The EPP and the ECP
The EPP and the classical ECP have the common property of singling out the subject position of clauses: the EPP says that the subject position is obligatory; the ECP says that the subject position is unmovable (in certain environments that permit object extraction, such as embedded clauses).
(1)

(2)

We can now connect the two properties: if the EPP requirement can be reduced to a criterion, we expect subjects to be unmovable, under Criterial Freezing. Who must move to Spec,Subj to fulfill the Subject Criterion, but then it is frozen there, unavailable for further movement. Subject-object asymmetries thus reduce to the fact that there is a Subject Criterion, but no Object Criterion.
The possibility of moving the subject in certain environments now amounts to the ability of the thematic subject to avoid moving to the EPP position, thus escaping the freezing effect. The most straightforward way to achieve this result is to fill the criterial subject position with an expletive, thus making the thematic subject free to move from a lower position accessible to movement. The traditional analysis of the lack of asymmetries in null subject languages (Rizzi 1982, chap. 4) is straightforwardly interpretable in these terms: expletive pro fulfills the Subject Criterion in (3c); hence, the thematic subject remains available for extraction from a noncriterial position.
(3)

This option is also available in non–null subject languages—more marginally, though, because of the more stringent limitations on the occurrence of overt expletives like there.
(4)

(5)

The critical point of the derivation is this:
(6)

If DP is raised to Spec,Subj to satisfy the Subject Criterion, it is frozen there, under Criterial Freezing, and further movement as in (5a) is impossible. If the Subject Criterion is fulfilled by the expletive as in (4b), the DP subject of the small clause is accessible to further movement. Were one to adopt a theory of expletives as pro-predicates, as in Moro 1997, the result would be the same.
The strategies for subject extraction that different languages use can now be seen as strategies for avoiding the freezing effect connected to the Subject Criterion. See Rizzi and Shlonsky, in preparation, for an analysis of some major cases.