Government and Binding Theory/Barriers

Yet Subjacency is not sufficient to explain everything. Consider these examples:

(6a) [CP   —   —   [IP [CP   —   that John has found what1] is2 strange]]
(6b) [CP   —   —   [IP [CP what1 that John has found  t1  ] is2 strange]]
(6c) [CP What1 is2 [IP [CP   t1  that John has found  t1  ]  t2 strange]]
(7a) [CP   —   —   [IP I do2 think [CP   —   [IP John has found what1]]
(7b) [CP   —   —   [IP I do2 think [CP what1 [IP John has found  t1  ]]
(7c) [CP What1 do2 [IP I  t2 think [CP  t1   [IP John has found  t1  ]]

(6) and (7) are essentially the same story. An element moved across an IP node, left a trace and moved cross another IP node. Under Subjacency, both are correct. However, a native speaker can intuitively tell that (6) is categorically wrong, and (7) is correct. There is but one fundamental difference between the two: in 5a, what is moved from the subject position, and in 5d, what is extracted from the complement position.

Barriers

Chomsky has thus suggested the Barriers theory, which explains why it is easier to move an element out of a complement than a subject:

Bounding Principle under Barriers A constituent is L-marked if it is subcategorised by a lexical head. A constituent is a Blocking Category if it is not L-marked. A constituent is a barrier if it is a Blocking Category other than IP. A CP inherits barrierhood from the IP and it is a barrier unless its specifier position is filled, in which case the it is not a barrier if not L-marked. A movement may not move across a barrier.

We can look at our previous examples to verify this principle. In (6), the CP is not subcategorised by the head but by its first projection, so it is not L-marked and will be a blocking category under any situation, preventing movement. In (7), the IPs are of course not barriers. The CP is subcategorised by the head, so after what moves to its specifier position, there aren't any barriers in sight. The movement is deemed grammatical.