Subjacency¶
A generative-syntax locality constraint under which one movement step may cross at most one designated bounding node.
Core Idea¶
Subjacency is a locality constraint in a historical family of generative-syntax theories. A single movement step may cross at most one bounding node, where the counted phrase categories are specified by the analysis. It offers a way to explain how a wh-phrase can be interpreted far from its pronounced position while each derivational step remains local: the phrase moves through intermediate clause-edge positions. If a required intermediate position is unavailable, a subsequent step may have to cross two bounding nodes and fail the constraint.[1][2]
The claim is theory-internal. It is not an independently observed physical barrier inside a sentence, nor an agreed complete explanation of every island effect. Historical analyses vary the bounding-node inventory, and contemporary work also examines structural, functional and processing explanations of extraction judgments. The entry describes the formal device and the predictions it makes under stated assumptions, rather than declaring all speakers' judgments to be categorical consequences of one rule.[2][3]
Structural Signature¶
Sig role-phrases:
- Movement dependency — A displaced phrase and an interpreted base position supply the relation whose derivation is being tested. Surface distance alone does not give the number of movement steps.[1]
- Bounding-node inventory — The grammar identifies which phrase categories count; MIT notes teach a CP/DP version and discuss IP/DP alternatives. The inventory is part of a formulation, not a timeless universal pair.[1][2]
- Local movement step — The unit checked by the rule. No one step can cross two counted nodes; a long dependency may contain several individually legal steps.[2]
- Intermediate escape site — A clause-edge position such as Spec,CP can host a temporary landing site, making successive-cyclic movement possible.[1]
- Island configuration — A filled edge or nested DP/CP can block the route so that a needed step crosses too much structure in the specified theory.[2]
- Acceptability evidence — Predicted contrasts must be compared with actual judgments and competing analyses; the constraint is an explanation to test, not the judgment itself.[3]
Condensed: movement dependency + declared bounding nodes + no more than one crossed per step + available intermediate edges → a locally licensed derivation.
What It Is Not¶
- Not a ban on all long-distance dependencies. Successive-cyclic movement can cross many clauses overall by taking several locally licensed steps.[1]
- Not identical to an island effect. An island is a pattern of difficult or prohibited extraction; subjacency is one proposed structural derivation of particular cases.[3]
- Not one fixed universal CP/DP or IP/DP inventory. Those are versions used in analyses, with historically different choices and language-comparison proposals.[1][2]
- Not a guarantee of categorical ungrammaticality from every alleged island. Acceptability can vary with construction and context, and alternative explanatory factors remain debated.[3]
- Not prime Non-Locality. That broad idea concerns relations spanning distance; subjacency is a specific formal restriction on steps in syntactic movement theory.
Scope of Application¶
The classic setting is wh-extraction. An English question can connect a fronted wh-phrase to an embedded object position through one or more clauses. In the cited CP-edge analysis, the phrase stops at an intermediate Spec,CP position; each step crosses no more than one bounding node. That analysis explains why the surface distance can be large while derivational locality is preserved.[1]
Complex noun-phrase and wh-island cases supply classic contrasts. In an extraction from a relative clause inside a DP, the DP and CP boundaries can both intervene, with no licensed analogous escape through the DP edge. In an embedded question, an intermediate wh-position may already be occupied. Under a particular inventory, either configuration can force an overlong step. These are model analyses of sentence judgments, not assertions that every apparent extraction contrast has this sole cause.[1][2][3]
Clarity¶
Consider a question roughly of the form “Who do you believe Mary said Sam will visit?” The wh-phrase who is pronounced near the front but is interpreted as the visitor. A one-jump representation would cross multiple clause boundaries; the successive-cyclic analysis instead routes it through intermediate clause edges. The rule evaluates those steps, not the total span of words.[1]
