If recent proposals about
the syntax-semantics interface are valid, the syntactic structure proposed above
should allow to make specific predictions about the semantic differences of
Bantu sentences with or without noun class marking on the verb. Since noun class
inflection of the verb can be shown to correlate with VP-internal vs.
VP-external argument saturation, the Mapping Hypothesis of Diesing 1992 becomes
immediately relevant. According to Diesing, material from the VP is mapped into
the nuclear scope and receives an existential reading, whereas material outside
the VP belongs to the restrictive clause and is thus interpreted as
presuppositional. Traditionally, definiteness was thought to be the linguistic
indicator of presuppositionality, but Enç 1991 showed that specificity is a
more accurate measure and argued that the so-called 'definiteness effect' should
more aptly be named 'specificity effect'. Since the bulk of relevant Bantuist
literature predates this development, the issue surrounding specificity remains
the Bantuist literature, it has long been conjectured that there is at least a
strong tendency for noun class markers on the verb to correlate with
definiteness of the argument. Doke 1955: 10-14, for example, observes that
'an indefinite subject or object is not represented in the predicate by its class concord, but by an indefinite concord or by none at all...Representation of a subject or an object by an absolute...pronoun ensures that the subject or object concerned is definite...When the simple object is indefinite, it is not represented by any concord with the predicate...When the simple object is definite, the objectival concord agreeing therewith is incorporated in the predicate, though this concord may be omitted if the object is a definite pronoun or a noun with definite significance'.
The problem with this
concentration on definiteness (rather than specificity) is, however, that the
picture is less clear-cut than one would hope for. Bresnan/Mchombo 1987 show
that direct objects may be interpreted as definite or indefinite regardless of
the presence of corresponding verbal inflection:
when the data are reinterpreted in terms of specificity, as defined by Enç 1991
as a form of discourse linking, i.e. presuppositionality, Bantu languages
provide rather nice evidence in favor of Diesing’s mapping hypothesis. Notice
that (24b), which according to Bresnan/Mchombo serves as counterevidence for
correlating verbal noun class marking with definiteness, presumably requires
some previous discourse in order for the interpretation of the noun class prefix
as ‘one’ to be felicitous. Thus, the relevant argument may not be definite,
but it is specific as expected. Keach 1995: 114 states explicitly for
Kiswhaili that object marking is usually topic-bound, i.e. presuppositional.
Since the correlation
between presuppositionality and verbal inflection thus seems to be strong, the
following semantic mappings are predicted from the syntactic structure:
a. In ‘impersonal
constructions’, both subject and object are non-specific. ku- in [Spec,
TP] blocks the subject from raising out of VP, which in turn blocks raising of
the object (in form of object NC). Since non-specificity rules out inherently
specific definite interpretations, common nouns in impersonal constructions are
predicted to be indefinite in translation, which is supported by the facts:
b. If subject and object
agree with verbal inflection, they must be specific. A specific interpretation
can in principle accomodate both definite and indefinite noun phrases, but
definite translations are preferred.
c. If the subject is specific and no object
marking is present on the verb, the object is predicted to be ambiguous. Without
overt evidence for movement to [Spec, AgrOP], the object will be interpreted as
in situ, i.e. VP-internal, and thus as non-presuppositional and non-specific. LF
movement to [Spec, AgrOP] is, however, still a possibility, such that a specific
interpretation remains an option. This explains the ambiguity between definite
and indefinite English translations.
 If the translations are definite, it provides nice support for my theory. However, one should keep in mind that English definiteness marking does not exactly match the specific vs. nonspecific distinction in Bantu languages. In translations where specificity is not considered relevant, the natural choice for nonspecific phrases in Bantu would be an indefinite description in English, while the translator would probably gravitate towards an English definite for Bantu specifics, even though indefinites would also be possible in suitable contexts. I believe that the following example provides such a context:
Babe tinja u-to-ti-nika ematsambo kusasa
father 5pl-dog subj-fut-obj-give 3pl-bone tomorrow
‘Father will give dogs bones tomorrow.’
Mahajan (1991) observes that scrambling in Hindi correlates with specificity. tinja in the above example is left-adjoined to TP, which essentially corresponds to IP-scrambling and has similar topicalization effects (as the translations of parallel examples show in Thwala, e.g. 40c). I will thus assume that this example does not provide counterevidence against my proposal.
 Keach makes a distinction between "animate" (class 1/2: reference to humans and animals) and inanimate class marking. While inanimate object marking requires a topic object, the animate object marking may cooccur with a non-topicalized overt object. While Keach takes this to be a characteristic of agreement, i.e. of a different functionality of class marking, I would nevertheless maintain that this behavior does not contradict my proposal, and a unified analysis remains the simpler alternative.