i.e. λx.wetalkedtothemanx. Furthermore, this is precisely the interpre-
tation that is needed for the CP. As is standardly assumed, relative clauses are
interpreted as predicates. However, if the PP remains in that position at LF,
that would not be the denotation of the CP. Rather, it would have the meaning
of a proposition, which is not the right type for a relative clause. Therefore,
the upper copy of the PP must be deleted at LF. The only (semantic) role that
the movement has in relative clauses is to create the predicate, as is standardly
assumed (see Heim and Kratzer 1998).
Note that this deletion essentially results in total reconstruction of the moved
phrase (cf. §3.2). The only difference with the cases of total reconstruction we
saw in §3.2 is that, crucially, in relative clauses the predicate created after the
movement remains a predicate after deletion of the upper copy.
5, 6
The fact that in relative clauses reconstruction is obligatory explains why
definite articles, as opposed to other determiners, can be used as relative op-
erators. Note that in (26), it is not necessary to apply Trace Conversion (cf.
§3.1), since the moved phrase is already a definite description, i.e. the man x.
If, on the other hand, some other determiner, such as quantifiers like each or
a, were to be used, Trace Conversion would be necessary. Trace Conversion in
this case would in effect destroy the chain created by movement, since the two
copies would now be different. As shown in §3.2, total reconstruction is blocked
in this context, due to the identity condition. Since, as shown above, total re-
construction is obligatory in relative clauses, we obtain the desired result that
quantifiers cannot be used as relative operators.
7
One possible objection to this analysis might go as follows. In the example
above, the lower copy must be interpreted as a definite description. Since it is
the internal argument of hablar ‘talk’, it must be of type e. However, consider
the case in (27), in which the quantifier each is used as a relative operator, and
it is heading a DP in subject position.
8
(27) [
CP
[eachboy]λx[
C
that [ each boy ] read his
x
copy of El Quijote]]
✻
Since the lower copy is in subject position and QPs can be interpreted in this
position, one might argue that Trace Conversion in this case is not necessary in
this case. Thus, deletion of the upper copy would not be blocked by the identity
condition on total reconstruction. Furthermore, in this specific example, the
lambda abstract created by the movement is binding a variable, i.e. the one
5
On this difference between standard cases of total reconstruction and the one argued for
here, see §4.3
6
This means that whatever operation makes C a predicate must take place before deletion.
A straightforward way to implement this is to assume, following Heim and Kratzer (1998),
that, as part of the movement operation itself, an index is adjoined to
C which turns it into
apredicate.
7
This raises the obvious question of how we can account for other kinds of relative operators,
such as wh-words (cf. §1and§2). Although, at the moment, I do not have a full answer to
this question, see §5 for some suggestions.
8
For the purposes of the argument, I am ignoring the movement of the subject from the
VP-internal position.
11