Lamarre, M. | Search bioRxiv

bioRxiv2025

Concreteness shapes semantic representations in bilingual brains

Behavioral studies show that humans process concrete words more similarly across languages than abstract words. This suggests that in bilingual brains, semantic representations may be more similar across languages for concrete concepts than for abstract concepts, but existing neuroimaging evidence is inconclusive. Here, we analyzed functional magnetic resonance imaging (fMRI) data from fluent Chinese-English bilinguals who read several hours of naturalistic narratives in both languages. We used encoding models to estimate voxelwise tuning towards concrete and abstract concepts in each language separately. We then quantified the similarity of cortical semantic representations across languages. First, we find that the cortical organization of concreteness tuning is consistent across languages. Second, semantic representations are similar across languages for both concrete-tuned voxels and abstract-tuned voxels. Third, we find that for abstract concepts, cross-language similarity of semantic representations may be driven by emotionality. Overall, these findings reveal how concreteness affects semantic representations in bilingual brains.

neuroscience↗

bioRxiv2022

Attention weights accurately predict language representations in the brain

In Transformer-based language models (LMs) the attention mechanism converts token embeddings into contextual embeddings that incorporate information from neighboring words. The resulting contextual hidden state embeddings have enabled highly accurate models of brain responses, suggesting that the attention mechanism constructs contextual embeddings that carry information reflected in language-related brain representations. However, it is unclear whether the attention weights that are used to integrate information across words are themselves related to language representations in the brain. To address this question we analyzed functional magnetic resonance imaging (fMRI) recordings of participants reading English language narratives. We provided the narrative text as input to two LMs (BERT and GPT-2) and extracted their corresponding attention weights. We then used encoding models to determine how well attention weights can predict recorded brain responses. We find that attention weights accurately predict brain responses in much of the frontal and temporal cortices. Our results suggest that the attention mechanism itself carries information that is reflected in brain representations. Moreover, these results indicate cortical areas in which context integration may occur.

neuroscience↗