The structure of null subject DPs and agreement in Polish impersonal constructions 149
move to D, as argued by both Cardinaletti (1993) and Progovac (1998). Instead, I will
claim, following the tenets of the Bare Phrase Structure, that the projection of D has
all the properties of D and because of that, the index [iR] that is bound by either one
of the speech features or by an A-topic will be present also on Dmax, which means that
it is the whole phrase that will carry the referential index. And if this is the case then,
there is no need for the pronoun to move to D, as the pronoun will be inside a phrase
that is bound by a referential element.^22 In other words, 3rd person pronouns will
become referential when [iR] on Dmin/max gets bound by an A-topic in the CP. As for
the 1st/2nd person pronouns, there are two possible options to consider. According to
the first one, [iR] gets bound by either the [speaker] or the [hearer] feature in CP, and
from there the index spreads onto pro as one of the subfeatures of the [participant] fea-
ture, that is either the [+speaker] or [+addressee]. According to the second possibility,
speech features bind both [iR] on Dmin/max and either [speaker] or [addressee] feature
at the same time.
Although at first sight the first proposal seems more appealing, the discussion
below reveals that it is the second one that should be adopted. We could then say
that both an A-topic and speech features in the CP bind all indices/features of the
same type within a given domain, similarly to generic operators that can bind a num-
ber of variables within a domain (Chierchia 1995a,b). The problem, however, with
both approaches is redundancy. In other words, the same feature with the same value
will be present on both elements Dmin/max and pro, as both [iR] and the [participant]
feature on pro will be marked as [+speaker]/[+addressee]. I will argue, however, that
this apparent redundancy is a necessity. First of all, if it is assumed, as I do here, that
pronouns are N heads in the NP complements of D and enter derivation as unvalued
bundles of features, then [iR] will be present on D from the very beginning. Secondly,
it appears that [iR] is necessary for a pronoun to be fully referential, even in the case
of the 1st and 2nd person pronouns. Without [iR] bound by the [speaker]/[hearer]
speech feature, it will not be able to surface as the 1st/2nd person pronoun. With just
the [speaker]/[addressee] feature bound on pro, the interpretation will be that of inclu-
sive/exclusive. It is only when [iR] on Dmin/max is bound as well, then the reading will
be that of precisely 1st/2nd person.^23 We could then further hypothesize that there are
two kinds of Dmin/max, one with [iR] and the other without it, and as is well known both
types can be overt or covert. The interpretation of a pronoun will depend on which one
- I remain agnostic as to whether the pro (or an overt pronoun) merged as complement of
D moves to [Spec DP] or not in the course of derivation. - See Choi (2012), for a similar hypothesis, based on Modern Greek data, that the semantic
person feature is encoded in D of pronominal DP.