The Branching Direction Theory of Word Order Correlations Revisited

In Dryer (1992a), I presented evidence from a sample of 625 languages on which pairs of elements correlate with the order of object and verb, and which do not (Greenberg 1963). I argued that the word order correlations reflect a tendency for languages to be consistently leftbranching or consistently right-branching, what I referred to as the Branching Direction Theory (the BDT), and proposed that this tendency reflects processing difficulties associated with mixing leftand right-branching. The predictions of the BDT depend heavily, however, on one’s assumptions about constituent structure. A number of the correlations require assuming fairly hierarchical constituent structures, and are not predicted by the BDT if one assumes flatter constituent structures. In this paper, I discuss a number of these correlations, arguing that some of them can be explained by a combination of processing considerations and other principles, while a few remain unexplained under assumptions of flat constituent structures.