principle that violates the Head Constraint: through a specification in the argument structure ofseem, its syntactic
subject is“referred down”to the subordinate clause tofind its correspondence with a semantic argument. (This is
formalized in different ways in different theories, but the basic idea is the same.)
As in many situations we have encountered, then, the issue is a trade-off between different rule types. On either
approach, the complexity of language does not go away; it just reveals itself in different ways. In particular, in either
approach, the possibility of this“raising”construction depends on a particular lexical property ofseem.
(55) represents a different type of violation of the Head Constraint, due to a long-distance dependency (discussed in
sections 1.7 and 3.2).Whois a semanticargumentof seebut is notwithin thesame clause. The“normal”position of the
object is a gap notated by∧.
(55) Who do you think [that John saw∧]?The reason this construction is called a long-distance dependency is that the gap can be deeply embedded within
subordinate clauses, as in (56).
(56) Which book do you think that [Harold told Miria m[to get Richard [to read∧next]]]?This construction,alongwithmanyotherssuchas indirectquestions, relativeclauses, topicalization, exclamations (How
elegant Bill seems to have become!), and comparatives (Fred is richer than Bill is thought to be), is licensed notbyparticular verbs,
but by conditions on the position of the gap and the way each successive clause is embedded in the one above it. The
Sentential Subject Constraint, which we met in section 3.2.3, is one such condition; others were illustrated though not
named in section 1.7.
In all of Chomsky's successive approaches, (55) satisfies the Head Constraint in underlying syntactic structure: thewh-
phrase is taken to be generated in D(eep) Structure in the position of the gap, then moved to the front. By contrast,
non-derivational theories generate thewh-phrase directly in its surface position, but place a“silent pronoun”or other
formal marker in the position ofthegap. Thus for these theories, long-distancedependencies are genuineviolations of
the Head Constraint. Whichever account is adopted, these violations of the Head Constraint have quite a different
character fro mthe raising cases.^74
146 ARCHITECTURAL FOUNDATIONS
(^74) Partlyas a matterof historicalaccident, long-distance dependencieshavetraditionallybeen characterized as a basically freerelationbetween aposition at thefrontand a gap
anywhereina sentence;thetask takenonbyRoss(1967) and mostsubsequentliterature has beentofind constraints thatrestrictthatfreedomappropriately. Postal(1997) ,
though,suggests thatitmay actually besimplerto characterizepositivelyjustthosepositionswheregaps are permitted.Theminimal casewouldprohibitgaps anywhere,and
a language woulddeviate fro mthisun marked case toso me degree or anotherdependinghowfreeitslong-distance dependenciesare. This proposal,ifitcan beworkedout,
has the distinct advantage of permitting learning by positive evidence and thereby simplifying UG to a degree.