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

Poster B19, Wednesday, November 8, 3:00 – 4:15 pm, Harborview and Loch Raven Ballrooms

A meta-analysis of seven FMRI-studies on artificial grammar learning

Julia Udden1;1Department of Psychology and Linguistics, Stockholm University, Stockholm, Sweden

The artificial grammar learning (AGL) paradigm enables systematic investigation of acquisition of linguistically relevant structures. It is paradigm of interest for natural language processing research interfacing with theoretical linguistics, as well as for comparative research on language acquisition and evolution. The FMRI literature on AGL spans at least fifteen original studies on distinct samples. We have used the GingerALE method in a meta-analysis of this literature. By forming the union across statistical landscapes from each study, centered at the reported peak coordinates (spread determined by the number of subjects), the activation likelihood estimation method determines the overall likelihood of activation of clusters in the union map by a permutation test. We only included peak activations from the contrast comparing activity for non-grammatical > grammatical sequences (NG > G) during the classification phase of grammaticality judgments, which resulted in seven included studies. When thresholding the meta-analysis of the NG > G contrast at an FDR-corrected level of 0.05 or alternatively correcting at the cluster level (using a cluster-forming threshold of p > 0.001) with permutation testing, six clusters were significant: the left and right frontal operculum, the left and right IFG, and the left and right middle frontal gyrus (MFG). When correcting for multiple comparisons with FDR, without any assumptions on the correlations in the data, only the bilateral frontal operculum (FOP - extending into the inferior frontal gyrus on the left), were significant. These results emphasize the relatively larger contributions from the right hemisphere seen in AGL, compared for example to complex syntax in natural language experiments. A majority of adult AGL studies use visual presentation, and the literature thus has to be carefully taken into consideration as being biased towards finding the activations related to visual processing. The most robustly activated region was the FOP, which has been implicated a causal node affecting cognitive control processes and which has also been described as a part of the salience network. It is possible that the non-grammatical sequences are more salient, perhaps even mildly threatening to participants (they are also dispreferred in preference tests). The results reveal a network markedly different from the regions most robustly involved in processing related aspects of natural language processing, for example natural language syntax (e.g. LIFG and left posterior superior/middle temporal gyrus). LIFG remains a region of overlap of artificial and natural syntactic processing. The robust activation of the nearby FOP region (bilaterally) is notable as a homologue pair of structures that are less often implicated in natural syntactic processing. However, the adjacent anterior insula, clearly left-lateralized, has repeatedly been implicated as one of the most reoccurring sites to display lesions in aphasia patients with both production and comprehension deficits. In summary, the results point to the crucial involvement of the bilateral frontal operculum and several (mostly perisylvian) regions in the right lateral hemisphere. I interpret the degree of activity as dependent on grammar and as interacting with input modality.

Topic Area: Grammar: Syntax

Back to Poster Schedule