There is substantive evidence to show that Gender and Class can
be unified. The prefix type classifiers of the Bantu family show agreement
with the noun class they are attached to. This agreeing behaviour of certain
classifiers is evidence that class is a gender like category. Although
in South Asian languages the Classifiers do not morphologically trigger
or participate in patterns of agreement, there are overall typological
grounds for saying that Classifiers and Gender are two different shapes
of the same thing. We further conjecture that the loss of ergativity in
Eastern IA languages was accompanied by a loss of the system of gender
classification in these languages. Bangla and other Magadhan languages
are presumed to have developed the system of classifiers subsequent to
this attrition of the agreement system. With this sort of a background
we suggest a neutral term like Badge for this site since we are talking
about the noun's identity indications. In Class languages it houses the
classifier. In Gender languages it holds the Gender and Number information.
The importance of Badge is also reflected in its ability to respond to
the process of Case marking. It is a site where the Case information spills
over. There is evidence that noun subtype properties interact with theta/
Case properties in an obvious way. Larson (1985) has demonstrated this
for bare NP adverbs. He argued that adjunct NPs get their theta roles and
Case through the N since nouns have intrinsic semantic properties; that
is, certain lexical items get their theta properties from being what they
are and where they are. The idea that noun subtype information can feed
Case and theta features becomes easy to implement if we postulate a Badge
site housing noun subtype properties.
We observe that Bangla, but not Hindi, has a system of Classifiers and employs the postnominal placement of a classifier (with or without a numeral prefix) as a device which signals nominal definiteness. The account of these and related phenomena which we develop elsewhere (Dasgupta and Bhattacharya 1994) proposes that the syntactic environment of a noun phrase is conditioned by two forces. Its external grammar shows up at the Declension (or D) head of the DP and registers Case, definiteness and other relational properties. Its internal grammar resides in what we have called the Badge (or B) enclosure of the noun complex. The D slot mediates the instructions of the external world. The Badge decides how the internal world, that is, the N is to be organized.
Badge comprises Gender and Number in Gender languages and is formally
instantiated in the feature composition of the Noun. In Class languages
of the Classifier subtype, we suggest, Badge may be a site, between the
Noun word proper and the Declension, where a Classifier (with or without
a numeral prefix) may appear. On such an analysis, both Hindi and Bangla
manifest definiteness by strengthening Det (the Spec of B') or D in both
languages, or B in Bangla where it may be weak (zero B) or super strong
(overt). Options for B in Hindi are limited; gender is lexically fixed
and number depends upon speaker's choice. Thus Hindi cannot strengthen
B to mark definiteness. We propose that in both types of language the content
of B interacts with that of D to determine the often fused shapes of the
relevant inflectional morphology.
DP
3
Spec D'
3
BP D
2 Case
Spec B'
2
DP B
2 !
Spec NP Cla
Agr in D in Abney (1987) links DP to NP, that is, two NPs. The phi features of B proposed here have to do with just one NP, the one that the B serves. Presumably Bangla B, if null, has Chomsky "weak" (PF invisible) phi features, while Hindi B always has Chomsky strong phi features. Consequently a Hindi N must raise to B in the overt syntax to go through feature checking and to make the derivation converge as otherwise these Chomsky strong phi features of the Hindi B will survive till PF. These naked features are not legitimate PF objects and so the derivation will crash. Bangla raises N to B at LF because the Bangla B is, if overt, a classifier morpheme with a distinct phonological shape of its own and posing no PF licensing problems. If null, it has no PF visible material at all, again posing no problems and requiring no N raising in the overt syntax.