is binary branching.  In the absence of binary branching, no linear
order is defined, and thus it follows that all relevant syntactic structure
is strictly binary branching.
	I will argue on minimalist grounds that the fundamental syntactic asymmetry
in language is not one based on hierarchical structure, but on linear order.
On this view, the hierarchical structure is a reflex of the linear order and
the correspondence with Conceptual Structure (CS).
	The minimalist perspective calls into question much of what has been taken
for granted in earlier theories of grammar.  There are varieties of
minimalism that must be distinguished, however.  Minimalism as a general
approach is simply good science - the most parsimonious account of the
phenomena.  If we are going to be minimalists, we cannot choose what parts
of the theory are going to be minimalist and what parts are not.  So we
cannot restrict ourselves to theories that have particular properties (e.g.
some economy metric on abstract steps in a derivation, as in Chomsky 1995,
or some particular degree of abstractness), or remain fixed on some commonly
held but never empirically validated assumption, without considering the
adequacy of more parsimonious alternatives.
	The variety of minimalism that I will pursue is in some sense the most
radical one.  On this view, which I call Concrete Minimalism, I take the
perspective of the language learner, who is exposed minimally to information
about linear sequences of sounds and their corresponding meanings, which I
will assume here to be Conceptual Structure (CS) in the sense of Jackendoff
1990.  I take these to be concrete manifestations of language, as
distinguished from abstract syntactic representations.  From the perspective
of the language learner, linear order and conceptual structure are
epistemologically prior to branching syntactic structure that is not
directly projected from CS.  It is evidence about linear order and CS that
the learner is presented with in language learning, and it is on this basis
that the learner would form hypotheses, even hypotheses about richer
structure, if such structure existed.   The minimalist perspective requires
that any additional structure, levels of representation, notational devices,
etc. beyond the two concrete ones just enumerated must be justified on the
basis of compelling empirical evidence.  In the case of branching structure,
we must ask what work additional branching structure does, or whether linear
order is sufficient to the task of explaining the facts.  While it is true
that structure or linear order may be taken to be primitive in a given
syntactic theory, linear order is a weaker notion than branching structure
in terms of expressive power, and hence should be preferred, other things
being equal.
	The structure of this paper is as follows.  First I review very briefly the
historical arguments for binary branching structure, arguing that the
primary motivations have been theory-internal, not factual.  Then I focus on
a range of empirical evidence that has been adduced in the literature to
demonstrate that branching is richer than the minimalist flat branching
structure.  I will argue that to the extent that there is evidence for
richer structure, it is actually evidence for CS, not syntactic structure.
There are of course cases in which the branching is binary, but there are
those in which there is multiple branching.  The evidence comes from such
constructions as VP ellipsis, VP anaphora, gapping and pseudo-gapping, bare
argument ellipsis, right node raising, conjunction reduction and VP
topicalization.  The core argument is that the antecedents in these
constructions are not always syntactic constituents, although they do appear
to correspond to CS objects.  A typical example is a sentence like

(1)	Mary will cook the potatoes for 15 minutes in the morning, and Susan
will for 25 minutes.

where the antecedent for the ellipsis can be "cook the potatoes in the
morning", which is not a constituent of the left conjunct.  Differences
between argument and adjunct PPs with respect to ellipsis (e.g., Mary
*put/read the book in the kitchen and Susan did in the living room) are not
evidence for differences in structure, but in interpretation, which is
represented as CS.  Evidence from right node raising (due to Postal)
suggests that there are cases where the structure may be flat up to IP;
e.g., John cooked and Mary ate the same kind of potatoes. It appears in this
case that "John cooked and Mary ate" is predicated of the same kind of
potatoes, as suggested by Steedman (1997) for the general case of RNR.
	I then turn to Phillips' proposal for dealing with the apparent
indeterminacy of constituency tests (Phillips to appear).  Phillips proposes
that syntactic structure built in the course of left-to-right processing is
available for certain relations, and that it may subsequently be destroyed
and reconstituted, in which case a different structure for an already
processed part of the sentence can serve as input to other relations.  I
argue that Phillips' approach cannot deal with the fact that the
constituency tests in question are not tests for syntactic constituency, but
CS constituency, and so in part is a solution to a problem that does not
exist.
	 To summarize, minimalist principles and in particular Concrete Minimalism
would lead us to assume that syntactic structure is flat in the absence of
compelling evidence to the contrary.  There is in fact evidence that
supports this conclusion.  One important consequence is that the
Antisymmetry Thesis cannot be a correct generalization about linguistic
structure, since it presumes linguistic structure that does not exist.  It
is preferable to take linear order to be the primitive asymmetry and to
treat structural asymmetry as following from and not determining linear
order.