Jump to ContentJump to Main Navigation
Biological Foundations and Origin of Syntax$

Derek Bickerton and Eörs Szathmáry

Print publication date: 2009

Print ISBN-13: 9780262013567

Published to MIT Press Scholarship Online: August 2013

DOI: 10.7551/mitpress/9780262013567.001.0001

Show Summary Details
Page of

PRINTED FROM MIT PRESS SCHOLARSHIP ONLINE (www.mitpress.universitypressscholarship.com). (c) Copyright The MIT Press, 2018. All Rights Reserved. An individual user may print out a PDF of a single chapter of a monograph in MITSO for personal use. Subscriber: null; date: 24 October 2019

Some Elements of Syntactic Computations

Some Elements of Syntactic Computations

Chapter:
(p.62) (p.63) 4 Some Elements of Syntactic Computations
Source:
Biological Foundations and Origin of Syntax
Author(s):

Rizzi Luigi

Publisher:
The MIT Press
DOI:10.7551/mitpress/9780262013567.003.0004

Abstract and Keywords

This chapter focuses on the nature of syntactic computations and current syntactic models, including “Principles and Parameters” and Minimalist models. It first considers the expression of the open-ended character of natural language syntax before turning to Minimalism, which has introduced an extremely simple structure building rule, Merge. Merge expresses a property called syntactic recursion, which allows the generation of an unlimited number of sentences, and builds hierarchical structures that may be altered by movement. After discussing the typology of movement processes, the chapter demonstrates how Merge and Move interact to determine basic word-order properties of natural languages. It also illustrates certain interplays between morphology and syntax before concluding with a discussion of invariance and variation and how the universality and variability of human language is expressed by parametric models.

Keywords:   syntactic computations, natural language, syntax, Minimalism, Merge, syntactic recursion, movement, morphology, invariance, parametric models

Abstract

This chapter focuses on some current directions of research on the nature of syntactic computations. I illustrate how certain issues have taken shape in the course of the last half century and discuss how they are addressed in current syntactic models, with special reference to “Principles and Parameters” and Minimalist models. The first issue concerns the expression of the open-ended character of natural language syntax. Minimalism has introduced an extremely simple structure building rule, Merge, which is able to reapply indefinitely to its own output, thus expressing syntactic recursion, a property permitting the generation of an unlimited number of sentences. Merge builds hierarchical structures, which may be modified by the other fundamental syntactic operation: movement. I discuss the typology of movement processes and illustrate how Merge and Move interact to determine basic word-order properties of natural languages. As is to be expected given the parsimonious, economy-based design of natural langue syntax, movement takes place to satisfy requirements of other components: phonology-phonetics, requiring full linearization of the hierarchical structures of syntax; morphology, requiring the formation of well-formed words; and semantics-pragmatics, dealing with the meaning and use of linguistic expressions. Focusing on verb movement in the functional structure of the sentence, certain interplays between morphology and syntax are illustrated which generate significant diachronic and comparative predictions. The final part of the chapter is devoted to the issue of invariance and variation, and to how the universality and variability of human language is expressed by parametric models.

Introduction

In this chapter I present current directions of research on the nature of syntactic computations. After illustrating the roots of these directions in classical work in generative grammar, I discuss current understanding of the issues. Coverage of topics is not exhaustive, the selection being based on my competence and taste. I will assume the basic conceptual structure of the Principles and Parameters framework, and will shape much of the presentation in terms of (p.64) the formalism and tools of the Minimalist Program. According to its general guidelines, Minimalism puts a strong emphasis on sticking to the bare minimum of the formal apparatus required to express the fundamental generalizations of syntax, and this effort of conciseness and simplicity seems to be particularly appropriate when posing questions as to the biological foundations and evolutionary origins of mechanisms.

Creativity and Recursion

One salient property of the human knowledge of language is the so-called creativity manifested in normal language use. We constantly produce and understand new linguistic objects, sentences that we had never encountered in our previous linguistic experience, and our linguistic capacities give us the possibility of expressing an indefinitely large number of messages. No other species possesses a communication system with such characteristics.

The importance of this creative aspect is not a new observation: it was clear, in essence, as early as in the seventeenth century. René Descartes pointed out that the capacity to organize words into an unlimited number of contextually appropriate sentences distinguishes the dumbest man from the most intelligent ape, and from the most sophisticated machine. What is the essence of this “familiarity of the newness,” and of the unbounded character of the human linguistic capacities?

What was missing until the middle of the twentieth century was a technical device to address these questions in a precise manner. In the first half of the century, structural linguistics conceived of language (saussurean “langue”) as a systematic inventory of linguistic signs, each of which consisted of a sound—meaning pairing, basically a theory of the lexicon. An inventory, however, is limited by definition; therefore, this approach was intrinsically unable to address the fundamental question of creativity, except through some vague notion of analogy: the infinite possible combinations of linguistic signs were, at least in part, relegated by Saussure to “parole” the actualization of the system of “langue” in individual linguistic acts, and “langue” basically contained a repertoire of frozen idioms, not productive syntax. Saussure was probably dissatisfied with this conclusion, as certain oscillations in the Cours de linguistique générale suggest (Saussure 1916/1985).1 Natural language syntax is (p.65) clearly regular, a rule-governed process, but linguistics at the beginning of the twentieth century did not possess a formal device to express this regularity.

Introducing such a device was Chomsky’s first critical contribution to the study of language. Chomsky showed that the core notions of the theory of recursive functions, developed in the study of the foundations of mathematics, could be adapted to language. A recursive procedure is one that can indefinitely reapply to its own output, giving rise to a hierarchical structure. The following examples illustrate some simple cases of recursion in natural language, in which a phrase of a given type can be embedded into a phrase of the same type (phrases are delimited by brackets); this can go on indefinitely:

  1. (1) The preface [of the first book [on the discovery [of the …]]]

  2. (2) I believe [that people wonder [whether Mary thinks [that someone said …]]]

  3. (3) I met [the boy [who bought [the book [which pleased [the critics [who wrote [the review …]]]]]]].

Notice that it is not the mere iterability of the procedure which makes it recursive, but its capacity to create a hierarchical structure, as happens at different levels of organization of linguistic structures. Thus, for example, the iteration of the motor program activated in walking, one step after the other, can go on indefinitely, but it does not give rise to any hierarchical structure. By contrast, the stringing together of words in a sentence does, determining the bracketed representations as in (1)–(3) or, more perspicuously, tree-like representations such as the ones we will consider below. Such representations, far from being mere artifacts of the adopted formalism, contribute crucially to determining properties of form and meaning of linguistic expressions.

