Recent developments in syntactic theory have seen a rapidly diversifying range of approaches to the treatment of verb-initial syntax. While VSO languages were viewed for a considerable period of time as a coherent typological class (derived via verb-raising or some equivalent), this appears to no longer be the case. Furthermore, VOS languages have generally been considered a rare residue fitting rather uncomfortably into the typologist's view of the world. In this talk I examine the analysis within HPSG of grammatical relations and constituency in Tongan, a well-known verb-initial language of the Polynesian family, while also drawing on comparative data from related languages.

Starting from some proposals made by Bob Borsley for dealing with verb-initial structures in Welsh and Syrian Arabic, it is shown that the categories needed to account for some basic properties of Tongan morphosyntax involve something of an amalgam of the two types used by Borsley. Furthermore, as has now become familiar from work on numerous Austronesian languages, it is convenient to appeal to multiple conceptions of subjecthood to account for key structural properties of these languages. Grammatical variation associated with these notions of subjecthood within Polynesian suggests that even in such a homogeneous family, the 'verb-initial type' is only characterizable in surprisingly superficial structural terms and the (sometimes difficult) choice of what counts as subject determines whether a language is considered VSO or VOS.