The Role of Syntactic Priming in Dialogue

2008-03 till 2012-02
Research Areas: 

SynPrime tests key assumptions of the Interactive Alignment Model, a recent theory of dialog some psycholinguists subscribe to. For this purpose, we employ and develop novel empirical approaches primarily based on dialog corpora which we analyse using automatic methods from computational linguistics and statistics.


Methods and Research Questions: 

The Interactive Alignment Model of dialog (IAM; Pickering and Garrod, 2004) claims that situation-model alignment is caused by automatic and resource-free alignment at all levels of linguistic processing. The IAM further presupposes that at each level, alignment is caused by priming of representations at that level.

Lab studies seem to support the IAM: They find that at least in task-oriented dialog, people tend to repeat structural choices they or their interlocutors made. This is interpreted as syntactic priming. Recent corpus studies of naturally occurring dialog, however, show inconsistent results. This project aims to improve methods for analysing syntactically annotated dialog corpora (treebanks), and to empirically answer the question whether syntactic priming and alignment really happen in natural, spontaneous dialog.

We employed a recent method for measuring syntactic priming in dialogue corpora developed by David Reitter in his PhD thesis (Reitter 2008), and improved this approach by adding more robust methods for analyzing the results.



Our preliminary results suggest that there is no other-priming and no syntactic alignment in natural dialog. If corroborated, this would invalidate a key assumption of the Interactive Alignment Model of dialog.