The structure of null subject DPs and agreement in Polish impersonal constructions 151
argument but some kind of a functional element. Here I assume, more specifically, that
it is a functional element that heads its own projection dubbed SIĘP. The functional
head SIĘ takes as its complement a bundle of unbound but interpretable phi-features,
i.e. pro. This is illustrated by the figure in (30).
(30) SIĘP
SIĘ proThe whole complex is merged as an external argument in [Spec,VoiceP]. As a func-
tional element, SIĘ is devoid of any interpretable Φ-features. I assume that it is also
devoid of any valued uninterpretable features, and as such it does not constitute a
viable goal for T. This means that it blocks neither the movement of pro to [Spec,TP]
nor the agreement with T. It is assumed that pro has to move out of the VoiceP (which
is located between the TP and the vP and is assumed to be a phase) to the TP, as it
needs to be available for binding in the domain over which various operators located
in the CP take scope (Diesing 1992).^26 If pro were to remain in situ in the VoiceP, it
- This assumption runs counter to the one explicated in Holmberg (2010a) and according
to which, pro (a ΦP; deficient pronoun in Holmberg’s (2010a,b) terms does not raise to
[Spec,TP] but stays in [Spec,vP] from where it values the features of T. Because Holmberg
(2010a) assumes that the features of a ΦP are a proper subset of those on T, they get incor-
porated into T. This operation is reminiscent (a) of head movement, as noted by Holmberg
(2010a), and (b) of cliticization if it is assumed that clitics are Xmin/max at the same time in
accordance with the BPS (Roberts 2010). There are, however, a few problems with this account.
First, this suggests that agreement on T is pronominal (Alexiadou & Anagnostopoulou 1998)
which means that it is referential and therefore capable of picking out a unique referent. In
other words, the valued phi-features on T are phi-features of a referent that a nominal expres-
sion is used to point to. It is clear that even in consistent null subject languages it is not always
the case that the properties/phi-features of a subject are marked on T even if feature syncre-
tism is taken into account. An example of that are impersonal constructions in Polish and
other languages for that matter. That is, the 3sg/3sg.n and the –NO/–TO inflectional marking
on T certainly does not reflect a very rich feature specification of null impersonal pronouns.
What is needed to recover that feature specification is the presence of elements such as topics
both overt and covert, various operators, and logophoric features that bind and/or provide
values for complex variables that pronouns are considered to be. Secondly, although there is a
high degree of resemblance between incorporation and cliticization, it is clear that the two are
to be kept separate. The reason for this is that phi-specification of a pronoun cliticized onto a
V+T complex is always recoverable whereas this is not the case for null impersonal pronouns,
as argued above. Thirdly, if it is assumed, as Holmberg (2010a,b) does, that T has an unvalued
uninterpretable [uD] feature that refers to definiteness or the lack of it, it is not clear how the
pronominal inflection on T can be in any way referential if it is uninterpretable.