Neural Correlates of Cross‑Linguistic Syntactic Variation in Multilinguals

Abstract

Introduction

Syntax is a core property of human language: every known language is governed by structural rules that enable the construction of meaningful sentences. The cortical model of syntax by Matchin and Hickok [1] provides a framework for studying how the brain supports syntactic processing. The model distinguishes between two dimensions of syntax: morpho-syntactic linearisation and hierarchical structuring. Morpho-syntactic linearisation refers to how words are arranged in a sentence according to grammatical rules, such as word order and the addition of grammatical markers (e.g., the suffix -ed, as in walk → walked). This process is associated with the posterior inferior frontal gyrus (pIFG). In contrast, hierarchical structuring, i.e., the construction of syntactic trees — tree-like, nested representations that reflect syntactic relationships, is supported by the posterior middle temporal gyrus (pMTG).

Although this model captures key aspects of syntactic processing and its neural basis, it does not address cross-linguistic variation. Moreover, few studies have examined how bilingual and multilingual speakers process syntax, particularly when significant differences exist across the languages they speak.

To address these gaps, we ask the following research question: How do syntactic differences between languages relate to their neural processing in multilinguals?

Methods

For this study, we will use fMRI data from the NEBULA101 project [2] collected using a quasi-naturalistic listening task — the Alice language localiser [3]. Each of 101 participants in the study listened to both intact and acoustically degraded excerpts from Lewis Carroll’s “Alice in Wonderland” and its translations in two self-selected languages (L1 and L2), with 19 unique L1–L2 pairs spanning 13 languages.

Each language pair will receive a syntactic similarity score based on two components: (1) text-derived syntactic metrics (e.g., mean dependency distance between syntactically related words) computed from the localiser stimuli; and (2) typological grammatical features (e.g., presence of non-adjacent relative clauses) drawn from the Grambank database. 

We will compute individual-level contrasts in brain activity for L2 > L1 processing and use mixed-effects linear models to predict these activation differences. We hypothesise that including the syntactic similarity score as a regressor will explain more variance than models with only sociodemographic covariates. Since our syntactic similarity score mostly reflects differences in morpho-syntactic linearisation across languages, we expect stronger effects in the pIFG than in the pMTG, consistent with the cortical model of syntax [1].

References

[1] W. Matchin and G. Hickok, “The Cortical Organization of Syntax,” Cerebral Cortex, vol. 30, no. 3, pp. 1481–1498, Oct. 2019. doi: 10.1093/cercor/bhz180.

[2] A. Rampinini, I. Balboni, O. Kepinska, R. Berthele, and Narly Golestani, “NEBULA101: an open dataset for the study of language aptitude in behaviour, brain structure and function,” Scientific Data, vol. 12, no. 1, Jan. 2025. doi: 10.1038/s41597-024-04357-y.

[3] S. Malik-Moraleda et al., “An investigation across 45 languages and 12 language families reveals a universal language network,” Nature Neuroscience, vol. 25, no. 8, pp. 1014–1019, Jul. 2022. doi: 10.1038/s41593-022-01114-5.