In the so-called
‘impersonal construction’, the subject DP does not trigger noun class
marking on the verb. Instead, a noun class 15 marker appears, which seems to
have expletive characteristics:
Since ku- also
servers as a locative marker, it seems to correspond to the English expletive
The important fact in (14)
is that the (non-agreeing) subject must occur in post-verbal position.
Kinyalolo 1991: 85ff. argues on the basis of the word order and agreement
facts of topicalization and relativization in Dzamba that the subject is
generated in [Spec, VP], and that a subject which fails to trigger verbal
agreement remains in situ. Thwala 1996 makes the same claim for Siswati and
presents the following word order facts:
As (16a) shows, the subject
cannot occur in preverbal position, so there is good reason to believe that it
has to remain within the VP. Presumably, ku- is generated in [Spec, TP]
and checks off all relevant features that would otherwise trigger
chain-formation between [Spec, TP] and [Spec, VP], where the subject is
seems to me that the word order issue has a straightforward solution if
non-thematic elements are allowed to adjoin to the right.
(16b-d) would have the structures in (17b-d) respectively (assuming that the
verb has moved out of VP):
Of course, there is no way
to tell for sure whether the adverb in (17d) is adjoined to VP or to V’. It
only seems clear that the adverb is not much higher than VP, given that
temporal adverbs consistently follow manner adverbs (cf. Thwala 213-216 uses kamalula
‘quickly’ and kusasa ‘tomorrow’).
(16f) shows that the object can precede the VP-internal subject, [Spec, AgrOP]
does not seem to be its location in such configurations, since VP-adjoined
manner adverbs cannot intervene between object and subject (cf. 16e). Since
the thematic hierarchy (and some version of the UTAH, cf. Baker 1996) would
disallow base generation of O-S order in normal transitive structures,
VP-internal object scrambling seems to be the only option. Crucially, the
arguments of the verb in all impersonal constructions are non-specific, which
is straighforwardly explained by the fact that they remain VP-internal.
 This is contra Thwala (1996: 230ff), who follows Kayne (1994) in insisting on strictly right-recursive structures. Among other problems, temporal (sentence) adverbs always occur on the right of the sentence, so Kaynian proposals would have to argue for complex "pied-piping" mechanisms in order to derive the correct word order. I cannot see what the motivation for such movements would be, so I stick to the traditional right-adjunction here.