It is sometimes said that recursion is not as critical to natural language syntax as the approach just introduced assumes, because the normal use of language, as emerging from corpora of ordinary conversation, typically consists of rather short sentences. This objection does not, however, take into account the fact that a system capable of producing very simple phrases, like John’s book or the picture of the girl, is already recursive, as it allows a nominal expression to be embedded within a larger nominal expression. Thus, a system capable of generating such simple structures already yields automatically the unbounded character of language. Imposing a limitation to sentences of a fixed length would complicate the system in an arbitrary and unwarranted manner: it would be about as arbitrary as defining the number system as ending with a particular number N on the basis of the observation that people use only fairly small numbers in everyday life.

Let us now try to express the role of syntax in a more comprehensive model of the human linguistic capacities. Phrasing things at a very general level, we can say that to know a language means to possess certain inventories of elements, somehow stored in memory, and the computational procedures to (p.66) combine the elements of the inventories to form entities of a higher order. This fundamental inventory is the lexicon and consists of two major systems of lexical items: (a) the contentive lexicon, consisting of nouns, verbs, adjectives, etc. (i.e., elements endowed with descriptive content characterizing events, arguments, qualities, etc.) and (b) the functional lexicon, consisting of grammatical words and morphemes such as determiners, complementizers, auxiliaries and copulas, expressions of tense and aspect (i.e., elements that have a more abstract semantic content and somehow define the configurational structure in which the contentive elements are inserted).2

There are other lists of elements which must be stored in memory, ready to be used in linguistic computations (e.g., features, phonemes, syllable structures, morphemes, idiomatic expressions) and which define the structure of the lexicon. When this cascade of levels leaves the lexicon and enters productive syntax, and we start putting words together, the computational procedures become recursive and give rise to higher-order entities, phrases, and sentences, which are indefinitely extendable.

More generally, language is sound with meaning (abstracting away from languages that use different modalities, such as sign language; the generalization to these cases is straightforward). Thus, a model of language must be able to connect representations of sounds with representations of meanings over an unbounded domain. The following structure is quite generally assumed in the tradition presented here:

  1. (4) Some Elements of Syntactic Computations

Items are selected from the functional and contentive lexicon and strung together through the recursive procedures of syntax. Hence, interface representations of sound (PHON) and meaning (SEM) are computed (also called Phonetic Form and Logical Form, respectively) and accessed by other systems: the auditory—articulatory systems and the conceptual-intentional systems. In this conception, syntax is the generative heart of the system, the device that generates an unbounded number of linguistic representations; it is also, in a sense, ancillary to the external systems that deal with sounds and meanings as it subserves the needs of such systems. This is clearly expressed by certain concepts of economy which are assumed to apply to syntactic computations within Minimalism: the assumption is that there is no true syntactic optionality; a syntactic device is used only when it is needed to obtain a certain interface effect (Fox 2000; Reinhart 2006). This conception led to a reanalysis of many apparently optional syntactic processes and yielded significant empirical results: very often an apparent optionality reveals detectable interpretive (p.67) differences (in phenomena like scrambling, “free” inversion; e.g., Belletti 2004) upon careful analysis. The fundamentally ancillary character of syntax makes intuitive sense. What really matters, for the expression of thought and communication, is the articulation of sound-meaning pairings. Syntax is the powerful mechanical device which makes the generation of the pairing possible over an unbounded domain.

A model like (4) raises the question of the “timing” of the transfer to the interface systems. How is it done? Traditional models of the Extended Standard Theory assumed that complex syntactic structures (with embeddings, etc.) are computed entirely by the syntactic component; thereafter, the entire configuration is transferred to PHON and SEM (or the equivalent interface levels). Among other more technical drawbacks, this radical “syntax first” assumption had the effect of divorcing the model of linguistic competence from the functioning of the processor: clearly, when we parse and interpret a complex utterance, we do not complete the syntactic analysis before starting to build the interpretation, even though it is plausible that the syntactic analysis is the necessary initial step (e.g., Frazier 1987a; Friederici 2000). In current minimalist models, the syntactic computation is assumed to proceed by phase (Chomsky 2001, 2007; Nissenbaum 2000): relatively small chunks of syntactic structures, the phases, are computed (roughly corresponding to simple clauses, but assumptions vary on the exact size of the phase) and sent to the interface, and then the syntactic component computes another phase and sends it to the interface, and so on.3

Merge and Structure Building

Various recursive techniques have been adopted in the different linguistic models which have been proposed since the 1950s, ever since Syntactic Structures (e.g., generalized transformation, rewriting rules, X-bar theory; Chomsky 1957). Jumping ahead almost fifty years of syntactic research, the ultimate distilled format of syntactic recursion is the operation Merge, the fundamental structure-building procedure assumed by the Minimalist Program (see Chomsky 1995, 2000 and related work), which takes two elements, A and B, to form a composed expression C:

  1. (5) Some Elements of Syntactic Computations

(p.68) The operation is recursive in that it can reapply indefinitely to its own output, generating a hierarchical structure. Thus, A and B can be two elements taken from the lexicon or complex expressions already formed by previous applications of Merge.

Merge strings words together and, at the same time, expresses the hierarchical structure of the sentence giving rise to tree representations. Merge, as in (5), creates a minimal subtree with two sister nodes, A and B, and a mother node, C. In the very impoverished computational system assumed by minimalist syntax, the computational component cannot introduce new labels (i.e., labels not already present in the lexical items involved: the inclusiveness principle). Thus the label C of the AB constituent in (5) must be inherited from one of the merged elements, the one which “projects”: either C = A, or C = B. The element that projects is the “head” of the construction.

Let us consider a concrete case. For Merge to apply there must be some kind of “affinity”: some selectional relation between A and B. IfA is a transitive verb and B is a noun, Merge can apply, forming a transitive verb phrase (say, meet Bill). In the obtained configuration, the selector is the head, the element possessing the label which projects, and gives a name to the whole structure; thus, in the case of a verb—object construction, the obtained C constituent would be a verbal projection, a verb phrase in more traditional terminology:

  1. (6) Some Elements of Syntactic Computations

Successive applications of Merge can give rise to complex structures like the following, expressed here in terms of the approach known as Bare Phrase Structure, a component of Minimalism: The verb meet is merged with the noun Bill to give rise to the verbal constituent meet Bill, which is then merged with the tense-bearing element, here the modal will. After merger of the subject (cf. later discussion for an important refinement), the sentence thus created, Mary will meet Bill, is merged with the complementizer that, an element which transforms a sentence into a complement, available to be selected by, and merged with, a higher verb, said, and so on.

