Loading field map
Syntax is the subfield of linguistics that studies how words combine into larger structures—phrases, clauses, and sentences—and what principles govern those combinations. Where morphology examines the internal structure of words, syntax takes the word as its basic unit and asks how sequences of words are organized, how that organization contributes to meaning, and why some sequences are possible in a language while others are not. The central object of study is not the meaning of individual words but the systematic patterns by which words are arranged, grouped, and related to one another.
At its heart, syntax asks a deceptively simple question: what makes a string of words a sentence of a given language? A speaker of English knows that the cat chased the mouse is a sentence, that chased the cat mouse the is not, and that the mouse chased the cat is a different sentence with a different meaning, even though it uses the same words. This knowledge is not memorized word by word; speakers can produce and understand sentences they have never heard before. Syntax seeks to characterize the system that makes this possible.
Several enduring questions organize the field. First, what are the units of syntactic structure? Words group into phrases—noun phrases, verb phrases, prepositional phrases—and phrases combine into clauses. The question is whether these groupings are merely convenient descriptions or reflect something real about how language works. Second, what are the rules or principles that license some combinations and rule out others? Early work tried to write explicit rules; later work asked whether the apparent variety of rules across languages might reduce to a smaller set of universal principles. Third, how does syntax connect to meaning? The same words can be arranged to express different who-did-what-to-whom relations, and syntax must explain how structural position determines interpretation. Fourth, how do languages differ, and what do they share? Every human language has syntax, but languages differ in word order, in how they mark grammatical relations, and in what they require to be expressed. A central question is whether these differences are superficial variations on a common theme or reflect fundamentally different organizational principles.
The modern study of syntax emerged in the twentieth century, but it has deeper roots. Traditional grammar, inherited from classical antiquity and developed through the medieval and early modern periods, provided a vocabulary for talking about parts of speech and sentence functions—subject, object, predicate—that remains in use today. This tradition was largely prescriptive and taxonomic: it classified constructions and taught correct usage, but it did not ask why the patterns hold or how speakers acquire them.
In the early twentieth century, structural linguistics, particularly in the American tradition associated with Leonard Bloomfield and his successors, made syntax an object of scientific study. Structuralists treated a language as a system of contrasts and distributions: they asked where a given element could appear relative to others, and they developed procedures for discovering the units of a language from its surface patterns. Their methods were discovery-oriented, aiming to segment and classify utterances without appealing to meaning or to the minds of speakers. This work established the idea that syntax could be studied rigorously and empirically, but it tended to produce inventories of patterns rather than explanations of them.
The decisive break came in the late 1950s with the work of Noam Chomsky. Chomsky argued that the goal of linguistics should be to characterize the implicit knowledge that lets speakers produce and understand an unbounded number of sentences. He proposed that a grammar is a finite system that generates all and only the grammatical sentences of a language, and he introduced the idea of transformations: operations that map one syntactic structure onto another. This generative turn reframed syntax as the study of the mental grammar underlying linguistic competence, and it set the agenda for the field for the next half century.
The generative tradition has dominated academic syntax since the 1960s, but it has never been monolithic, and it has not been the only game in town. The field is best understood as a set of overlapping research programs that share some questions but differ in their assumptions about what a syntactic theory should look like.
Chomsky's early theory, developed in Syntactic Structures (1957) and refined through the 1960s and 1970s, proposed that a sentence has two levels of representation. A deep structure captures the basic grammatical relations—who does what to whom—and is generated by phrase-structure rules. Transformations then move, delete, or insert elements to produce the surface structure, the form that is actually pronounced. For example, the passive sentence the mouse was chased by the cat was derived from a deep structure roughly like the cat chased the mouse by a transformation that rearranged the noun phrases and adjusted the verb.
This architecture explained a striking fact: sentences that look very different on the surface can be systematically related, and sentences that look similar can have different underlying structures. The classic example is the ambiguity of flying planes can be dangerous, which has one reading where flying planes is the activity and another where planes that fly are the danger. Transformational grammar could represent these as two different deep structures converging on one surface form.
