Antisymmetry and Word Order in Serial Constructions

Strict head-final surface order derives from underlying left-headedness in Ijo̩ a Niger-Congo language of Nigeria. A word order anomaly in Ijo̩ serial verb constructions (SVCs) strongly suggests this, and left-to-right asymmetric c-command among internal arguments of SVCs confirms it. The anomaly is universal among surface right-headed languages with SVCs, indicating that deep left-headedness is universal, as antisymmetry theory predicts (Kayne 1994). Assuming complements are in Specs, and that a light verb v selects every VP (Chomsky 1999), I derive VOVO from OVOV by two instances of V-to-v movement. I argue for a nonuniform approach to SVCs, involving relations of both raising (Campbell 1989) and control (Collins 1997). Other aspects of SVC word order are predictable from a universal thematic hierarchy nontheme > theme, and short scrambling (Takano 1998).*

1. A word order anomaly

Muysken (1987) points out a surprising fact: the word order in serial verb constructions (SVCs) in head-final languages is not the mirror image of that found in their head-initial counterparts. Consider 1–3 (from Muysken’s 17–20).¹

(1)

(2)

(3)

It has been persuasively argued from head-initial data that, in the core cases of SVCs, one verb has as its sister the projection of another (Larson 1991a, Lefebvre 1991, Campbell 1989, Veenstra 1996, Collins 1997).³ I will call the higher verb V1 and the subordinate verb V2. V1 Case-marks the highest nonsubject argument, under this approach (ideas on its theta-marker vary); V2 Case- and theta-marks a lower argument if one is present. Assuming these points and binary branching, 4a schematically represents VP in the Sranan 1a, 2a, and 3a. Extending the analysis to Ijo̩, something like the representation in 4b follows (cf. Muysken 1987).

(4)

The relative ordering of verbs in the Ijo̩ examples 1b, 2b, and 3b is entirely inconsistent with 4b, as Muysken points out. If V1 of Ijo̩ corresponds hierarchically to V1 of Sranan, we expect the latter to be final where the former is initial, and similarly for O1s and O2s as shown in 5. But this is not the case.

(5)

Subjects are initial in Ijo̩ suggesting that specifier positions are generated to the left. If we assume that O1 is specifier to V1 we might alternatively expect a structure like 6a,b for Ijo̩ SVCs. This hypothesis derives the correct relative ordering between O1 and O2, but wrongly predicts the word order [O1 (O2) V2 V1].

(6)

Although serial verbs are found in a number of OV languages, the linear orders corresponding to 4b, 5b, and 6 are completely unattested. Instead pairs of verbs and of objects exhibit identical relative orderings in SVCs of both VO and OV languages. Head-initial Yorùbá, Akan, Ewe, and Haitian Creole all show the word order pattern exemplified in 1a, 2a, and 3a (see Collins 1997, Campbell 1989, Déchaine 1993, among others). SVCs in OV languages pattern with Ijo̩ in their word order (see Lee 1992 on Korean, Nagarajan 1991 on Tamil, Hale 1991 on Misumalpan, Wheatley 1985 on Yi, Laughren 1976 on Senufo, and Pawley 1994 on Kalam).

I will refer to this problem as Muysken’s anomaly:

(7) Muysken’s anomaly: Head-finality has no correlates in the relative orderings of verbs [V₁ ... V_n] and arguments [O₁ ... O_n] in SVCs.

As 7 indicates, more than two verbs and internal arguments are possible in SVCs; for expository convenience I restrict my attention to pairs.

In this article I argue that Muysken’s anomaly is powerful evidence for Kayne’s 1994 proposal that directionality is not parametrized. Specifiers always precede heads, and heads must precede structural complement positions, in Kayne’s view. Ijo̩ and Sranan thus have the same underlying representation; local raising of verbs and internal arguments differentiate OV and VO orders. Tense morphology appears on V2 because the complex VP raises to [spec, TP] making V2 left-adjacent to T⁰, which cliticizes.

(8)

The argument is based on a comparison of SVCs in Ijo̩, on the one hand, and Yorùb...