  1. (7) Some Elements of Syntactic Computations

(p.69) These representations express key properties of the structure of sentences (e.g., what words are related to other words, what units are formed) and enter directly into the determination of form and meaning of the sentence. These representations are transferred to the interface systems at the end of a phase, in a phase-based model, and determine, on the PHON side, intonation and other prosodic patterns; on the SEM side, they determine properties of argumental semantics (who does what to whom), of referential dependencies (the interpretation of pronouns, anaphors, and the like), and of the scope-discourse semantics (e.g., scope of operators, informationally related properties such as topicality and focus). We will return to this shortly.

Going back for a moment to the properties of Merge, we can observe that it is an extremely general formal operation. Minimalism accepted the difficult challenge of showing that all the fine details of syntactic structures uncovered in half a century of formal syntax could be traced back to this operation, in interaction with reasonable assumptions on lexical specifications and interface requirements. The challenge is remarkably successful, even though many problems remain.

The hypothesis that Merge represents the core of syntax has opened new perspectives for the study of language evolution. Hauser et al. (2002) speculate that the availability of recursion in the right “spot” of the human cognitive map, perhaps in the form of the Merge operation, may have been a sudden and recent event in evolutionary history, perhaps the major computational consequence of a minor reorganization of the brain, and that this single evolutionary event may be at the root of the emergence of the language faculty and possibly, even more broadly, of what paleoanthropologist Ian Tattersall calls “the human capacity,” the collection of cognitive capacities that distinguishes our species from the others (Tattersall and Schwartz 2001). The following are examples of the many questions raised by this fascinating hypothesis:

  • Is there an identifiable neural substrate which implements the recursive property for language, and, if so, at what granularity of analysis could it emerge?

  • If so, how does it relate to the other major human capacity—namely, the capacity to count (Dehaene 1997), which deals with discrete infinities?

  • Do the mechanisms underlying linguistic and numeric capacities relate in a nontrivial way to the mechanisms responsible for other kinds of hierarchical structures in other cognitive domains (e.g., vision, motor control, the theory of mind and the other cognitive capacities that govern social interactions)?

  • How did the mastery of recursion for communication and other human cognitive systems evolve in the natural history of the species?4

(p.70) These and many related questions define a broad and ambitious long-term program, but it is imaginable that partial answers to some of these questions may be within reach, through the conjoined efforts of formal modeling of cognitive capacities, the study of pathology, and brain imaging techniques.

Movement

A pervasive property of natural language syntax is movement. We use this term to refer to the fact that linguistic expressions are very often pronounced in positions different from the positions in which they are interpreted (or, more exactly, in which they receive crucial elements for their interpretation). Consider:

  1. (8)

    1. (a) [Which book] did Mary want to buy …?

    2. (b) Mary wanted to buy [this book].

    3. (c) Which book did you say…John believed…Mary wanted…. to buy —?

To understand a sentence like (8a), it is necessary to interpret the expression which book as the object of the verb buy, much as this book in (8b); however, which book has been displaced to the initial position and, in fact, it may end up being, in the surface configuration, indefinitely far away from the immediate structural context of buy, as in (8c). Still, it must be interpreted as belonging to the argument structure of buy.

To illustrate the pervasiveness of movement, consider the following French example:

  1. (9) Qui rencontreras-tu ?

”Whom will-meet you-NOM? = Whom will you meet?”

(p.71) French is an SVO (subject-verb-object) language like English, but in this kind of structure the order is reversed to OVS by various applications of movement, as indicated by the arrows:

  1. (10) Some Elements of Syntactic Computations

This gives rise to a surface configuration like the following:

  1. (11) [Qui rencontr+eras C [tu ____ … [____ _____ _____]]].

where the blanks indicate the positions vacated by movement, or the “traces” of movement. In (11) the verb phrase, the initial nucleus in which thematic roles like agent and patient are assigned, is completely vacated by movement, and this is by no means an exceptional situation. There are at least three kinds of movement involved in (10):

  • Head movement, forming the complex word rencontrera, hence associating the lexical verb to the tense affix,

  • A-movement, which moves the subject from its thematic position to its canonical clause-initial position (in languages like French),

  • A’-movement (read “A-bar”), which moves the interrogative pronoun qui to clause-initial position.

I will now illustrate these in the appropriate structural context.

Clausal Structure, Head Movement, and A-movement

Let us follow a derivation step by step, looking first at an example in English. The verb meet takes two arguments: an agent and a patient. The verbal root is thus initially merged with two nominals, which receive the two “theta roles,” or argument roles, of agent and patient; this operation satisfies the argument structure of the verb, giving rise to the thematic nucleus of the clause: the verb phrase (VP, in traditional informal notation):

  1. (12) [you [meet Mary]].

Then, the functional structure is added, which forms the structural backbone of the clause; in particular, it includes the tense specification. Tense has a number of functions, which affect both the form and the interpretation of the expression.

In the system presented here, tense has at least a dual function of relevance for the interpretive systems. It locates the described event in time with respect to the speech time (present—past—future), and it (or some element close to it in the functional structure of the clause) creates the subject—predicate articulation.

(p.72) Suppose that the tense (T) element merged to (12) is a future marker, expressed in English by the modal will:

  1. (13) will [you [meet Mary]].

In languages like English, T attracts the closest nominal to its left-adjacent position (its specifier), thus creating the subject—predicate articulation, or “aboutness” structure (“about argument X, I’m presenting event Y concerning X”). This is an instance of A-movement, or movement of an argument to a subject position:

  1. (14) You will [ [meet Mary]].

In English, future T is expressed by an autonomous word, the modal will. In some languages, T is always expressed by an autonomous particle (e.g., typically in Creole languages), but this is by no means the general case.

In French, as in many other languages, future T is expressed by an affix, an element which does not form an independent word:

  1. (15) Tu rencontr-eras Marie “You will-meet Marie.”

In this, as in many other cases, linguists have profitably followed an intuition of uniformity and assumed that the clausal structure of French is exactly the same as in English. Thus, future T, expressed by the affix -eras, is merged to the verb phrase nucleus, much as in (13):

  1. (16) -eras [tu rencontr-Marie ].

Example (15) is derived by a double movement: tu A-moves to the canonical subject position. Moreover, as -eras is not a morphologically well-formed word in French, but an affix, something must happen to associate it to the verbal root. This is also done through movement. So, the verb moves to T, giving rise to the complex inflected verb rencontr-eras:

  1. (17) tu rencontr+eras [____ _____Marie].

