Island constraints prohibit extraction from certain syntactic domains (relative clauses, coordinate structures, adjuncts), limiting where wh-constituents can originate. Subjacency formalizes this by requiring moved elements to cross no more than one bounding node per derivational step.
Test each major island type systematically with grammaticality judgments; compare with free extraction contexts to identify the domains that block movement.
Islands are not absolute barriers; pragmatic context, processing cost, and language variation affect island sensitivity, suggesting the constraints are syntactic violability principles rather than hard rules.
Your prerequisites in movement and transformations, and in X-bar theory, give you the tools to understand island constraints. From movement, you know that syntactic derivations can displace constituents from their base positions — a wh-word like "what" can originate as the object of a verb deep in the structure and move to the front of the sentence: *What did she buy?* (from the underlying *She did buy what*). From X-bar theory, you know that phrases are built hierarchically, with specifiers, heads, and complements. Island constraints are restrictions on where that movement can begin — or more precisely, they identify certain syntactic domains from which extraction is impossible or severely degraded.
Consider three classic island types. First, the relative clause island: *What did she meet the man who bought?* — ungrammatical. The object of "bought" is trapped inside the relative clause "who bought ___". You cannot extract it. Compare: *What did she meet the man who bought yesterday?* — still ungrammatical, no matter how the sentence is arranged. Second, the coordinate structure island: *What did she buy a book and read?* — trying to extract the object of only one conjunct of a coordination is impossible. The "Coordinate Structure Constraint" requires that you extract from both conjuncts equally or neither. Third, the adjunct island: *What did she leave before she bought?* — extraction from the subordinate adjunct clause is blocked.
Ross (1967) identified these patterns empirically; Chomsky's Subjacency Condition (1973) attempted to provide a unified formal explanation. Subjacency states that a moved element cannot cross more than one bounding node in a single derivational step. Bounding nodes are typically specified as NP and S (or IP in later frameworks) — the main clause nodes that create structural barriers. A relative clause is embedded inside an NP, which is inside another S, so extracting from a relative clause would require the moved element to cross two bounding nodes in one step — a subjacency violation. The elegance of subjacency was that a single formal principle could explain multiple empirically distinct island effects.
However, subjacency immediately faced problems. First, which nodes count as bounding nodes varies across languages — in Italian and Spanish, extraction from relative clauses is somewhat more acceptable than in English, which is difficult to explain if islands are purely syntactic. Second, not all island effects are equally strong: extracting from a Complex NP ("What did you believe the claim that she bought?") feels worse than extracting from a simple adjunct, and these gradations are hard to capture in an all-or-nothing syntactic constraint. Third, processing accounts have argued that "island violations" partly reflect working memory costs of maintaining long-distance dependencies through complex structures, not a categorical grammatical prohibition.
These complications have led to several successor theories: the Phase Impenetrability Condition in Minimalism treats phases (vP and CP) as the relevant barriers; Optimality-theoretic approaches treat islands as violable ranked constraints; experimental linguists have shown that acceptability judgments are gradient and context-sensitive, not binary. What survives across all these approaches is the core empirical observation: human languages systematically restrict movement from certain syntactic domains, and these restrictions are not arbitrary but fall into identifiable categories. Island constraints are thus a window into the nature of syntactic computation — evidence that grammar has internal architecture that limits what operations can reach what positions, making syntax not just a combinatorial system but one with structural memory of its own derivational history.