Poster C31, Wednesday, August 21, 2019, 10:45 am – 12:30 pm, Restaurant Hall
Representational analysis of abstract word meaning
Karen Meersmans1, Rose Bruffaerts1, Simon De Deyne2, Gert Storms3, Patrick Dupont1, Rik Vandenberghe1;1Laboratory for Cognitive Neurology, Department of Neurosciences, KU Leuven, 2Computational Cognitive Science Lab, University of Melbourne, 3Laboratory of Experimental Psychology, KU Leuven
Concrete and abstract word are processed differentially in the semantic brain network, with concrete words activating bilateral posterior regions, reflecting their association with the visuoperceptual system, and abstract words activating left inferior frontal and temporal regions, reflecting their relation with the symbolic system. To further clarify the representation of meaning of abstract and concrete words, we conducted an event-related fMRI-experiment and investigated how semantic similarity between concrete and abstract words is coded in the brain. Twenty-four subjects performed a word repetition task on visual and auditory stimuli (32 abstract and 32 concrete nouns, matched for valence, arousal, frequency, age-of-acquisition, prevalence). Consonant strings and contorted speech (rotated spectrogram) were included as control stimuli. The experiment comprised 8 runs of 80 pseudorandomised trials each (64 stimulus and 16 control trials) with an intertrialinterval of 8.25s. Every stimulus was presented 4 times in each modality. Structural and functional data were acquired using a 3T Philips Achieva, equipped with OptoActive-II ActiveNoise Cancellation sytem (OptoAcoustics Ltd.). Functional images were subsequently realigned, slice-timing corrected, normalised to MNI-space and smoothed. A univariate analysis using SPM12 showed activation for abstract relative to concrete nouns in left lateral anterior temporal pole extending into anterior superior temporal sulcus (aTP/aSTS; -48,8,-22 kE=171), in left rostroventral inferior parietal cortex, extending into posterior middle temporal gyrus (rvIPL/pMTG; -54,-43,26; kE=255), bilaterally in medial superior frontal gyrus (mSFG; RH 15,47,5, kE=61; LH 63,-13,-4, kE=92) and in right cerebellum (22,-52,-43, kE=176; all voxel-level uncorrected p<0.001, cluster-level FWE-corrected p<0.05). Before Multivariate Pattern Analysis, pMTG and rvIPL were separated into two unconnected regions by excluding white matter voxels (inclusive masking with average across subjects grey matter map; threshold>0.5). The activated regions (minus the cerebellum) were then superimposed with the subject-specific grey matter probability maps (threshold>0.5). fMRI-similarity matrices were constructed by calculating the cosine similarity between the trial-specific response patterns defined as the integral of the BOLD-response 2-8 seconds post-stimulus onset. These matrices were averaged across trials, modalities and subjects in order to obtain a stimulus-specific fMRI-similarity matrix, which was correlated with a semantic similarity matrix obtained from word association data. For abstract words, this yielded a significant semantic similarity effect in rvIPL (Spearman's rho=0.14, p=0.0006), pMTG (rho=0.11, p=0.006) and mSFG (rho=0.10, p=0.01) but not aTP/aSTS (rho=0.07, p=0.06). No semantic similarity effect for concrete words was found in these regions (p>0.1). Our observation of increased BOLD-amplitude in lateral aTP corresponds to the previously described conreteness-related gradedness of this region from dorsolateral (social/abstract concepts) to ventromedial (concrete object information). However, our similarity analysis, verging on significance, does not unequivocally confirm the representational nature of the anterior temporal effect. Abstract words are characterised by means of social, event, and introspective features and rely on thematic relations. Both mSFG and rvIPL are part of the Default Mode Network, involved in automatic semantic processing. pMTG/rvIPL have previously been implicated in the processing of thematic associations, event and relational semantics. We hypothesise that pMTG and DMN regions support the retrieval of abstract semantics by providing thematic, event and relational information.
Themes: Meaning: Lexical Semantics, Methods
Method: Functional Imaging