You are viewing the SNL 2017 Archive Website. For the latest information, see the Current Website.

Poster D12, Thursday, November 9, 6:15 – 7:30 pm, Harborview and Loch Raven Ballrooms

Decoding the P600: late ERP positivities to syntactic mismatch share neural patterns with nonlinguistic oddballs, but not face or semantic manipulation patterns

Jona Sassenhagen1, Christian J. Fiebach1,2;1Goethe University Frankfurt, 2IDeA Center for Indidivudal Development and Adaptive Education, Frankfurt

The late positive P600 component of the Event-Related Potential/ERP is often interpreted as a correlate of high-level linguistic processing (e.g., syntactic or semantic integration). We aim to test the alternative hypothesis that the P600 is an instance of the low-level, domain-general P3 component. Multivariate Pattern Analysis/MVPA of EEG data (King & Dehaene, 2014) can be used to probe questions about cognitive and ERP phenomema that cannot be addressed via standard univariate ERP analyses, complimenting previous findings. For example, pattern classifiers can be trained on one experimental contrast and then tested on their performance on a different manipulation. If the pattern classifier performs well on this kind of cross-study task, this indicates the ERPs, neural process, and fundamentally cognitive processes characteristic of the first experimental contrast also govern the second. In a pre-registered analysis (, we collect EEG data for the same subjects (n=25) for three experiments: a sentence processing experiment with semantic and syntactic (agreement) violations, an auditory Oddball (rare tone) experiment, and a face processing task. Thus, we combine the study protocols typically used to elicit P3 (Oddball), N170 (face), N400 (semantics) and P600 (syntax) components. With MVPA, we investigate to what extent the neural patterns characteristic of syntactic mismatches in our (standard) P600 paradigm overlap with those of processing faces, semantic mismatches or Oddball tones. MVPA decoders from the Oddball experiment (regularized logistic regression, cross-validated, fit for each time sample) successfully classify trials from the syntactic manipulation (ROC AUC ~.65, p < .01); they perform about as well as when classifying based on P600 patterns themselves (ROC AUC ~.65), and much better than semantic and face processing-sensitive components (both p < .01), whose performance is around chance (ROC AUC ~.5). This pattern is expected if the EEG responses to non-linguistic Oddballs (P3) and to syntactic mismatches (P600) exclusively share a neural substrate (Coulson et al., 1998; Sassenhagen et al., 2014; Sassenhagen & Bornkessel-Schlesewsky, 2015). Consequently, functional interpretations of the P600 should be aligned with models of the P3. We also caution against "reverse inference": observing neural patterns equivalent to those of the P600 does not necessarily entail any specific high-level linguistic process.

Topic Area: Grammar: Syntax

Back to Poster Schedule