What is Basic Word Order?
In order to reconstruct something, we must first know what it is so that we can identify it in each daughter language and thereby compare it across the family. As the definition of “basic word order” has been the subject of some dispute, I first briefly review the uses to which the term has been put.
Greenberg (1966:74) originally defined the notion of “basic Word Order” by identifying the terms “subject” and “object” semantically, then concluding that the statistically dominant order would be considered “basic”. Hawkins (1983:12) employed the same semantic criteria to identify units; then he used three tests to identify which order amongst these units might best be called “basic” (p.13): (i) most frequent in attested sample (i.e. of natural discourse) = basic; (ii) most frequent in grammatical contexts (e.g. occurs with more lexemes) = basic; (iii) unmarked (either grammatically or pragmatically) = basic. For a large majority of the languages in his database, these somewhat intuitive criteria revealed (at least to Hawkins) a single basic order.2 Dryer (1983) advocates reducing the criteria to nothing more than attested frequency in discourse, this being presumably the test with which he has created the largest extant word order typology database (cf. the findings reported in Dryer 1988, 1992).
An increasing number of researchers are finding that the notion of a basic grammatical order of A, S and O vis-à-vis the verb may not be the most insightful. On the one hand, in many languages, the highest discourse frequency belongs to a verb-only clause, in which any order of subject and object vis-à-vis the verb appears less basic, since subjects and objects rarely occur as explicit noun phrases (Payne 1986:453, 459); on the other hand, when subjects and objects do occur explicitly, their order vis-à-vis the verb often cannot be predicted from their grammatical relation, but are instead better predicted from pragmatic information associated with each participant (Mithun 1987; Payne 1987; Payne (ed) 1992, inter-alia, Downing and Noonan (eds) 1995, inter-alia).
From the other direction, definitions of basic order have also come from more formal traditions, in which basic order is equated with “deep” structure, usually that from which all the surface variants can be most readily derived by movement rules (McCawley 1970; Derbyshire 1977, 1985) or theory-internal principles (Hale 1992). In this tradition, the notions of subject and object are considered derivative from constituency in the VP, object internal to the VP and subject external. Thus any assymetry between subjects and object are taken as evidence for constituency, regardless of the evidence of more traditional tests for constituency (Hale 1982, 1990, 1992; Jelinek 1984). While frequency might be expected to follow more-or-less directly from constituency facts (e.g. an [OV] verb phrase constituent might be expected to produce a high frequency of OV order), Hale (1992) argues convincingly from Papago data that discourse frequency may be completely unrelated to this notion of basic order; Payne (1994) argues the same point from Panare data.
Little attention has been paid to the possibility — so prevalent in the Cariban languages — that different constructions in a single language may present “split configurationality”, in which different constructions in a single language partake of different VP structures (cf. Comrie 1989:89 for German; Payne 1994, 1999:153–57 for Panare; Gildea 1999 for the entire Cariban family; Bresnan 1995 [cited in Payne 1999] for a theoretical position on split configurationality). As might be expected, the determination of a single “basic word order” can be controversial in such cases, where syntactic constituency is inconsistent and frequency facts might also vary across the different construction types.
In this paper, I examine order and constituency across the modern languages of the Cariban family in search of reconstructible patterns. If basic word order is defined in terms of frequency, I have no principled means of using comparative evidence to reconstruct a prior most frequent order — we have no prehistorical texts on which to use our quantitative methodologies, nor can we reconstruct any. If basic order is defined in terms of a single deep structure from which all surface variants are to be derived, it is somewhat difficult (and excessively theory-dependent) to know first, which deep structure to choose for each synchronic language, and second, what sort of deep structure to reconstruct, given that we do not know how deep structures evolve, and we do not yet have a detailed knowledge of the surface structures which would presumably be derived from our reconstructed deep structure. Thus, simple comparison across related languages is insufficient to allow us to reconstruct either of the abstractions that go under the name “basic word order”.
However, a few voices in the literature point out the possibility of recon structing modern order by reconstructing certain nonverbal constituents which later evolve into main clause verb phrases. Bybee (1988:354) points out that in principle, “One grammatical order is not established on analogy with or to harmonize with another order, rather a new grammatical construction develops in a language out of the constructions that already exist and shows ordering consistent with the construction from which it developed.”3 Following this same line of reasoning, Aristar (1991) shows that a number of Greenbergian word order correlations could be traced not to synchronic “harmonic principles” (cf. Hawkins 1983), but to common origin, as a single agreeing modifier construction in the Agaw (Cushitic) languages gave rise to relative clauses, genitival phrases and adjectival phrases, all of which then “harmonize”. In the example most germane to the present paper, Claudi (1994) shows that a change in “basic order” from SVO to SOV in Mande and from VSO to SVO in Egyptian Coptic both result from the reanalysis of gerunds (verbal nouns) which were possessed by their notional O arguments: in both cases, N > V and Genitive > O. The genitive-N order in Mande thus yielded an innovative OV order, whereas the N Genitive order in Coptic yielded the innovative VO order. As she summarizes (Claudi 1994:218–19), “the fact of word order change comes from the fact that the Genitive N order encodes the O and V meanings, respectively, but that the order of O and V in main clauses is DIFFERENT from the order of genitive and noun.” [emphasis in original]
After a review of previous word order studies in the Cariban language family, I show the parallel Genitive N source for all of the innovative VPs in modern Cariban languages; I also show how one source construction that lacked a genitive phrase evolved into a nonconfigurational VP.
