(1) The book may be put on the table.
(2) *On the table may be put the book.

Optional Reading: Here's Vergnaud's letter, if you'd like to read it.

The problem with (1-2) is that, as things stand, we expect both examples to be well-formed. In both, the EPP is met, by movement of some phrase (an NP in (1), and a PP in (2)) to the specifier of TP. So we need some way of ruling out (2).

Vergnaud suggested that NPs have a special need which is not met in (2). In particular, he suggested that we should take advantage of another special property of NPs, namely that they often bear Case morphology:

(3)	a.	Yukarrbarda warnawu yak-in.	[Lardil]
		man-NOM     cook    fish-ACC  
		'The man cooked the fish'

	b.	Yaka     mutha.
		fish-NOM big
		'The fish is big'

(4)	a.	Gakusei-ga  hon-o    yonda.	[Japanese]
		student-NOM book-ACC read
		'The student read the book'

	b.	Hon-ga   omoi.
		book-NOM heavy
		'The book is heavy'

(5)	a.	Puer    vidit dom-um.		[Latin]
		boy-NOM saw   house-ACC
		'The boy saw the house'

	b.	Domus     magnus est.
		house-NOM big    is
		'The house is big'

In examples like the ones in (3-5), the nouns have suffixes on them that tell you what they're doing in the sentence; the subjects above, for example, all have Nominative case morphology (abbreviated NOM), and the objects have Accusative morphology (abbreviated ACC). So this Case morphology is apparently conditioned somehow by the NP's position in the sentence. Vergnaud's idea was that once we understood the relationship between positions in a sentence and particular types of Case morphology, we would also have a theory of why NPs can show up in some parts of the tree but not others; they can only appear where they can be assigned a Case.

In other words, Vergnaud was proposing a Case Filter:

Case Filter
An NP must have Case.

As part of developing a theory of Case, we will have to figure out where NPs can get Case. And one of the things we'll say is that a transitive verb assigns Case to its complement, but a passive verb does not. As a result, the NP object of a passive verb cannot get Case where it is, and must move. And we'll apparently also want to say that the specifier of TP is a place where NPs can get Case. Having said all that, we can handle the data in (1-2) above; once the verb put is passive, its NP object cannot get Case where it is, and must move to the specifier of TP, where it gets Case. In (2), the PP moves to the specifier of TP instead, and while that satisfies the EPP, it violates the Case filter; the object of put is left without Case.

Data like (1-2) are not the only reasons to posit a Case filter; there are a lot of places where NPs seem to have special conditions on their distribution. For instance, as we've been talking about selection, we've mainly been talking about verbs. Verbs can select a number of different categories:

(6) John likes [kumquats]
(7) John said [that kumquats are tasty]
(8) John lives [in a yellow submarine]

In (6), the verb likes selects for an NP complement; in (7), the verb said selects for a CP complement; and in (8), the verb lives selects for a PP complement.

Nouns and adjectives can also select for complements, and their complements can be CPs or PPs--but apparently not NPs:

(9) *John's fondness [kumquats]
(10) John's fondness [for kumquats]
(11) John's belief [that kumquats are tasty]

(12) *John is fond [kumquats]
(13) John is fond [of kumquats]
(14) John is sure [that kumquats are nutritious]

We can describe facts like these by saying that verbs, but not nouns or adjectives, can assign Case to their objects.