Abstract
This article relates results in data-driven dependency parsing of Swedish to linguistic generalizations regarding syntactic argument status, such as tendencies regarding animacy and definiteness, as well as properties more specific to the Scandinavian languages, such as finiteness. We show how data-driven modeling in combination with labeled dependency representations enable the acquisition of functional preferences that are evident as statistical tendencies in language data. We present an in-depth error analysis of a data-driven dependency parser with a particular focus on assignment of core syntactic arguments and show how a data-driven parser provides an experimental setting where the influence of various linguistic properties may be evaluated and investigated further.