Now place the gap inside “the claim that Sam will visit ___.” A DP containing a CP stands between the gap and the outer question position. In the classic example, the phrase cannot escape via a DP edge as it can via the relevant CP edges. The next move crosses both DP and CP and is disallowed in the taught formulation. The explanation depends on the assumed constituent structure, bounding-node set and available landing sites.[2]
Manages Complexity¶
The constraint compresses several classic extraction restrictions into one local accounting rule: count designated nodes crossed by each movement operation. A linguist need not invent a wholly unrelated ban for every long dependency. At the same time, compressing distinct acceptability patterns into one principle risks obscuring differences in gradient judgments, meaning and processing load. The right use of the rule is to derive a particular contrast and test it, not to infer that all island phenomena have been exhaustively unified.[1][3]
The parameterized bounding-node inventory also exposes a trade-off. Changing which categories count can improve one set of language comparisons while altering predictions elsewhere. Thus a clear analysis records the inventory and each intermediate step, rather than just saying “subjacency applies.”[2]
Abstract Reasoning¶
Draw the proposed constituent tree and identify the wh-phrase's interpreted base position and final pronounced position. State the bounding-node inventory. If intermediate edges are available, decompose the dependency into successive steps and count crossed nodes for each. If a step crosses two designated nodes, the derivation violates subjacency in that formulation. If no step does, subjacency does not by itself rule out the sentence; another grammatical or processing factor might still matter.[1][2]
This last point makes the rule falsifiable in a useful way. An unacceptable sentence that satisfies the count is not explained by subjacency alone. A highly acceptable sentence that the formulation bans pressures its tree, node inventory, landing-site assumptions or the applicability of the theory itself. The comparison should use actual acceptability evidence rather than a label attached after the fact.[3]
Knowledge Transfer¶
The transferable analytic habit is to distinguish overall distance from locally composed steps. In a syntactic derivation, a chain can be long even if each move is short. However, importing the exact one-bounding-node rule into other linguistic frameworks or languages requires arguments for movement, constituent categories and available edges. The MIT notes document variant inventories, and Rizzi's historical work is associated with cross-language differences; this entry does not turn a specific Italian judgment into a universal rule without the original full analysis.[1][2][4]
Subjacency remains domain-specific because its countable objects are syntactic bounding nodes within a particular model. A generic locality metaphor does not carry the same empirical predictions.
Examples¶
Licensed wh-chain through two embedded clauses¶
The MIT handout's (2a)/(14a) licensed question has the bracketed shape “Who do you believe [that Mary said [that Sam will visit __]]?” Here who is interpreted as the visitor. In the handout's CP/DP inventory, the proposed chain goes from the object gap to the nearest embedded Spec,CP, then to the next clause edge, and finally to the matrix question position. Each individual step crosses one CP, not two; the whole chain spans several clauses. This is the source's theory-internal derivation, not a claim that CP stops are audible in the sentence.[1]
Mapped back: mover = who from the visit-object gap; counted nodes = CP and DP in the handout's historical version; escape sites = the two intermediate CP edges; step count = one CP per move; verdict = permitted by this subjacency analysis.
Extraction from a sentential subject¶
The same handout marks its (3)/(13) attempted subject extraction with a star: “Who did [[that Sue spoke to __] surprise me]?” The gap lies in a subject CP; the historical analysis also treats that subject as dominated by a DP. There is no available intermediate escape stop between that CP and its containing DP, so the movement from the gap toward matrix *who crosses both CP and DP in a single required step. The star records the source's judgment; the two-node diagnosis is its theoretical account, not an observed material barrier.[1]
Mapped back: mover = who from the complement of spoke to; nodes = subject CP and containing DP; missing route = no permitted step that stops after only the CP before leaving the DP; verdict = violation of the one-node limit in the specified model. A different constituent analysis would need its own count.