This is a case of head movement: a head, the verb, moves to the next higher head and combines with its content. The reason for this process is to align syntax and morphology: syntactic units (the heads) and morphological units (the words) do not match perfectly; in particular, there are heads which are not complete words. Head movement aligns the two systems by forming complex inflected words: a lexical root moves to higher heads “picking up” the morphological specifications expressed in them. The system is illustrated in (17) with a single functional head for T, but it readily generalizes to more complex cases involving a richer functional structure expressing mood, tense, aspect, voice, the markers of agreement with the subject, and other arguments (Cinque 1999).

(p.73) Important evidence supports this syntactic conception of how inflectional morphology works, and offers straightforward explanations for complex cross-linguistic patterns of adverb distribution (Emonds 1978; Pollock 1989).

Consider the different position of a frequency adverb like often/souvent in English and French:

  1. (18)

    1. (a) You often meet Mary.

      You will often meet Mary.

    2. (b) Tu rencontres souvent Marie.

      Tu rencontreras souvent Marie.

The adverb precedes the verb phrase (VP) in English, whereas it interpolates between the inflected verb and the object in French. Following again a fundamental intuition of uniformity, Emonds (1978) and Pollock (1989) have proposed that the adverbial position is the same in the two languages: the adverb is merged with the VP it modifies. Moreover, in both languages the subject moves to the initial position, the specifier of T. What varies is the independent difference we have just seen in how the morphology—syntax interface is addressed: French involves verb movement to T, which raises the verbal root past the adverbial position, whereas in English the lexical verb does not move. Thus, here we have a single movement in English and a double movement in French:

  1. (19)

    1. (a) Some Elements of Syntactic Computations

    2. (b) Some Elements of Syntactic Computations

This mode of explanation connecting morphology and syntax has proven extremely fruitful. First, it has prompted an in-depth exploration of syntactic structures, giving rise to the so-called cartographic projects, which look at the fine details of the structural articulation of clauses and phrases using, as guidelines, the morphological properties of the expression of tense (but also mood, aspect and voice) and the distributional properties of adverbials and other kinds of elements. Second, it has produced an effective technique to address major cases of variation in word order within and across languages. Third, it has favored the exploration of diachronic patterns of change in the morpho-syntax of languages.

Comparative and Diachronic Implications

It does not seem to be a pure syntactic accident that French requires head movement of the lexical verb to the inflectional system, while Modern English (p.74) does not. There seems to be a relation between the richness of the paradigm of verbal morphology (in particular, the well-differentiated expression of agreement) and the syntactic movement of the verb: roughly speaking, a rich inflectional system is able to attract the verb out of the VP, a more impoverished system does not. This is immediately suggested by certain comparative facts. Consider, for instance, Icelandic in comparison with the mainland Scandinavian languages Swedish, Norwegian, and Danish.

Icelandic has a well-differentiated agreement paradigm both in the present and preterite tenses; mainland Scandinavian varieties, illustrated here by Danish, lost the agreement specification completely and have a single verbal form that co-occurs with subjects with all persons and numbers:

  1. (20) Icelandic (heyra “hear”) present: heyr-i, heyr-ir, heyr-ir, heyr-um, heir-ið, heyr-a preterite: heyr-ði, heyr-ði-r, heyr-ði, heyr-ðu-m, heyr-ðu-ð, heyr-ðu

  2. (21) Danish (høre “hear”) present: hør-er preterite: hør-te.

Not surprisingly, in Icelandic the lexical verb raises to T past the negative adverb ekki, much as in French, whereas in Danish the lexical verb does not leave the VP and appears lower than the negative adverb ikke (we have to use embedded clauses to see these phenomena in order to control for Verb Second, a major phenomenon radically modifying word order in main clauses in most Germanic languages):

  1. (22) …að hann  keypti ekki bokina (Icelandic) “that he bought not the book”

  2. (23) …at han ikke kobte  bogen (Danish) “that he not bought the book.”

There are immediate diachronic implications. Platzack (1987) showed that until the seventeenth century Swedish had rich inflection and verb movement, much as modern Icelandic; both properties were lost, however, in the following history of the language. Faroese, a Scandinavian variety spoken in the Far Oer islands, roughly halfway in between Iceland and the continent, seems to be in an unstable transitional state, with dialectally variable morphological richness and verb movement.

A clear diachronic effect is also straightforwardly observable from the history of English, as highlighted by Roberts (1993). He observes that sixteenth-century English showed clear signs of verb movement, as illustrated by the interpolation of negation and various kinds of adverbials between the lexical verb and the direct object:

  1. (24)

    1. (a) If I gave not this accompt to you (1557)

    2. (b) In doleful way they ended both their days (1589).

(p.75) In parallel, Roberts observes, verbal morphology expressed a richer paradigm of agreement, with some variation between different varieties:

  1. (25) Is: cast Ip: cast(-e) IIS: cast-est IIp: cast(-e) IIIS: cast-eth IIIp: cast(-e).

Again, loss of verb movement and loss of a rich morphological expression of agreement seem to have gone hand in hand. Much research has been devoted to the exact characterization of the notion of “morphological richness” (Roberts 1993; Vikner 1997; Rohrbacher 1999) as well as to the important question concerning the direction of the causal effect, from morphology to syntax, or vice versa. According to the first view, the properties of inflectional morphology cause the syntactic behavior directly: if the morphology reaches a certain threshold of “richness,” syntactic movement is automatically triggered. According to the alternative view, morphology merely registers what syntax does, and “rich agreement” is simply the way in which morphology typically underscores the fact that syntactic movement took place. The debate revolves around certain exceptions to the generalization “Rich agreement if and only if syntactic movement” in Scandinavian dialects and other regional varieties, and represents a very lively chapter of current research.

VSO Languages

The same ideas that have been proposed to address the syntax—morphology interface and the position of adverbials have shown a clear explanatory power with respect to other major questions of word order across languages. One concerns the analysis of VSO (verb—subject—object) languages, traditionally a serious puzzle for syntactic theory.

  1. (26)

    1. (a) Cheannaigh siad teach anuraidh (Irish) ”Bought they a house  last year”

    2. (b) Chuala Roise go minic an t-amharan sin ”Heard Roise often this song.”

