Consequences
We have presented a number of related arguments that wh-trace is not subject to Opacity. Chomsky (1980) is largely compatible with this position, but has one argument to the contrary, to which we now turn. Note that here we follow Chomsky’s presentation in treating Opacity as conditions on logical form rather than as reindexing rules. The argument could readily be restated in the latter format, that of the appendix to Chomsky (1980).
Comparing (30) to (31), Chomsky notes that both violate Subjacency with respect to binding to the matrix COMP, yet (31) is somewhat more acceptable.
(1)
(2)
If wh-trace were subject to the NIC, then the contrast could be described as follows. While both (1) and (2) violate Subjacency, the complement subject in (1), but not in (2), would also be subject to the NIC. (Note that if the PIC is taken to be the relevant condition, (1) and (2) would not be distinguished in this way since the PIC would affect the complement object as well.) Thus, (1) violates two conditions, whereas (2) violates only one—granting that wh-trace is not subject to the SSC.
The validity of this account requires that Opacity be split into two quite distinct constraints. The NIC must be taken “to be an “inviolable” constraint, as compared with Subjacency (and [the SSC]…),” the latter two being “weaker” constraints with wh-trace as anaphor (Chomsky (1980, 37–38)). The small difference in acceptability between (1) and (2) seems rather slim evidence for such a weakening and complication of the theory. It should also be noted that there are examples quite parallel to (1) in acceptability, but identical to (2) in relevant derivational properties. Consider (3):
(3)
While (3) is just as unacceptable as (1), like (2) it does not involve the NIC, regardless of how wh-trace is to be treated. (4) demonstrates that the unacceptability of (3) cannot be attributed to a crossed binding constraint of the type frequently discussed in the literature (cf. Fodor (1978)).
(4)
Though (3) is of the form x1 x2 y1 y2, while (33) is of the form x1 x2 y2 y1, they are equally unacceptable—some new constraint, which presumably would generalize to (1), is required.
Thus, if there is in fact a linguistically relevant contrast between (1) and (2), the NIC does not seem to be responsible. Given that wh-trace is not subject to the NIC (PIC), it follows that the *[that–e] filter of Chomsky and Lasnik (1977) cannot be reduced to the NIC as proposed in several recent articles and unpublished papers. In fact, given our conclusions above, there is no overlap at all between the filter and the condition. Thus, one argument for eliminating the filter, namely its alleged partial redundancy with the condition, is without force. Another consideration that has come up in discussions of the filter is its complexity. The “unless”-condition particularly has been criticized.
(5)
If this complexity actually is a problem, it is interesting to note that it can be completely eliminated. The purpose of the entire “unless”-clause is distinguishing between (on the one hand) that verbal complements, which always lead to ungrammaticality when in violation ((6a)), and (on the other hand) relative clauses and clefts ((6b)), which never lead to ungrammaticality when in violation.
(6)
Thus, an alternative to the “unless”-clause is the reasonable separation of complementizer that into two lexical items. The filter would then be stated in terms of the verbal complementizer that and would need no “unless”-clause at all.
The conclusion that a wh-trace is not subject to either the SSC or the PIC bears on the analysis of strict cyclicity whereby the empirical effects of the Strict Cycle Condition are derived from independently motivated conditions on representations. In Freidin (1978), two accounts of strict cycle violations involving Wh Movement as in (7) are proposed.
(7)
One account assumes that wh-trace is subject to the PIC and SSC. Thus, (7) is prohibited because e1 in (7a) is subject to the PIC and e1 in (7b) is subject to the SSC. Alternatively, the binding between e1 and who1 in both examples violates Subjacency. As argued above, the first alternative is not viable.
This situation provides an argument that Subjacency is properly construed as a condition on representations, and not a condition on movement. Note that (7a) can be derived in a way that does not violate Subjacency interpreted as a condition on movement, as in (8).
(8)
(A similar derivation can be provided for (7b).) Such derivations could be excluded given the Strict Cycle Condition (SCC) (see Chomsky (1973) and Freidin (1978) for details). However, the SCC is redundant for NP Movement, as shown in Freidin (1978), and unnecessary generally if Subjacency is taken to be a condition on representations rather than a condition on movements. It remains to be determined where in a derivation Subjacency applies.