The theory underwent major revisions. In the 1980s, Chomsky proposed the Government and Binding framework, which replaced the inventory of specific transformations with a small set of general principles and a single movement operation. The idea was that languages differ not in their rules but in a few parameter settings—binary choices such as whether the verb comes before or after its object—while the underlying principles are universal. This framework was succeeded in the 1990s by the Minimalist Program, which asks how much machinery is actually necessary. Minimalism tries to derive syntactic properties from the demands of the interfaces between the grammar and the systems of meaning and sound, reducing the theory to the operations of Merge (combining two elements into a new unit) and Move (rearranging elements already combined).
The generative tradition's enduring contributions include the insight that syntactic knowledge is systematic and productive, the development of precise formal tools for describing structure, and a research agenda that connects syntax to language acquisition and to the biology of language. Its limits are also well recognized. The theory has changed substantially over time, and earlier versions are not simply superseded but partially abandoned; the field remains divided over how much of the current framework is correct. Critics have also argued that the focus on English and a handful of well-studied languages has led to theories that are less universal than claimed, and that the abstract representations posited are hard to connect to observable data.
In the 1970s, a group of linguists led by David Perlmutter and Paul Postal broke with Chomsky's framework to develop Relational Grammar. Instead of representing syntactic structure as a tree of phrases, Relational Grammar took grammatical relations—subject, direct object, indirect object—as primitive elements. A sentence is represented as a sequence of strata, each specifying the grammatical relations of its noun phrases, and rules describe how relations change from one stratum to the next. This approach was designed to handle phenomena like passivization, where the object of an active sentence becomes the subject of a passive one, across a wide range of languages.
Relational Grammar was influential in showing that many syntactic phenomena are best understood in terms of grammatical relations rather than phrase structure, and it inspired later frameworks such as Arc Pair Grammar and Role and Reference Grammar. Its emphasis on typological breadth—testing claims against many languages—was a corrective to the English-centric focus of early generative work. However, it never achieved the theoretical unification of the Chomskyan program, and its formal apparatus was less developed. Its insights have largely been absorbed into other frameworks, but its insistence that grammatical relations are real and central remains influential.
Developed by Joan Bresnan and Ronald Kaplan in the late 1970s and 1980s, Lexical-Functional Grammar (LFG) shares with generative grammar the goal of a precise formal theory but rejects the idea of transformations. LFG posits two parallel levels of representation: a constituent structure (c-structure) that captures the surface arrangement of words into phrases, and a functional structure (f-structure) that captures grammatical relations and features such as tense and agreement. The two levels are linked by correspondence functions, not by derivations.
This architecture has several consequences. Because there is no movement, LFG handles word-order variation differently: languages with flexible word order can have the same f-structure but different c-structures. Because much of the syntactic work is done in the lexicon, LFG is well suited to languages with rich morphology, where grammatical relations are marked on the verb rather than by word order. LFG has been particularly successful in computational linguistics, where its explicit formalization and its separation of levels make it implementable. Its limits are that it is less ambitious than the Minimalist Program in explaining why syntax is the way it is; it is more of a descriptive and computational framework than a theory of universal grammar.
Head-Driven Phrase Structure Grammar (HPSG), developed by Carl Pollard and Ivan Sag in the 1980s, is another nontransformational approach. HPSG represents syntactic knowledge as a system of constraints on feature structures—complex objects that specify the properties of words and phrases. There is no deep structure and no movement; instead, a single level of representation is constrained by general principles, and the lexicon carries much of the burden.
HPSG is notable for its formal rigor: it is designed to be computationally explicit, and it has been used extensively in natural language processing. It also handles local phenomena—agreement, subcategorization, unbounded dependencies—in ways that are often more transparent than transformational accounts. Its weakness is that it has not engaged as deeply with the acquisition question or with the search for universal principles; it is better at describing the syntax of particular languages than at explaining why languages are constrained in the ways they are.
A different tradition, with roots in the work of Lucien Tesnière in the mid-twentieth century, is Dependency Grammar. Where phrase-structure grammars treat sentences as nested constituents, dependency grammars treat them as networks of binary relations between words. Each word is a dependent of exactly one other word (its head), and the structure of a sentence is a tree of these dependency relations. The verb is typically the root of the sentence, with its arguments and modifiers attached to it.