In the other major language types, SVO and SOV, the verb and object are adjacent and seem to always form a constituent, the verb phrase, merged with the subject. In fact there are very good reasons to assume structures like S [VO] and S [OV]. But how can such a binary subject—predicate articulation be expressed in a VSO language, in which the subject apparently interpolates between the two constituents of the verb phrase? A traditional approach was to assume that VSO languages are different in that they instantiate a flat ternary structure, with subject and object generated as sisters of the verb. This approach is suspect, both on theoretical and empirical grounds. Theoretically, if the structure-building operation is Merge, structures can only involve binary branching; the point that tree branching is binary was forcefully asserted by (p.76) Kayne (1984). Empirically, all of the standard evidence showing that the subject is structurally higher than the object applies to VSO languages as well. For instance, the subject typically can bind a reflexive in object position (John washed himself), but not vice versa (*Himself washed John). This property can be shown to be sensitive to structural prominence, not just linear order (e.g., in constructions with a clause-final subject in Italian, the subject can bind a preceding object, and not vice versa: ha votatoper se stessa anche la sorella di Gianni, “voted for herself also Gianni’s sister”). The asymmetry clearly holds in VSO languages as well, with subjects binding objects but not vice versa. Thus, subject—object asymmetries emerge which would not be expected under a “flat structure” analysis. Moreover, phrasal idioms typically involve the VO sequence, with the subject remaining a freely referential position (John kicked the bucket) but virtually no idiom involves the SV sequence leaving the object position referential, a fact which follows from the fact that V and O form a constituent (to which a special idiomatic interpretation can be attached), while S and V do not. Idioms pattern exactly like that in VSO languages.

The natural solution is that the VSO order is derived via verb movement to an inflectional head X from an underlying SVO (or SOV) order (Emonds 1980):

  1. (27) Some Elements of Syntactic Computations

Perhaps X is T, and VSO languages have V to T movement as in French. However, they differ from French in that the subject does not move to the specifier of T (i.e., only one of the two movements involved in (19b) takes place here). More plausibly, X is a functional head higher than T, and V-movement takes this extra step to X with respect to a language like French, where it stops in T (that the subject moves in VSO languages is suggested by the possibility of adverb interpolation between the subject and the object in examples like (26b).

When an auxiliary verb is raised, as in the Welsh example (28b), the SVO order with the lexical verb resurfaces:

  1. (28)

    1. (a) Cana i yfory (Welsh) ”Will-sing I tomorrow”

    2. (b) Bydda i ’n canu yfory ”Will-be I singing tomorrow.”

Thus, recalcitrant VSO languages can be traced back to familiar ingredients: a basic order permitting the subject—predicate articulation and head movement of the verb to the functional system.

A’-Movement and the Interface with Semantics and Pragmatics

Let us now consider the class of cases that most straightforwardly illustrates the phenomenon of movement, the displacement of an operator-like element (p.77) to the beginning of the clause. This is what linguists call A’-movement and is illustrated by the English constructions in (29), which give rise, respectively, to a question (a), a topic-comment structure (b), a focus-presupposition structure (c), a relative (d), and an exclamative construction (e):

  1. (29)

    1. (a) Which book should you read ____?

    2. (b) This book, you should read ____.

    3. (c) (It is) THIS BOOK (that) you should read____(rather than something else).

    4. (d) The book which you should read____is here.

    5. (e) What a nice book I read____!

A’-movement, in the clear cases, has a very straightforward “teleological” motivation. It is a device that associates to a linguistic expression two types of semantic properties:

  • properties of argumental semantics (e.g., thematic roles, who does what to whom), and

  • properties which Chomsky refers to as expressing scope-discourse semantics: the scope of various kinds of operators (e.g., interrogative, relative, exclamative) and such discourse-related and informationally related properties as topicality and focus.

Natural language expressions may be assigned both properties. How is this done? Among the many solutions that would be a priori possible, natural languages opt for movement: the expression is inserted in a position dedicated to argumental semantics and “picks up” the scope-discourse property through movement, by moving to a position dedicated to this kind of interpretive property. So, for instance the nominal expression which book in (29a) is to be interpreted both as the patient of the verb read and as an interrogative operator with main clause scope in a direct question, to yield a semantic representation which is something like “for what x, x a book, you read x.” This is achieved through Merge and Move: the phrase is merged with read, and in the local configuration with the verb it receives the thematic role of patient; then it is moved to the initial periphery of the clause, where it “picks up” its scope property.

How is movement triggered? In the other cases of movement (head movement and A-movement to a subject position), we have seen that it is always a functional head which attracts another element, a head or a phrase, so it is natural to assume that in movement to a scope-discourse position, this would also hold. Here I will make the somewhat controversial assumption that in fact all kinds of movement work in this manner. In other words, I will assume that the representations of (30) all involve a functional head in the left periphery of the clause which attracts the question operator in (30a), the topic in (30b), the focus in (30c), the head of the relative clause in (30d), and the exclamative operator in (30e).

  1. (30)

    1. (a) Which book Q should you read ____?

    2. (b) This book TOP you should read ____.

    3. (p.78) (c) THIS BOOK FOC you should read ____.

    4. (d) The book which R you should read ____.

    5. (e) What a nice book EXCL I read ____.

This system of heads governing A’-movement is silent in English. None of them is pronounced. However, often the corresponding constructions show overt counterparts in other languages, italicized in the following examples:

  1. (31)

    1. (a) Ik weet niet [wie of [Jan_____gezien heeft]] (Dutch varieties; Haegeman 1996) ”I know not who Q Jan seen has”

    2. (b) Un sé [do [dan lo  [Kofi hu í (Gungbe; Aboh 2001) ”I heard that snake the TOP Kofi killed it”

    3. (c) Un sé [do [dan lo  [Kofi hu_____]]] (Gungbe; Aboh 2001)”I heard that snake the Foc Kofi killed”

    4. (d) Der Mantl [den wo [dea Hons____gfundn hot]](Bavarian; Bayer 1984) ”The coat which R the Hans found has”

    5. (e) Che bel libro che [ho letto____]! (Italian) ”What a nice book Excl I read!”

Many languages overtly manifest a Q marker (e.g., the dialectal Dutch of), topic and focus markers (yá and wé in Gungbe), a relative marker (the dialectal German wo), and an exclamative marker (che in Italian). It is tempting to assume that the difference between English and these languages is very superficial and has mainly to do with the phonetic realization of a system of heads that is always present and syntactically active across languages, but pronounced only in some. These constructions have analogous (sometimes identical) syntactic properties across languages, are interpreted uniformly at the SEM interface (apart from a few possible parametrizations), and the fundamental variation only involves the superficial property of being pronounced or not at the PHON interface.

