Measuring distributional semantic effects in syntactic variation

Introduction and previous work A number of papers in the quantitative syntax literature have proposed that semantic dependencies between words affect their relative order in a sentence [13, 15]. This idea is an extension of the very general principle stating that (syntactically) related words tend to occur next to each other [4]. This principle is known under various names in syntax and in the language processing literature, such as early immediate constituents, locality, and dependency length minimisation [12, 8, 18, 7]. Semantic and syntactic relations between words are highly correlated notions and it is difficult to tease apart the syntactic from the purely semantic effects on word adjacency. Lohse et al. (2004) [15] unambiguously identifies a semantic pressure favouring adjacent orders in verb-particle construction in English. A verb-particle transitive phrase can have two alternative word orders: verb-particle split order (1) and verb-particle adjacent order (2).