Grammar- Versus Frequency-Driven Syntactic Ambiguity Resolution : The Case of Double-Object Constructions

Abstract

With regard to argument serialization, a number of prominence hierarchies have been proposed. We discuss possible ways how such hierarchies might be involved in the reverse process, the process of recovering syntactic functions from a given input string during human language comprehension. We present three experiments and an accompanying corpus study focusing on German sentences with two locally ambiguous objects, both filled by animate NPs. Temporarily, the two objects can be parsed as ‘accusative object preceding dative object’ and as ‘dative object preceding accusative object’. Our results show a strong preference for the order ‘ACC before DAT’. We explore the consequences of this finding for both grammar and frequency based approaches to the human parsing mechanism. We show that accounts in which units of size greater than a single DP can compete with each other are difficult to maintain given our experimental results. We argue instead that our results are best explained in terms of a ranking between cases/syntactic functions, reflecting both the case hierarchy and the frequency ranking that we found in our corpus data.