<<   >>   TOC

3.2 Non-Nominal Constructions


I argued above that simplicity suggests that class markers are always elements of category D. Hence, adjectival constructions would have the following general form:


Since this introduces a further referential pro into a given sentence structure, it has to be assured that no binding violation can occur. I will argue below that Bantu languages are largely non-configurational, i.e. certain thematic and all non-thematic elements occur in adjoined positions. Thus, syntactic dependencies directly reflect semantic dependencies. Since pro is embedded in an adjunct, it cannot govern into the remainder of the sentence. Overt lexical arguments are shown to be either VP-internal or themselves in adjoined position. In the first case, the overt argument is lower than pro, and in the second case, the overt argument may not be a governor. Thus, a pro embedded in an adjunct cannot be governed by an overt NP either, which means that binding will not pose a problem. Numerals, possessives, relative pronouns etc. can also occur with noun class marking: I assume that the basic structure is essentially the same as for adjectival constructions.

            A verbal complex containing a noun class marker (i.e. subject or object "agreement") would be constructed along the same lines, ignoring surrounding details:


This analysis provides an explicit answer the question around which much of the relevant literature revolves: are noun class markers pronouns or do they license pro? I claim that the latter is the case. However, while typical GB analyses tend to assume that NCs are Agr and license pro in their Spec position, I claim that NC and pro form an argument together and are thus both located in [Spec, TP] or [Spec, AgrOP], depending on their function.

            A potential problem for this analysis is the fact that pro is commonly regarded as [+pronominal, -anaphor]. According to Baker 1996, this circumstance explains straightforwardly that Mohawk does not have independent anaphors like reflexive her/himself or reciprocal each other, since they would create binding violations with the pro's introduced by the verbal complex. Like Mohawk, Bantu languages usually employ reflexive/reciprocal verbal affixes instead of independent anaphors (see e.g. Chichewa in Mchombo 1993). However, Zulu also seems to exemplify selection for an anaphoric null complement. Deviating from the general Bantu pattern, in which verbal extensions mark the reflexive, it uses an object noun class prefix instead. Example (11) is ambiguous:


The prefix /zi-/ can thus receive regular pronominal interpretation (identifying alternatively noun class 8 or 10) or it can refer back to the subject like a reflexive. (12) shows the phrase structure given my assumptions on the nature of noun class markers:


            Rather than stipulating a language-specific type of pro, it seems more reasonable to assume that the ambiguity between pronominal and reflexive interpretation can be attributed to a lexical idiosyncrasy of zi-. The fact that such anaphoric constructions are impossible with NCs markers of other classes seems to support this view.

            Myers 1987, 1998 argues that selectional restrictions hold between subject and tense markers. Current syntactic models contradict his assumption of an AUX node unifying both morphemes, but the Spec-head relationship commonly assumed for Subject and Tense (as shown in 13) would certainly allow to capture the relevant selectional restrictions without special assumptions.


            In line with Halle and Marantz's 1993 arguments, I assume that syntax manipulates abstract feature bundles, and that phonological manifestation of these feature bundles is a spell-out mechanism. The systematic relationship of the surface phonological forms is thus a natural consequence of the structural equality of NCs in their various environments. The morphological material available to fill [NOUN CLASS] is monomoraic and thus prosodically deficient: the minimal word requirement in Bantu languages is typically a bimoraic foot. Thus, NC markers must cliticize to adjacent lexical heads. Notice that cliticization rather than prefixation explains the promiscuity of noun class markers, with which they occur on lexical heads of all major syntactic categories (A, N, V). The fact that the syntactic literature often does not carefully distinguish between clitics and affixes seems to be one of the main reasons for the multitude of incompatible analyses for NC markers.

            This raises the question why VP-adjuncts such as manner adverbs may not intervene between noun class (and tense) markers and the verb. The literature does not address this problem, which may already signal the hidden difficulties. The clitic status of NC markers rules out one possible line of argument, namely that morphological properties of NC markers require immediate adjacency to their host (as would be true for e.g. affixes). A straightforward explanation would be to assume verb movement to Agr. If elements adjoin to VP, they would still not intervene between the NC or tense clitics and V. However, as shown below, non-agreeing objects must stay in situ within the VP, and adverbials may not intervene between V and object either. One may want to draw upon Pollock's 1989 analysis of French and English and argue an adverbial may block the required overt verb movement to Agr and cause the derivation to crash. This move is made impossible by the fact in (16b): manner adverbs can intervene between V and non-agreeing subject (and object). If movement constraints are at work, they are thus tied to the morphosyntactic structure of the verbal complex. Since the facts available to me are not terribly clear, I will not ponder this point further. <<   >>   TOC