In conclusion, according to this approach the functional lexicon of every language specifies a number of heads, creating positions dedicated to scope-discourse semantics. These heads attract phrases from their thematic positions, and the created configurations are handed over to the interpretive systems where specific interpretive routines are triggered (e.g., scope, topic-comment, focus-presupposition). Thus, a typical A’-construction connects a thematic position and a scope-discourse position. A phrase moved from one to the other is interpreted as carrying both kinds of semantic properties; and both kinds of interpretive properties are assigned in analogous configurational structures: dedicated head of the substantive lexicon (for thematic roles) or of the functional lexicon (for scope-discourse properties) assign such properties to their immediate dependents, specifiers and complements:

  1. (p.79) (32) … ____Xscope-discourse … ____Xthematic

In an example from a topic-comment construction:

  1. (33) This book TOP [you should [____ read ____]]

the lexical head read assigns the role of agent to its specifier you (eventually moved to subject position) and the role of patient to its complement this book. The head TOP assigns the property of Topic to its specifier this book and the property of comment to its complement, the rest of the clause you should [___ read ____].

The next observation is that A’-positions typically tend to pile up in the left periphery of the clause, often in a fixed order, partly universal and partly subject to parametric variation. For example, the left periphery of Italian permits the co-occurrence of a topic, a focus and a preposed adverbial modifier in the space in between the declarative complementizer and the rest of the clause:

  1. (34) Credo [che [a Gianni TOP [QUESTO FOC [oggi Mod [gli dovreste dire]]]]], non qualcos’ altro.

    ”I believe that to Gianni THIS today you should say, not something else.”

There are then “cartographic” issues that arise in this domain as well. The attempts to draw maps as precise as possible of the left periphery of the clause have given rise to a very lively area of research, with the identification of different positions and ordering constraints, ruled by specific principles and parameters (e.g., Rizzi 1997; Cinque 2002; Belletti 2004; Rizzi 2004b).

The following example illustrates a reasonable approximation to a map of the left periphery that is valid quite generally across the Romance languages:

  1. (35) [ Force … [ TOP … [ INT … [ FOC … [ Q … [ MOD … [ FIN [Clause]]]]]]]].

This zone of structural layers is then assumed to include the clause, specifying a space of positions dedicated to scope-discourse semantics. A’-movement is movement of an element to one of these external layers, triggering the expression of the relevant interpretive property.

Movement: A Subcase of Merge

What kind of formal operation is movement? In traditional generative approaches, movement was performed by a class of formal rules, the transformations, which were completely different from the formal rules building the structural representations, the phrase structure rules, or X-bar theory. This sharp distinction raised a serious conceptual problem. Emonds (1970) observed that the core cases of transformations are “structure-preserving” in that they create configurations that are independently generated by phrase structure rules. Why, however, would two completely different types of rules converge to create exactly the same kinds of structures? The problem of structure preservation (p.80) led a number of researchers to explore the possibility that it was a mistake to postulate two distinct rule systems for creating and modifying structures, and that phrase structure and transformational rules could be unified.

A full unification is achieved in the Minimalist Program, which contained a single structure building operation, Merge, repeated here for convenience:

  1. (36) Some Elements of Syntactic Computations

What varies is the origin of A and B, the elements undergoing Merge. If they come directly from the lexicon, or one or both are independent complex entities already created by previous applications of Merge, we have external merge; if one (let’s say A) is taken from within the other (B in our case) we have internal merge, which amounts, in traditional terms, to moving A from within B to the position sister of B. Using the format of (36), we could depict the global operation of internal merge as follows:

  1. (37) Some Elements of Syntactic Computations

This way of expressing things is, however, misleading: (37) is not an operation distinct from (36). What differs in the two cases is simply the way in which the two candidates of Merge, A and B, are selected through a search in the available computational space: they are separate objects in (36) whereas one is contained within the other in (37). Once the two candidates are selected through some kind of external or internal search, the formal operation that strings A and B together is the same.

Consider the (simplified) derivation of an interrogative: What will you say? This will illustrate how a series of applications of external and internal merge are interspersed in the computation of a clausal structure from a selection from the lexicon like the one given in (38):

  1. (38) Selection from the lexicon: {say, what, you, will, Q}

  2. (39) Derivation:

    1. (a) [say what] Ext Merge

    2. (b) [you [say what ]] Ext Merge

    3. (c) [will [you [say what ]]] Ext Merge

    4. (d) [you [will [___[say what ]]]] Int Merge

    5. (e) [Q [you [will [___[say what ]]]]] Ext Merge

    6. (f) [will+Q [you [____[____[say what ]]]]] Int Merge

    7. (g) [what [will+Q [you [____[____[say____]]]]]] Int Merge.

In this view, the structure-preserving property of movement is immediately explained: movement is structure preserving because it is a particular case of the fundamental structure-building operation, Merge.

(p.81) Locality

Let us now focus on the procedure through which the candidates of Merge are selected. The significant case is internal merge. So, consider again the derivational stage in which the Q head, indicating an interrogative, is merged with the rest of the structure. The Q head starts a search for an interrogative operator, also endowed with the question feature.

  1. (40) Some Elements of Syntactic Computations

This is, in essence, the operation that Chomsky calls “agree,” but I opt here for the more general term “search.” Once an element is selected through the search operation (the wh-word what in our case), the search is terminated and the selected element becomes a candidate for internal merge with the whole structure. Through other operations, a sentence like (39g) is derived.

Search is a “local” operation involving a “probe” (i.e., the head activating the search) and a “goal” (i.e., the element which is reached); respectively, Q and what in (40). The operation is local in that it is blocked when an element intervenes between the probe and the goal, and the intervener bears some kind of structural similarity, to be precisely defined, to the elements involved in the relation. For instance, a wh-operator can be freely extracted from an embedded declarative (42a), but not from an indirect question, as in (42b). The structures from which extraction is attempted are given in (41):

  1. (41)

    1. (a) I think Bill behaved like that.

    2. (b) I wonder who behaved like that.

  1. (42)

    1. (a) How do you think Bill behaved _____?

    2. (b) *How do you wonder who behaved _____?

This is, in essence, the effect of Relativized Minimality: the locality principle barring local relations when an element intervenes which bears some structural similarity to the elements that should be connected.

  1. (43) Relativized Minimality: in a configuration like X … Z … Y a local relation connecting X and Y is blocked if Z has the same feature specification as X and Y (Rizzi 1990, 2004a).

