(4) Universal endocentricity:
As a result of (2)-(3), every SO is endocentric, leaving no room for non-
endocentric structures.
(2)-(4) have been very widely presumed throughout the history of generative
linguistics (see Fukui and Narita 2014 for an overview). However, it should be
noted that these hypotheses are no longer necessary – nor are they innocuous – in
the contemporary theory of “Merge” (“bare phrase structure”) launched by
Chomsky (1995 et seq.). Merge is a simple set-formation operation that takes
n (typically two) SOs, Σ 1 ,.. ., Σn, as its input and generates an unordered
set {Σ 1 ,.. ., Σn} (but see Kato et al. 2016 for a further analysis of this
operation):
(5) Merge(Σ 1 ,.. ., Σn) = {Σ 1 ,.. ., Σn}
According to bare phrase structure theory, all linguistic expressions reduce to
set-theoretic objects generated via Merge. Structural descriptions provided by
Merge-based syntax are therefore quite different from those of earlier phrase
structure grammars (PSGs) or various versions of X-bar theory. Most notably,
no notions of projection or labels are represented in the output of Merge.
The stipulation of universal projection (2) therefore loses its ground, which
is in signifi cant contrast to the automatic and exceptionless generation of
projection/labels in X-bar syntax. Then, if we were to somehow incorporate
(2) into bare phrase structure, some additional stipulation would have to be
provided. To take a familiar example, Chomsky (1995) proposes to encode
the notion of label into the defi nition of Merge as in (6) (modulo a binarity
restriction, n = 2).
(6) Merge(α, β) = {γ, {α, β}}, where the label γ = the label of α or β.
In contrast, Chomsky (2004, 2008, 2013) argues for the simpler version of
Merge in (5), while hypothesizing that endocentricity/projection is determined
by an independent algorithm of “labeling.” See also Fukui (2011), who attempts
to decompose the two components of (6), with Merge as in (5) serving for
hierarchical organization of constituents and an operation called “Embed”
providing a label in a way similar to (6). However, whether we complicate the
defi nition of Merge as in (6) or propose a labeling algorithm independent of
Merge, those approaches count as departures from the simple Merge system
formulated in (5). Thus, as a natural move, linguists of minimalist persuasions
have begun to cast serious doubts on the notion of labels/projection (Collins
2002, Chomsky 2007, 2013, Narita 2014). For example, Chomsky (2007:23)
remarks that “reference to labels (as in defi ning c-command beyond minimal
search) is a departure from S[trong]M[inimalist]T[hesis], hence to be adopted
only if forced by empirical evidence, enriching UG.” Then, we should ask, do
we have to keep the hypothesis in (2) even in the bare phrase structure
10 Hiroki Narita and Naoki Fukui