Search Abstracts | Symposia | Slide Sessions | Poster Sessions
The neural geometry of metaphor multimodality over time investigated via Representational Similarity Analysis
Poster Session A, Wednesday, September 30, 11:00 am - 1:00 pm, Wangari Maathai
Chiara Battaglini1, Veronica Mangiaterra1, Paolo Canal1, Valentina Bambini1; 1University School for Advanced Studies IUSS Pavia
ERP studies consistently show enhanced N400 responses to metaphors, sometimes followed by a later positivity, reflecting semantic and inferential processing demands [1]. This electrophysiological pattern varies as a function of interacting factors, including the degree of physicality of the metaphor, with more physically grounded expressions typically eliciting larger N400 amplitudes [2]. Converging evidence indicates that metaphor processing recruits a distributed, multimodal neural network extending beyond language regions to include areas involved in theory of mind, affective evaluation, mental imagery, and sensorimotor simulation [2-4]. However, relatively little attention has been devoted to characterizing how these representational dimensions unfold over time and jointly shape neural responses during online processing. To address this gap, the present study adopted a time-sensitive multivariate approach based on time-resolved representational similarity analysis (RSA) applied to EEG data, allowing us to track the temporal evolution of representational structures during metaphor comprehension. Thirty-five Italian speakers read 124 metaphoric sentences while EEG activity was recorded, following the paradigm in [2]. Time-resolved representational dissimilarity matrices (RDMs) were computed from group-averaged centroparietal EEG activity. Neural RDMs were compared against eight theoretically motivated model RDMs encoding pairwise distances between stimuli according to different representational dimensions: (i) non-propositional metaphor-level indices, such as physicality and mentality; (ii) non-propositional interpretation-level features, including embodiment and mentalizing; and (iii) distributional semantic measures, including familiarity, interpretation relevance, topic–vehicle distance, and metaphor–interpretation distance computed through text-embedding techniques. Cluster-based permutation analyses were conducted across the 200–800 ms interval to identify time windows in which each model significantly aligned with the neural representational structure. Significant correlations emerged between EEG-derived RDMs and most theoretical models, although their temporal profiles differed substantially, highlighting the dynamic nature of metaphor processing. No reliable alignment was observed for metaphor physicality or interpretation mentalization. In contrast, all other models showed significant correspondence with neural activity: between 200 and 400 ms, neural representational geometry was most strongly aligned with distributional and semantic properties, including topic–vehicle distance, metaphor familiarity, and metaphor mentality. Between 400 and 700 ms, interpretation embodiment emerged as the best predictor of the neural data. Finally, between 700 and 800 ms, metaphor–interpretation distance provided the strongest fit to the neural representational structure, suggesting a later-stage refinement process in which the relationship between the literal expression and its intended interpretation becomes fully resolved. The results reveal a dynamic progression in metaphor comprehension, moving from early semantic access and inferential elaboration to later stages of embodied meaning construction and global pragmatic integration. Distributional semantic properties and conceptual dimensions contribute at earlier stages, whereas embodiment becomes more prominent during intermediate latencies, consistent with semantic–pragmatic exploration and sensorimotor grounding of metaphor meaning. Later neural activity is instead best explained by broader distributional relations between metaphor and interpretation, reflecting pragmatic reanalysis. More generally, these findings support the view that language comprehension relies on the dynamic interaction between distributional, embodied, and inferential representational systems unfolding over time. [1] Canal & Bambini, 2023; [2] Canal et al., 2022; [3] Bambini et al., 2011; [4] Citron et al., 2014
Topic Areas: Meaning: Discourse and Pragmatics, Multisensory or Sensorimotor Integration