This paper sets out to substantiate three claims. First, there is (conceptual and emiprical) evidence for adopting a local derivational approach to syntax (as opposed to non-local derivational, local declarative, and non-local declarative approaches). Second, the general counter-bleeding effect with remnant movement provides an argument for a derivational approach as such, but raises problems for a local derivational approach as it is presupposed in phase-based minimalist syntax. Third, this challenge can be met if the movement-related features of moved items (like [wh]) have values that act as buffers, storing minimal aspects of the derivational history on the moved item.
Background
Two DichotomiesCurrent theories of syntax can roughly be grouped into four classes according to two dichotomies: An approach can be derivational (such that syntactic structures are generated by structure-building and structure-manipulating operations) or declarative (such that syntactic structures are viewed as potential objects that have to comply with the grammatical constraints of a language); and an approach can be local (such that all syntactic dependencies are confined to local domains, e.g., subtrees, and syntactic constraints have access only to these subtrees) or non-local (such that syntactic dependencies can span larger portions of structure, potentially the whole sentence, and syntactic constraints also have access to these large domains). A cross-classification yields four models of grammar: Current local derivational approaches include most prominently the Minimalist Program (Chomsky (2001;), where syntactic structures are generated bottom-up, by alternating operations like Merge, Move, and Agree, and the accessible window of a derivation is quite small throughout (it is standardly assumed to be confined to the minimal phase). GPSG (Gazdar (1981), Gazdar et al. (1985)) and HPSG (Pollard & Sag (1994), Sag & Wasow (1999), Müller, St. (2007) belong to the class of local declarative approaches: Dependencies and constraints are evaluated in small subtrees (essentially, mother-daughter relations), and syntactic structures are assumed to be licensed by constraint satisfaction, rather than being generated successively by syntactic operations. Next, there are non-local derivational models,