The semantic effects of sequence of tense are known to differ as between object relative clauses and complement clauses, both types having been discussed in original work by Bill Ladusaw. Thus, whereas (1) is fine, (2) is absurd as an indirect discourse report:

(1) Mary found a unicorn two days ago that was walking yesterday.
(2) Mary said two days ago that Bill was in New York yesterday.

The reason for the status of (2) cannot be that Mary's utterance, say "Bill will be in New York tomorrow," fails to match the speaker's embedded clause in truth conditions --- for, manifestly, it does. Nor, I will argue, can it be for metaphysical or purely semantic reasons, as suggested (in different ways) by both Toshi Ogihara and Dorit Abusch (and by me in earlier work). Finally, the remark that is sometimes made, that it is a matter of "scope," is not coherent as applied to (2) (as Ladusaw already noticed); and, I will argue, even the scopal solution proposed by Ladusaw, and modified in Ogihara, for the acceptability of the relative clause case (1) fails in general.

In earlier work, I argued that the ambiguities in (3) and (4), for example, followed from various possibilities for anaphora between the matrix and embedded implicit arguments of Tense:

(3) Mary found a unicorn that was walking
(4) Mary said that a unicorn was walking

Michela Ippolito takes the argument further, in connection with the Italian imperfect and other constructions. If so, then the problem of (1) vs. (2), or the 3-way ambiguity of (3) versus the 2-way ambiguity of (4), can be put as follows: tense-anaphora are obligatory in (4) and the like, but optional for (3) and the like; and the question is why. The answer, following up on an original suggestion of Tim Stowell, and partly developed in recent work by Alessandra Giorgi and Fabio Pianesi, is one that combines syntax and semantics, exploiting a certain interpretation of movement rules as "copying."

The perspective I adopt, however, is quite the inverse of that in Giorgi and Pianesi, in that I suppose that it is the copy of Tense in the complementizer position C of a complement that is anaphoric to the matrix Tense. But, as will be seen, the hypotheses of my discussion taken together imply not only the asymmetry between complement and relative clauses, but also the semantics of the so-called "double-access" cases.