Near miss: a merely long dependency¶
A dependency spanning three embedded clauses may look “nonlocal” on the page. If the proposed movement chain passes through an available edge in each clause, no individual step crosses two bounding nodes. Surface length alone is not a violation.[1]
Structural Tensions¶
The apparent opposition between distant interpretation and local derivation is resolved by successive cyclicity, not a cost tradeoff: the licensed question spans several clauses because it has several short steps. Diagnostic: can every proposed intermediate step be licensed and counted?[1]
Unified rule versus empirical coverage is a genuine model-choice pressure. A compact one-node condition derives the licensed-chain versus subject-island contrast with one inventory, reducing ad hoc prohibitions; applying it as an exhaustive account risks predicting categorical failure where judgments are graded or a different factor is active. Enlarging the inventory of exceptions may improve local fit but erode the proposed unification and make the theory less predictive. Diagnostic: with the same declared tree and nodes, which additional constructions does the rule correctly classify, and which require independently motivated qualifications?[1][3]
Structural–Framed Character¶
Subjacency is a formal structural constraint embedded in a contested theory of syntax. Once a tree, movement chain and bounding-node inventory are fixed, a step's node count is definite. Evaluative weight enters when linguists judge how well that analysis explains an acceptability pattern, not in arithmetic over the stipulated tree. Human practice supplies elicited judgments, argument structure and tree proposals; historical syntactic scholarship introduced the name and inventory, but no institution makes a speaker accept or reject a sentence by decree. The vocabulary travels literally among analyses that specify movement steps and counted nodes, though inventories can differ; calling any obstacle to communication “subjacency” would import a metaphor. Recognizing a new sentence whose independently motivated derivation obeys the same one-node check is literal within the framework. The structural calculation must remain distinct from the stronger empirical claim that it alone explains every island effect. Its character: a precise but theory-framed locality test whose explanatory value is answerable to linguistic evidence.[1][3]
Structural Core vs. Domain Accent¶
The skeletal relation is local-step accounting over designated boundaries; a broader locality abstraction might bear that skeleton only after independent multi-domain admission. The indispensable domain mechanism is syntactic movement chains, CP/DP or IP/DP categories, Spec,CP landing sites and wh-extraction judgments. Other domains may have bounded-step paths, but without syntactic movement and bounding nodes they do not instantiate subjacency itself. The named entry fails the prime bar because its one-node prediction is typed to this historical grammatical framework, not a portable constraint on arbitrary transformations. No broader parent is asserted merely from the word “local.”[2]
Instantiates / Related Primes¶
This entry is a kind of Constraint.
Constraint is the accepted strict genus: subjacency restricts a syntactic movement step to crossing at most one designated bounding node in its historical framework. This explanatory rule is not a physical barrier or a universal account of every island effect. Prime Non-Locality is related motivation, not a parent.
Relationships to Other Abstractions¶
Current abstraction Subjacency Domain-specific
Parents (1) — more general patterns this builds on
-
Subjacency is a kind of Constraint Prime
Subjacency formally restricts syntactic movement steps across designated bounding nodes, a specialized constraint.The staged linguistic principle rules out or limits certain syntactic dependencies across specified bounding nodes, directly instantiating live Constraint's admissibility restriction. Other constraints operate in different domains and need not involve movement or bounding nodes. The edge classifies a formal explanatory restriction, not an independently observed physical barrier.
Hierarchy path (1) — routes to 1 parentless root
- Subjacency → Constraint
Neighborhood in Abstraction Space¶
Subjacency sits in a moderately populated region (55th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.
Family — Formal Grammar & Syntactic Structure (16 abstractions)
Nearest neighbors
- Chain Shift — 0.87
- Adjunct (Grammar) — 0.87
- Cross-Serial Dependencies — 0.86
- Pursuit Curve — 0.85
- Yoked Control Design — 0.84
Computed from structural-signature embeddings · 2026-10-08
Not to Be Confused With¶
An island effect is an acceptability pattern; subjacency is one syntactic explanation. Successive cyclicity is a route through intermediate positions that can satisfy subjacency, not the same rule as the bounding-node limit. A phase-theoretic or processing account may analyze similar data with different primitives. The CP/DP versus IP/DP choice is part of a formulation, so a claim using “the” bounding nodes without naming its version is underspecified.[1][2][3]
References¶
[1] Anagnostopoulou and Fox, MIT Advanced Syntax, “Subjacency” notes (2007), pp. 2–4. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m ↩n ↩o ↩p ↩q ↩r ↩s
[2] MIT Language and Its Structure II syntax notes, pp. 5–6. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j ↩k ↩l ↩m
[3] Liu and colleagues, original synthesis of island-effect explanations (2022), §1.2. Original Chomsky texts not checked here. registry ↩a ↩b ↩c ↩d ↩e ↩f ↩g ↩h ↩i ↩j
[4] Rizzi, Issues in Italian Syntax (1982), publisher record; full book not checked. registry ↩