An fMRI study of canonical and noncanonical word order in German

Abstract

Understanding a complex sentence requires the processing of information at different (e.g., phonological, semantic, and syntactic) levels, the intermediate storage of this information and the unification of this information to compute the meaning of the sentence information. The present investigation homed in on two aspects of sentence processing: working memory and reanalysis. Event-related functional MRI was used in 12 healthy native speakers of German, while they read sentences. Half of the sentences had unambiguous initial noun-phrases (masculine nominative, masculine accusative) and thus signaled subject-first (canonical) or object-first (noncanonical) sentences. Noncanonical unambiguous sentences were supposed to entail greater demand on working memory, because of their more complex syntactic structure. The other half of the sentences had case-ambiguous initial noun-phrases (feminine gender). Only the second unambiguous noun-phrase (eighth position in the sentences) revealed, whether a canonical or noncanonical word order was present. Based on previous data it was hypothesized that ambiguous noncanonical sentences required a recomputation of the sentence, as subjects would initially commit to a subject first reading. In the respective contrasts two main areas of brain activation were observed. Unambiguous noncanonical sentences elicited more activation in left inferior frontal cortex relative unambiguous canonical sentences. This was interpreted in conjunction with the greater demands on working memory in the former condition. For noncanonical ambiguous relative to canonical ambiguous sentences, an activation of the left supramarginal gyrus was revealed, which was interpreted as a reflection of the reanalysis-requirements induced by this condition.

Figure 1

Simplified syntactic analysis of sentences with unambiguous masculine noun‐phrases in the initial position (linguistic elements have been represented by filled circles, for a detailed description of the syntactic structure, see [Gorrell, 2000]). The canonical sentence on the left has a simpler structure than the noncanonical, object‐first sentence on the right. At positions marked t_x previous elements of the sentence (coindexed) have to be reactivated from working memory.

Figure 2

A: Comparison of all sentence conditions against fixation/baseline. Activations are rendered on a T1‐weighted single subject MNI template as provided by SPM. For all differences shown, P < 0.01, cluster extent 20 voxels. B: Main effect of Word order: noncanonical against canonical sentences. An activation of the left anterior inferior frontal region (BA44/45) and the left supramarginal gyrus is seen. C: Effects of word order in the unambiguous sentences: noncanonical sentences are more active in the left anterior frontal region. D: Effects of word order in the ambiguous sentences: noncanonical sentences yield an activation of the left supramarginal gyrus. The axial views presented in this figure were superimposed on the mean anatomical image formed by averaging the T1 structural MRI scans mapped into normalized MNI space across all subjects.

Figure 3

Time course of the BOLD response in several regions of interest. The time‐points on the x‐axis are relative to sentence onset (0 s.) The ROI in the inferior frontal gyrus (upper panel, coordinates −44, 12, 20) showed a pronounced and early difference in the BOLD time course between noncanonical and canonical sentences for the unambiguous sentences. The corresponding difference for the ambiguous sentences was smaller and delayed. The second ROI encompassed the posterior part of the middle temporal gyrus (−56, −48, 0, middle panel) and showed a difference between noncanonical and canonical sentences in the ambiguous conditions only. The third ROI was located in the left inferior temporal gyrus (−56, −32, −16, lower panel): The difference between noncanonical and canonical sentences was more pronounced in the ambiguous sentences (see also Table III).