Presentation

Miriam Greidanus Romaneli¹, Saskia B. J. Koch¹, Margot Mangnus¹, Kexin Cai¹, Christine Cimpian¹, Sushmita Sadhukha², Ivan Toni¹, Jana Bašnáková¹, Arjen Stolk^1,2; ¹Radboud University, Donders Institute for Brain, Cognition, and Behaviour, ²Dartmouth College

Human communicative signals elicit stable semantic concepts due to repeated use in a community (Clark, 1996; Levinson, 2000). Yet, in dialogue, people continuously and dynamically adjust the relationship between signals and concepts, including their granularity (i.e. the amount of semantic detail referred to by an individual signal). Here, we build on the suggestion that communicative interaction requires coordinating a shared conceptual space (Stolk et al., 2022) and explore how interlocutors adapt the granularity of signal-meaning mappings in this space, a crucial but unexplored feature of this coordination process. This project expands on behavioral and neural (fMRI) observations obtained in a controlled yet open-ended non-verbal communication game (de Ruiter et al., 2010; Stolk et al., 2013, 2014). Analogous to everyday conversations, the game challenges dyads to construct signals that are referentially contingent on the ongoing communicative interaction, including current signal-meaning relationships. Differently from everyday conversations, communicators have no prior experience with these signals, providing researchers reliable access to signals’ referential contingencies. We reasoned that, under these circumstances, the granularity of communicators’ concepts can be inferred from the number of meanings mapped onto individual signals. We quantified signals produced by pairs of communicators during real-time non-verbal interactions on a digital game board. In each trial, two players were tasked with jointly reproducing a target configuration of two given shapes, shown to one player only. This required that player to construct a signal using her shape that conveys to her partner the target configuration of his shape. These circumstances drive pairs to converge on a limited but idiosyncratic number of signals among an open-ended set of possibilities, such that different dyads use different signals to refer to the same meaning (i.e. shape configuration; Stolk et al., 2014). Preliminary behavioral results (N=8 dyads, 94 trials/dyad) indicate that, when having to convey a new meaning, some dyads introduce and converge in using unique signals for that meaning while retaining existing signal-meaning mappings. In contrast to these “splitters”, so-called “lumpers” introduce and converge in using new signals for both new and existing signal-meaning mappings. These observations suggest that lumper and splitter dyads spontaneously coordinate their conceptual space at different levels of granularity. We plan to test whether this behavioral characterization of conceptual granularity is reflected in the representational dimensionality (RD) of neural activity in the right superior temporal gyrus (rSTG), a cortical area previously found to support conceptual alignment between communicators (Stolk et al., 2014). RD of a neural population has been shown to balance the computational trade-off between generalizability and separability of mental representations (Fusi et al., 2016; Badre et al., 2021). We hypothesize that coarser representations in the meaning space of lumper dyads are supported by lower RD, whereas the finer-grained representations of splitter dyads are paralleled by higher RD in the rSTG. These inquiries will advance our neurocognitive understanding of how communicators negotiate and move between levels of conceptual granularity in their signal-meaning mappings, a fundamental yet largely unexplored aspect of human communicative interaction and language use.

Topic Areas: Meaning: Discourse and Pragmatics, Meaning: Lexical Semantics