Dependency grammar has a long history in European linguistics and has become prominent in computational linguistics, where dependency parsing is a standard technique. Its advantages are that it handles free word order naturally—the dependency relations remain the same regardless of linear order—and that it aligns well with the way grammatical relations are often conceived. Its limits are that it has less to say about constituency and about the hierarchical structure of phrases, and it has not produced a unified theory of the kind that generative grammar has.
A very different set of approaches rejects the core assumption of generative grammar that syntax is an autonomous mental module governed by abstract rules. Usage-based linguistics, associated with researchers such as Joan Bybee and Michael Tomasello, argues that grammar emerges from language use. On this view, syntactic patterns are not innate principles but generalizations over the utterances a speaker has heard and produced. Frequency matters: frequently occurring sequences become entrenched as units, and children acquire syntax by learning specific constructions and then abstracting patterns from them.
Construction Grammar, developed by Charles Fillmore, Paul Kay, and Adele Goldberg, is the most fully articulated version of this idea. A construction is a pairing of form and meaning, ranging from single words to fully general patterns like the ditransitive (give someone something) or the passive. There is no distinction between the lexicon and the grammar; both are inventories of constructions of varying complexity and abstraction. This approach handles idioms and irregular patterns naturally—they are just constructions—and it has been influential in language acquisition and in the study of language change.
These approaches are genuinely different from the generative program: they deny that syntax is an autonomous system, they reject the poverty-of-the-stimulus argument that motivates innate universal grammar, and they emphasize the role of input and use. Their limits are that they have been less successful at capturing the subtle structural generalizations that generative grammar handles well, and their formal apparatus is often less precise. The debate between generative and usage-based approaches remains one of the deepest divisions in the field.
These approaches are not simply rivals that have succeeded one another in time. Transformational generative grammar has been the dominant framework since the 1960s, but it has coexisted with alternatives throughout. Relational Grammar emerged as a direct challenge in the 1970s; LFG and HPSG developed in the 1980s as explicit alternatives to transformations; dependency grammar has a separate lineage that never engaged deeply with the Chomskyan program; usage-based approaches have roots in functional linguistics and have grown in influence since the 1990s.
The relationships among them are complex. LFG and HPSG share with generative grammar the goal of a precise formal theory and the assumption that syntax is a systematic, rule-governed system, but they reject the specific architecture of transformations and the strong claims about universal grammar. Dependency grammar is orthogonal to the generative-transformational debate: it is a different way of representing structure, compatible with various theoretical commitments. Usage-based approaches challenge the foundational assumptions of all the formal schools, arguing that the very idea of a distinct syntactic system is mistaken.
There is also meaningful cross-fertilization. Work in LFG and HPSG has influenced computational linguistics, which has in turn influenced how all theories are evaluated. Dependency grammar has become the standard in natural language processing, and its success has led formal linguists to take it more seriously. Usage-based insights about frequency and entrenchment have been incorporated into some formal theories, even if the larger claims remain contested.
Contemporary syntax is characterized by pluralism. The Minimalist Program remains the most visible and most widely practiced approach, particularly in North America and in theoretical linguistics more broadly, but it no longer has the near-monopoly it once enjoyed. LFG and HPSG have active research communities, especially in Europe and in computational circles. Dependency grammar is dominant in natural language processing and has a growing presence in theoretical work. Usage-based and construction-based approaches are strong in language acquisition, psycholinguistics, and typology.
Several developments have reshaped the field in recent decades. The rise of large-scale typological databases has made it possible to test claims about universal grammar against hundreds of languages, and this has both supported and challenged specific proposals. Experimental syntax has emerged as a subfield, using judgments, reading times, and brain imaging to test theoretical claims. The growth of computational linguistics has created a demand for syntactic theories that can be implemented, favoring frameworks with explicit formalizations. And the increasing attention to understudied languages has revealed phenomena—such as ergativity, where the subject of an intransitive verb patterns with the object of a transitive verb—that complicate theories built on English.
The field remains divided over its most fundamental questions. Whether syntax is an autonomous mental system or a byproduct of general cognitive processes; whether the diversity of languages reflects a shared universal grammar or a set of general learning mechanisms; whether the right way to represent structure is in terms of constituents, dependencies, or constructions—these questions are not settled, and they may not be settleable by the methods currently available. What is not in dispute is that syntax is a real and central aspect of language: every human language has systematic principles of word combination, and understanding those principles is a necessary part of understanding what language is and how it works.