3 On the primitive operations
of syntax*
Takaomi Kato, Hiroki Narita, Hironobu Kasai,
Mihoko Zushi and Naoki Fukui
1 Introduction
Human language in its essence is a system that generates an infinite set of
hierarchically structured syntactic objects (SOs) which in turn can be used to
fulfill various cognitive needs, such as to relate “meaning” and “sound” (the
“Conceptual-Intentional” (CI) and “Sensorimotor” (SM) systems, respectively).
It has been widely assumed in the literature that the simplest characterization
of the relevant computation builds upon the notion of Merge, a combinatory
operation that takes n (often two) SOs (lexical items or composite phrases) and
constructs an unordered set of these objects: Merge(Σ 1 ,.. ., Σn) = {Σ 1 ,.. .,
Σn}. We can safely assume that iterative application of Merge is sufficient to
yield an infinite set of hierarchical SOs (Chomsky 2004 et seq.). Thus, it has
been asked in various manners whether we may ultimately conclude that Merge
is all we need to stipulate in the theory of human language (cf. Boeckx 2009,
Berwick 2011, Fujita 2013). In so doing, the literature seems to share the
conclusion that Merge counts as a “virtual conceptual necessity,” and hence
may be the most elementary and the most basic operation of syntax.
The purpose of this chapter is to challenge this established assumption. We
will put forward the hypothesis that Merge is not the most elementary syntactic
operation, and that it should rather be understood as a composite of two primi-
tive operations, i.e., (i) the selection of n elements, α 1 ,.. ., αn, from a designated
domain of computation, and (ii) the formation of an unordered set of these n
elements, {α 1 ,.. ., αn}. We will argue that the proposed analysis is not only
required in terms of descriptive preciseness, but also favorable on several empiri-
cal grounds. Among other things, we will argue that the relevant operations (i)
and (ii) can be further generalized to capture a number of linguistic “relations”
that Merge alone as usually construed cannot capture, including Agree(ment),
chains formed by Internal Merge (IM), binding, and also headedness (or label-
ing; cf. Chomsky 2005, 2007, 2008).
This chapter is organized as follows. In Section 2, we will review the notion
of Search discussed by Kato et al. (2014), which was proposed as a way to unify
the otherwise disparate relation-formation operations, including Agree, chain-
formation, and binding. We will show that the relevant notion of Search can
be decomposed into two more primitive operations: the first operation, which