Model representational geometry and brain potentials jointly reveal relativity in word meanings
Abstract
Whether word meanings partition reality uniformly across languages or reflect language-specific categorization remains debated. Universalist accounts assume shared conceptual repertoires whereas relativist accounts emphasize language-specific partitions of conceptual space. However, evidence has relied largely on a few semantic domains and behavioral measures, leaving unclear whether and how graded semantic overlap of translation-equivalents affect the neural dynamics of concept processing. Here we combined computational semantic alignment (SA), embedding-based representational similarity analysis, and electroencephalography to test whether graded SA predicts cross-language representational correspondence and neural correlates of implicit semantic processing. Thirty-five Chinese–English bilinguals completed a translation-recognition task involving 160 translation-pairs across 20 semantic domains and 160 unrelated controls. High-SA translations showed stronger cross-language representational correspondence than low-SA translations, whereas controls showed near-zero correspondence. High-SA translations elicited less negative N400 amplitudes and larger N400 translation-priming effects. Moreover, millisecond-by-millisecond comparisons across the 0–1,000 ms epoch detected high–low SA differences within the canonical N400 window for translation-pairs, but no corresponding differences for controls at any time point across the entire epoch, indicating that the N400 finding was detectable without restricting comparisons to a preselected time window and unique to concepts represented by translation-pairs rather than non-translation controls. The SA effects from translation-equivalents were further modulated by second-language proficiency. Overall, by linking cross-linguistic distributional semantic structure to model representational geometry and electrophysiological responses, this work convergingly supports a nuanced relativist account of word meanings, in which bilinguals are neurophysiologically sensitive to SA, with greater sensitivity at higher levels of second-language proficiency.
Related
- The representational geometry of emotional states in basolateral amygdala
- Representational geometry of punishment learning across basolateral amygdala neurons
- Case Report: Representational geometry of multi-digit coordination in the macaque grasping network for intracortical BCIs
- Representational geometry as a fidelity metric for connectome-constrained networks: evidence from the Drosophila visual system
- NeuralFieldManifold: Uncovering the Manifold Geometry of Local Field Potentials
BibTeX
@misc{liu2026model,
title = {{Model representational geometry and brain potentials jointly reveal relativity in word meanings}},
author = {Yufeng Liu and Shifa Chen and Yuqing Zhang and Xingyu Yang and Brent Wolter},
howpublished = {PsyArXiv},
year = {2026},
month = sep,
doi = {10.31234/osf.io/s4qfr_v1},
url = {https://osf.io/preprints/psyarxiv/s4qfr_v1/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Model representational geometry and brain potentials jointly reveal relativity in word meanings
AU - Yufeng Liu
AU - Shifa Chen
AU - Yuqing Zhang
AU - Xingyu Yang
AU - Brent Wolter
PY - 2026
DA - 2026/09/07
PB - PsyArXiv
DO - 10.31234/osf.io/s4qfr_v1
UR - https://osf.io/preprints/psyarxiv/s4qfr_v1/
AB - Whether word meanings partition reality uniformly across languages or reflect language-specific categorization remains debated. Universalist accounts assume shared conceptual repertoires whereas relativist accounts emphasize language-specific partitions of conceptual space. However, evidence has relied largely on a few semantic domains and behavioral measures, leaving unclear whether and how graded semantic overlap of translation-equivalents affect the neural dynamics of concept processing. Here we combined computational semantic alignment (SA), embedding-based representational similarity analysis, and electroencephalography to test whether graded SA predicts cross-language representational correspondence and neural correlates of implicit semantic processing. Thirty-five Chinese–English bilinguals completed a translation-recognition task involving 160 translation-pairs across 20 semantic domains and 160 unrelated controls. High-SA translations showed stronger cross-language representational correspondence than low-SA translations, whereas controls showed near-zero correspondence. High-SA translations elicited less negative N400 amplitudes and larger N400 translation-priming effects. Moreover, millisecond-by-millisecond comparisons across the 0–1,000 ms epoch detected high–low SA differences within the canonical N400 window for translation-pairs, but no corresponding differences for controls at any time point across the entire epoch, indicating that the N400 finding was detectable without restricting comparisons to a preselected time window and unique to concepts represented by translation-pairs rather than non-translation controls. The SA effects from translation-equivalents were further modulated by second-language proficiency. Overall, by linking cross-linguistic distributional semantic structure to model representational geometry and electrophysiological responses, this work convergingly supports a nuanced relativist account of word meanings, in which bilinguals are neurophysiologically sensitive to SA, with greater sensitivity at higher levels of second-language proficiency.
ER -