Poster C66, Friday, August 17, 10:30 am – 12:15 pm, Room 2000AB
Break the ice vs. boa constrictors: Do they have different neural bases?
Shohini Bhattasali1, Murielle Fabre1, John Hale1;1Cornell University
Introduction: Language comprehension is widely viewed as being subserved by a left-lateralized perisylvian network of frontal and temporal brain regions. Several compositional and memory-related processes occur incrementally to accomplish sentence comprehension. We address the neuroanatomical basis of these processes by focusing on the frequently co-occurring word sequences, known as Multiword Expressions (MWEs) in computational linguistics. MWEs are considered a single unit, rather than a structurally composed combination (Sag et al., 2002). Thus, MWEs are a perfect testing ground to understand how expressions like break the ice, boa constrictor, safe and sound, see to it, in spite of are processed in the brain. This study investigates whether different types of MWEs evoke different patterns of activation in the brain using fMRI. Specifically, we ask if the strong relationship between verbs and their arguments are encoded in different brain areas compared to non-verbal MWEs featuring no argumental structure. Method: Participants (n=51, 32 female) listened to The Little Prince’s audiobook for 1hour 38min. Participants' comprehension was confirmed through multiple-choice questions (90% accuracy, SD = 3.7%). Functional scans were acquired using multi-echo planar imaging sequence (ME-EPI) (TR=2000ms; TE's=12.8, 27.5, 43ms; FA=77 degrees; FOV=240.0mm X 240.0mm; 2X image acceleration; 33 axial slices, voxel-size 3.75 x 3.75 x 3.8mm). Preprocessing was done with AFNI16 and ME-ICAv3.2. MWEs were identified using a statistical tagger trained on Children’s Book Test dataset. Presence/absence of verbal expression yielded two categories of MWEs (i.e. 56% verbal vs. 44% non-verbal). Additionally, we entered four regressors of non-interest into the GLM analysis (SPM12): word-offset, word frequency, pitch, intensity. To account for sentence-level compositional processes, we included a regressor formalizing syntactic structure building based on a bottom-up parsing algorithm. These regressors were not orthogonalized. Contrasts were inclusively masked with the main effect of all MWEs and FWE-corrected (T-score>5.32). Results: The main effect for presence of MWEs elicited activation mainly in bilateral Supramarginal Gyrus, right Angular Gyrus, right MFG, and right Precuneus Cortex (Fig1A). Whole-brain contrasts show that these two types of MWEs activate different brain regions with no overlap. Verbal MWEs appear right-lateralized compared to non-verbal ones in IPL and in IFG triangularis (Fig.1B). The opposite contrast yielded a mostly right-lateralized and wider pattern of activation, including bilateral Supramarginal Gyrus extending to STG and right SMA together with smaller activation clusters in Pars Opercularis and MTG (Fig.1C). Conclusion: The results indicate posterior Supramarginal and Parietal areas and SMA as involved in lexical-semantic memory (Binder et al., 2009) and show that these non-compositional MWEs mostly implicate a right-lateralized network. Our findings confirm that bilateral Supramarginal Gyrus is sensitive to co-occurence frequency of word combinations as reported previously for semantically meaningful and frequent word-pairs (Graves et al., 2010; Price et al. 2015). Additionally, the significant clusters for verbal and non-verbal MWEs illustrate spatially distinct patterns of activation and a dorso-ventral gradient is observed in Broca’s area for verbal versus non-verbal MWEs. Finally, activation patterns for verbal-MWEs indicate that verb-argument selectional relations in frequent verbal collocations exclusively involve right hemisphere activity in Broca’s area and IPL.
Topic Area: Computational Approaches