Merge in the Mind-Brain

Essays on Theoretical Linguistics and the Neuroscience of Language

Naoki Fukui

• Publisher: Routledge
• Year: 2017
• ISBN:
  9781315442785
• Edition: 1
• Topic: Langues et linguistique
• Subtopic: Linguistique

This collection of nine papers brings together Naoki Fukui's pioneering body of work on Merge, the basic operation of human language syntax, from the two distinct but related perspectives of theoretical syntax and neurosciences. Part I presents an overview of the development of the theory of Merge and its current formulations in linguistic theory, highlighting the author's previously published papers in theoretical syntax, while Part II focuses on experimental research on Merge in the brain science of language, demonstrating how new techniques and the results they produce can inform the study of syntactic structures in the brain in the future. By combining insights from theoretical linguistics and neurosciences, this book presents an innovative unified account of the study of Merge and paves new directions for future research for graduate students and scholars in theoretical linguistics, neuroscience, syntax, and cognitive science.

Part I

Merge in the Mind

1 Merge and Bare Phrase Structure*

1 Introduction

Since Aristotle, language has been taken as a system of associating sound—or sign, as recent research has shown—and meaning over an infinite range. One of the most important discoveries in linguistic sciences is that this association is actually not direct, but rather is mediated by “structure” whose exact nature remains to be clarified by empirical investigations. Modern linguistics has identified certain fundamental properties of the “structure” of human language and the system that generates it. These properties can be summarized as follows.

1. (1) a. hierarchical structure
   1. b. unboundedness/discrete infinity
   2. c. endocentricity/headedness
   3. d. the duality of semantics

There is a fair amount of consensus by now that these are the properties that ought to be captured, in one way or another, in any linguistic theory that aims to explain the nature of human language. The questions are: How much mechanism is needed to account for these properties elegantly, and is it possible to figure out what is behind these properties?

The rest of this chapter is organized as follows. In section 2, I briefly review the history of modern linguistics (particularly generative linguistics), to see how these characteristics have been captured by various different components of grammar. In section 3, I focus on the operation Merge, which is assumed in bare phrase structure theory to be the fundamental operation in human language, and discuss its properties and problems. In this section I also explore a few different interpretations of Merge and related operations (if such operations exist), and discuss some implications for comparative syntax, particularly Japanese syntax. In the concluding section (4), I summarize the discussion, trying to figure out the current stage of our understanding of the relevant issues, and speculate on future directions. Throughout the discussion, I confine myself to those issues directly related to phrase structure theory, particularly bare phrase structure theory. Accordingly, I cannot pay sufficient attention to various other important problems of minimalism that may in principle be related to the issues at hand. The reader is referred to introductory books such as Hornstein et al. (2005) for a more comprehensive discussion on minimalist syntax at large, in which the following discussion is couched.

2 A Brief History¹

Let us consider (1a) “hierarchical structure” first. That linguistic expressions have abstract hierarchic structures, not merely sequences of words and formatives, is one of the fundamental discoveries of modern linguistics. This discovery goes back to pre-generative structural linguistics, particularly in the form of “Immediate Constituent (IC)” analysis (Wells 1947). IC analysis is couched in the “procedural” approach developed in (American) structural linguistics, and as such cannot be carried over to the theory of generative grammar, which explicitly denies the procedural approach (see, e.g., the introduction to Chomsky 1955/1975). However, the insights of IC analysis, along with important concepts drawn from historical phonology—the concept of “ordered rewriting rules” in particular—can be incorporated into the theory of phrase structure grammar. The theory of phrase structure grammar is developed on the basis of Post’s combinatorial system (Post 1943), with an important modification regarding the notion of “vocabulary” (the terminal vs. non-terminal distinction), and is a set of rules (phrase structure rules) of the following form, where A is a single symbol and X, Y and Z are strings of symbols (Z non-null; X and Y possibly null):

1. (2) XAY → XZY

Phrase structure rules express the basic structural facts of the language in the form of the “P(hrase)-markers” they generate, with terminal strings drawn from the lexicon. P-markers generated by phrase structure rules express three kinds of information about a linguistic expression:

1. (3) a. the hierarchical grouping of the “constituents” of the structure (Dominance);
   1. b. the “type” of each constituent (Labeling);
   2. c. the left-to-right order (linear order) of the constituents (Precedence).

Thus, the specific kind of hierarchical structure of a linguistic expression (i.e., the labeled hierarchic structure), along with how the elements are stringed (linear order), is explicitly expressed by phrase structure grammar generating a set of P-markers.