In the case under consideration, the derivation of (42b) is barred because the search connecting the main Q element and how in the embedded clause is blocked by the intervention of who, which also is an interrogative pronoun:

  1. (p.82) (44) Some Elements of Syntactic Computations

Search is often discussed in the context of analyses of movement, as a prerequisite for internal merge. There are reasons to believe, however, that it is a much more general operation, perhaps encompassing all the cases in which a local relation is established between two positions, independently from movement. For instance, the binding of a reflexive element by an antecedent, or the control of the null pronominal subject (PRO) of an infinitival clause in so-called “obligatory control” constructions—two relations, expressed by co-indexation in (45), constrained by a kind of intervention locality similar to the principle operative in (44)—may plausibly involve variants of the search operation:

  1. (45)

    1. (a) Johni saw himselfi in the mirror.

    2. (b) Johni decided [PROi to leave].

Thus, the search operation that identifies the candidate for internal merge may well be a particular case of a more general operation which can connect two positions in the tree if intervention locality is respected.

Invariance and Variation

How can this kind of approach address the problem of invariance and variation in natural language? This is a fundamental question for virtually every aspect of the study of language, clearly of central relevance for the question of language evolution. The theoretical entities that have been referred to within the generative tradition to address this issue are the concepts of Universal Grammar (UG) and particular grammars. The traditional conception of the 1960s and 1970s, which was based on the idea that particular grammars are systems of rules specific to a particular language, did not provide adequate tools to factor out the invariant properties across languages (among many other drawbacks).

Things changed radically around the late 1970s with the Principles and Parameters (P&P) approach, based on very different ideas (e.g., Chomsky 1981; Rizzi 1982; Kayne 1984). The key notion became UG, which was construed as an integral component of particular grammars. UG was conceived of as a system of principles containing certain parameters, binary choice points expressing the possible cross-linguistic variation. Particular grammars could be seen as UG with parameters fixed or set in particular ways. This conception went with a particular model of language acquisition. Acquiring a language (p.83) meant essentially setting the parameters on the basis of experience. This is not a trivial task, as several researchers observed: in a number of cases, the evidence available to a child may be ambiguous between different parametric values, and there are complex interactions between parameters (Gibson and Wexler 1994). Still, despite such problems, parameter setting is a much more workable concept than the obscure notion of rule induction, which was assumed by previous models. More generally, the P&P approach introduced a very effective technical language to express, in a concise and precise manner, what languages have in common and where languages differ. Modern comparative syntax flourished once the P&P approach was introduced, and language acquisition studies took a new start.

Let me just mention here one basic example of parametrization. In some languages (e.g., VO languages), the verb precedes the object: love Mary (English) or aime Marie (French). Other languages (e.g., Japanese) have object-verb (OV) order. To address these properties, we need some kind of parameter operating on Merge and having to do with linear order. In some languages, the head (the verb) precedes the complement, whereas in other languages the head follows the complement. For a different approach to this kind of parametrization and deriving certain orders via movement, see Kayne (1994).

This simple ordering parameter has pervasive consequences in languages that consistently order heads and complements one way or the other. (Other languages, a minority according to typological studies starting with Greenberg (1963), are “incoherent” in this respect, as they opt for distinct ordering options for different types of heads.) Thus, two examples like the English sentence (46a) and its Japanese counterpart (46b) differ dramatically in order and structure, as illustrated by the two trees (47a, b):

  1. (46)

    1. (a) John has said that Mary can meet Bill.

    2. (b) John-wa [Mary-ga Bill-ni a - eru- to] itte-aru “John-Top [Mary-Nom Bill-Dat meet-can-that] said-has”

  1. (47)

    1. (a) Some Elements of Syntactic Computations

    2. (p.84) (b) Some Elements of Syntactic Computations

English expressions have a fundamentally right-branching structure, whereas Japanese expressions follow a fundamentally left-branching structure. The two are not perfect mirror images because certain ordering properties (such as the subject-predicate order) remain constant; however they are almost a mirror image of each other.

We have parameters on external merge, as the one just illustrated, and parameters on internal merge or movement. For example, we have seen that the lexical verb moves to T in French and Welsh, but not in Modern English. Also, the subject moves to the clause-initial position in English and French, but not in Welsh, yielding the VSO order. In addition, all Germanic languages except English have a Verb Second constraint in main clauses which involves a double movement of the inflected verb and another phrase to the left periphery.

A third parameter concerns Spell-out: the phonetic realization of the various expressions. There are certain elements that can or must be left unpronounced in particular configurations in some languages. One classical case is the Null Subject parameter: subject pronouns can be left unpronounced in languages like Italian and Spanish (e.g., in sentences likeparlo italiano; I speak Italian), and this property relates in a nontrivial manner to other properties of the language (e.g., Rizzi 1982).

After a few years of developing these ideas, a crucial question arose concerning how to express the format of the parameters. Is it the case that anything can be parameterized in UG, or is there a specific locus for parameters? The first idea on the locus for parameters was that parameters were expressed directly in the structure of principles. This was probably suggested by the fact that the first parameter discussed in the late 1970s had to do with a particular locality principle, Subjacency, the proposed parametrization involving the choice of the nodes that would count as bounding nodes or barriers for locality (the S/S’ parameter; see Rizzi 1982, chap. 2). On the basis of this case, it was assumed for some time that parameters were generally expressed directly in the structure of the principles, and that could be the general format. Among other things, this assumption raised certain expectations on the question of how many parameters one should expect in UG: as the UG principles were assumed (p.85) to be relatively few, if parameters were expressed in the structure of principles one could expect an equally (relatively) small number of parameters.

This view was abandoned fairly quickly, for a number of good reasons. One reason was that some principles turned out not to be parameterized at all. In no language, as far as we know, does a structure like He thinks that John is crazy allow for co-reference between He and John (principle C of the Binding Theory). That pronouns cannot be referentially dependent on some expression in their c-command domain is a general, invariable property of referential dependencies; many other principles are categorical in a similar way.

The second reason was that some macro-parameters (i.e., big parameters initially assumed to characterize basic cross-linguistic differences) turned out to require reanalysis into clusters of smaller parameters. One case in point was the so-called “configurationality parameter.” Some languages have a much freer word order than other languages. Originally it was thought that there was a major parameter that divided languages between those with and without free word order. It quickly turned out, however, that there are different degrees of free word order: some languages are freer in the positioning of the subjects, others are freer in the reordering of the complements (scrambling), etc. There is a sort of continuum: not in a technical sense, but in the informal sense that there are different degrees of freedom, so that the big “non-configurationality” parameters really needed to be divided into smaller parameters.

