Rizzi proposes an account for the asymmetries between adjuncts and arguments, based on a notion of "Relativized Minimality," where governement relations are blocked only if the intervening element has the correct type. This account is meant to explain differences in the acceptability of raising arguments and adjuncts out of various constructions, such as:
Negative operators block extraction of adjuncts:1a. How do you think that John talked to Mary t?1b. Who do you think that John talked to t?2a. How don't you think that John talked to Mary *t?2b. Who don't you think that John talked to t?
Wh Islands block extraction of adjuncts:3a. * How do you wonder whether John talked to Mary t?3b. ? Who do you wonder whether John talked to< i>t*?
Super-raising is blocked in A chains:4a. * John seems that it is likely to win.4b. * Advantage seems that it is likely to be taken t of John.4c. * John seems that it is likely to be taken advantage of t.
A government relation between X and Y is blocked if there exists Z such that: (a) Z is phrase-internal (not an adjunct/thematic phrase); (b) Z is a position of the same type as X; (c) Z c-commands Y; and Z does not c-command X. The possible "types" for (b) are head position, maximal projection in A position, and maximal projection in A' position.
(2a) is blocked because negP's spec is filled by a negative operator; (3a) is blocked because "whether" is in spec/CP; (4a-c) are blocked by "it" in spec/IP.
To explain why (2b) is ok, Rizzi claims that argument variables and their operators are connected by a long-distance relation, using a referential index. This long-distance relation lets them bypass reletivized minimality. But adjunct traces don't have referential indices, so they can't use this alternate mechansm. In the case of super-raising (eg 4a), referential indices aren't available because traces aren't arguments, and so they can't bear referential indices.
After introducing this basic account, Rizzi shows how it gives a consistent evaluation of the grammaticality differences (or lack thereof) between adjuncts & arguments in German and several other languages.
The account proposed by this paper seemed fairly stipulative to me; Rizzi had to invoke a fairly large set of theoretical devices (eg the referential indices) to make things work out right. It's unclear to me whether these devices are well motived, independantly of this problem.