In (2), X and Y need to be non-null, when the environment in which A is to be rewritten as Z needs to be specified. This situation arises when a lexical item is inserted into a particular terminal position of a P-marker. This type of “lexical insertion” is abolished in favor of the lexicon with subcategorization features (Chomsky 1965). The separation of the lexicon from the computational system—phrase structure grammar (PSG)—makes it possible to simplify the form of phrase structure rules for human language from the context-sensitive rule (2) to the context-free rule (4).

1. (4) A → Z

In (4), A is a single non-terminal symbol, and Z is either a non-null string of non-terminal symbols or the designated symbol “Δ” into which a lexical item is to be inserted in accordance with its subcategorization features (see Chomsky 1965 for details).

Thus, context-free phrase structure grammar (coupled with the lexicon) is responsible for expressing the properties of phrase structure of human language, particularly its labeled hierarchic structure with the designated left-to-right linear order. Property (1a) is thereby accounted for.

Let us skip properties (1b) and (1c) for the moment, and consider property (1d) next. This property, the duality of semantics, calls for a device other than phrase structure grammar. The duality of semantics refers to the fact (as has been noticed and studied from various points of view over the years) that generalized predicate-argument structure is realized in the neighborhood of the predicate (within the core part of a clause), whereas all other semantic properties, including discourse-related and scopal properties, involve an “edge” or a “peripheral” position of a linguistic expression (generally a sentence). This duality, particularly the latter fact, requires a device that relates two non-sister positions in the structural description of a sentence, i.e., a device that refers back to some earlier—not necessarily immediately preceding—step in the phrase structural derivation. However, a reference to constituent structure (i.e., to the past history of a phrase structural derivation) cannot be neatly expressed by context-free phrase structure grammar (see, e.g., Chomsky 1957). Thus, a new grammatical device has to be introduced to deal with the duality of semantics, and the notion of “grammatical transformation” is introduced for this and related purposes.²

Human language clearly exhibits the property of discrete infinity (1b), taken to be the most elementary property of the shared language capacity. Language is discrete, as opposed to dense or continuous, roughly in the sense that linguistic expressions are constructed on distinct and separate units (rather than continua), so that there are n word sentences and n+1 (or n-1) word sentences, but there are no n.5 (or n.3, etc.) word sentences (just like natural numbers). And language is infinite, since there is no n (in any human language) such that n is the number of words contained in the longest sentence (so that a sentence with n+1 words is a non-sentence). Most important cases of discrete infinity exhibited by human language are handled by special types of transformations—“generalized transformations”—in the early theory of transformational-generative grammar. These transformations are equipped with the function of embedding a structure (typically a sentence) into another structure of the same type (a sentence). With the abolishment of generalized transformations in the Standard Theory in the 1960s (Chomsky 1965), this function of (self-)embedding is transferred to phrase structure grammar with “recursive symbols” that appear both on the left-hand and right-hand sides of the phrase structure rules, allowing a kind of non-local recursion.

Toward the end of the 1960s, it became apparent that certain important generalizations about the phrase structure of human language, i.e., endocentricity/headedness (1c), cannot be stated in terms of phrase structure rules alone (nor in terms of transformations for that matter). Phrase structure in human language is generally “endocentric,” in the sense that it is constructed based on a certain central element—called the “head” of a phrase—which determines the essential properties of the phrase, accompanied by other non-central elements, thus forming a larger structure.³ This is the right intuition, but, as pointed out by Lyons (1968), the theory of phrase structure grammar cannot capture this. Phrase structure rules are too permissive as a theory of phrase structure in human language, in that they overgenerate phrase structures that are never actually permitted in human language, i.e., those structures that are not headed (“exocentric” structures). We thus need some other mechanism which correctly captures the endocentricity/headedness of phrase structure that appears to be a fundamental property of human language. X-bar theory is introduced mainly for this purpose.

The basic claims of X-bar theory, as it was introduced in Chomsky (1970), can be summarized as follows.

1. (5) a. Every phrase is headed, i.e., has an endocentric structure, with the head X “projecting” to larger phrases.
   1. b. Heads are not atomic elements; rather, they are feature complexes, consisting of primitive features.
   2. c. Universal Grammar (UG) provides the general X-bar schema of the following sort, which governs the mode of projection of a head:
      X’ → X …
      X” → [Spec, X’] X’

The version of X-bar theory presented in Chomsky (1970) was in a preliminary form, and numerous refinements have been made since then. However, it is also true that all the crucial and fundamental insights of X-bar theory were presented in this study, and have been subjected to few substantive changes since. While claim (5c), the existence of universal X-bar schema, has been subjected to critical scrutiny in recent years, claims (5a) and (5b) have survived almost in their original forms throughout the ensuing development of grammatical theory, and are still assumed in the current framework (but see the following for a recent proposal to eliminate the notion of projectio...