The third reason was that some parametric values turned out to be intimately related to specific lexical items. For instance, consider the long-distance anaphor parameter (i.e., that certain reflexives roughly corresponding to English himself in some languages allow for an antecedent that is not in the same local clause, e.g., in Icelandic). This turned out to be the specific property of certain lexical items: if the language has such special lexical items (i.e., anaphors of a certain kind), then these anaphors work long-distance. Thus, we are not looking at a global property of the grammatical system, but simply at the presence or absence of a certain kind of item in the lexicon. These considerations led to the general view that parameters are not specified in the structure of principles, but rather are properties specified in the lexicon of the language. In fact, assuming the fundamental distinction between the contentive and the functional lexicon, parameters could be seen as specifications in the functional lexicon.

To summarize, a reasonable format for parameters would be: H has F, where H is a functional head and F is a feature triggering one of the major syntactic operations. This view implies important differences with respect to the view expressing parameters in principles. For instance, the order of magnitude of parameters is now related not to the number of principles but to the size of the functional lexicon.

If one combines this view on parameters with the cartographic approach (Belletti 2004; Cinque 1999, 2002; Rizzi 1997, 2004b), assuming very rich functional structures, the implication is that there can be a very rich system of parameters. Putting together the theory of parameters as specifications in (p.86) the functional lexicon, some minimalist assumptions on linguistic computations and cartography, we end up with something like the following typology of parameters.

  1. (48) For H a functional head, H has F, where F is a feature determining H’s properties with respect to the major computational processes of Merge, Move (internal merge), and Spell-out.

Merge parameters: What category does H select as a complement? To the left or to the right? Move parameters: Does H attract a lower head?

Move parameters: Does H attract a lower phrase to its specifier? Spell-out parameters: Is H overt or null?

Spell-out parameters: Does H license a null specifier or complement?

We have parameters determining the capacity of a functional head to undergo Merge: What categories does it select, and does it take complements which precede or follow it? Perhaps even more fundamental properties include: Does the language use that particular functional head? Thus it may be the case that (certain) heads of the cartographic hierarchy may be “turned on” or “turned off” in particular languages. In terms of the move parameters, heads function as attractors: they may attract a lower head which incorporates into the attractor, or a phrase which moves to the attractor’s specifier. Does the tense marker attract the lexical verb, as it does in the Romance languages but not in English or most varieties of continental Scandinavian? For spell-out parameters, we ask: Is a particular head overt or not? For instance, the topic head is realized in some languages (e.g., Gungbe, possibly Japanese wa) but not in others (e.g., in Romance clitic left dislocation). Does a head permit null dependents? For instance, does the verbal inflection license a null subject? That is one of a number of possible ways of looking at the null subject parameter in current terms.

This general picture is rather widely accepted at present, though many points remain controversial. As there are many more parameters than was originally assumed in the early days of the P&P approach, it turns out that the different parametric choices will enter into various complex kinds of interactions, generating many possible configurations of properties, so that the superficial diversity to be expected is very significant. For this reason, the parametric approach turned out to be particularly well-suited for the comparison of historically and structurally close languages, in which it is easier to control for very complex interactions; the likelihood of pinning down truly primitive points of differentiation is high, as Kayne (2000) indicated. This also explains the remarkable success of the approach to account for dialectal varieties (e.g., Italian and Scandinavian dialects) within language families (e.g., Romance, Germanic) and the renewal of dialectological studies based on this theory-driven approach.

Despite the apparent vastness of the parametric space, the abstract structure of parametric options still reduces to very few schemata, perhaps along the lines of (48). The deductive interactions between P&P remain very tight, so (p.87) that there are many logical possibilities that are excluded in principle. A profitable research strategy remains the search for attested and unattested clusters of properties under large-scale surveys. Consider, for example, Cinque’s (2005) discussion of Greenberg’s (1963) Universal 20, which showed how certain systematic gaps in the ordering of various kinds of nominal modifiers can receive a principled explanation. Empirical claims about universals can now be checked against a data base that is enormously richer than it was 25 years ago. In addition, the search for nonaccidental gaps in the clustering of properties preserves all of its heuristic value and naturally complements the study of the micro-parametrization within very close systems, in the attempt to understand different facets of invariance and variation across languages.

Acknowledgments

I would like to thank Derek Bickerton, Noam Chomsky, and an anonymous reviewer for helpful comments on an earlier draft.

Notes:

(1) For example, “…des phrases et des groupes de mots [sont] etablis sur des patrons reguliers... repondent a des types generaux…” but “…ilfaut reconnaitre que dans le domaine du syntagme il n’y a pas de limite tranchee entre le fait de langue, marque de l’usage collectif et le fait de parole, qui depend de la liberte individuelle” (sentences and groups of words are established on regular patterns…but… it is necessary to recognize that in the domain of the phrase there is no sharp limit between the fact of langue, characterized by the collective usage, and the fact of parole, which depends on individual freedom). The interpretation of Saussure’s ideas about the open-ended character of syntax may be controversial, as an anonymous reviewer points out, but the fact that linguistic approaches of the early twentieth century did not work out an operative formal device to capture this property is clear (Saussure 1916/1985, p. 173).

(2.) I include in the functional lexicon the system of functional heads which structure the clause and trigger important operations, such as movement.

(3.) Standard minimalist models involve “bottom-up” derivations of the kind illustrated in what follows. Other models, motivated by both linguistic and psycholinguistic considerations, involve variants of Merge and phase theory consistent with “top-down” derivations (Phillips 2003; Chesi 2005). I do not address this issue here and will adhere to standard assumptions.

(4.) Derek Bickerton (pers. comm.) points out that Merge, so broadly construed, can hardly be seen as a domain-specific operation. For instance, the stone-chipping technique, which consists of successive applications of the same operation to build an organized object and was presumably available to hominids long before the advent of modern humans, is reminiscent of sentence building through successive applications of Merge. The point is well-taken. In fact, it is precisely the great generality of Merge which invites comparisons across cognitive capacities in search for cognitive and neural invariants involved in different types of computations. Such comparisons would have been hopeless with more complex models of Universal Grammar (UG), whose constructs (e.g., “the specified subject condition,” “the complex NP constraint”) sounded deeply rooted in language-specific notions and categories, but are perfectly sensible in a Minimalist, Merge-based conception of UG. Granting that Merge-like operations may well exist in other cognitive domains and in other species, perhaps supported by analogous cognitive and neural mechanisms, the question of the recent evolution of syntax clearly cannot be phrased in terms of the appearance of Merge tout court: what remains specific to modern humans is the availability of a Merge-like operation in the system of signals for communication and the expression of thought. If we phrase things in this way, a central question for the evolution of syntax then becomes: How and when did a Merge-like operation, previously available to other cognitive systems, penetrate the system for the expression and communication of thought?