Poster D28, Friday, August 17, 4:45 – 6:30 pm, Room 2000AB
Expectation modulations based on verb bias and grammatical structure probability shape sentence processing: An ERP study
Kirsten Weber1,2, Lena Henke1, Cristiano Micheli2;1Max Planck Institute for Psycholinguistics, Nijmegen, the Netherlands, 2Donders Institute for Brain,Cognition and Behaviour, Nijmegen, the Netherlands
Language processing is dynamic and is shaped and modulated by all kinds of factors. Here we investigated how prior knowledge and the current linguistic environment influence the processing of sentence structures. In this project we looked at how two types of information, 1) the statistics of the input (the proportion of two ditransitive sentence structures across different blocks) and 2) verbs biasing towards one of these structures or the other. We wanted to see how and when these different type of information influence sentence processing and whether they have similar or distinct electrophysiological markers. We acquired electroencephalography recordings while 32 Dutch native speakers were reading sentences word by word. The Dutch sentences either had a prepositional object (PO) or double-object (DO) structure. They contained verbs with either a DO or PO bias. The verb bias values were taken from a corpus analysis (Colleman, 2009) and were confirmed by a post test that participants filled in one week after the experiment. In this test participants had to generate ditransitive sentences for the different verbs resulting in their production preferences, and thus biases, per verb. In one block 75% of the sentences had a DO sentence structure and 25% a PO sentence structure, with the opposite pattern in the other block. We thus had three factors in this study, Global Structure Frequency, Verb Bias and Sentence Structure. We looked at ERP effects on the post-verbal noun, where our two expectation manipulations should influence the processing of sentence structure. Reported results are either from a linear mixed effect model of single-trial N400 data of central-posterior electrodes or cluster-based permutation tests over all electrodes using the fieldtrip toolbox (http://www.fieldtriptoolbox.org). Expectation violations based on verb bias (unexpected verb-structure combination) resulted in a larger N400 on the post-verbal noun. This was especially the case when the verb biased towards a DO structure but a PO structure was presented. Also a later frontal positive component showed a larger effect if the verb-structure combination was more surprising. Global Structure Frequency, and thus the current linguistic environment, had a marginally significant effect on a later positive component. In sum, different types of information that could be used for predictive language processing of grammatical structures can have different ERP effects. Local information from the verb modulates the earlier N400 component as well as a later positivity. The statistics of the current linguistic environment on the other hand has a weaker effect later on during processing. These distinct effects shaping sentence processing are in line with findings from a previous fMRI study with a similar design. Verb-based surprisal effects were found within the canonical language network, while the anterior cingulate kept track of the current linguistic environment.
Topic Area: Grammar: Syntax