Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Traces in the Brain: Neural Evidence for Syntactic Movement in English and Chinese
Poster Session C, Thursday, October 1, 10:45 am - 12:45 pm, Wangari Maathai
Yuhan Huang1, Zhengwu Ma1, Yuqi Jin1, Beth Chan2, Zheng Shen2, Jackie Yan-Ki Lai1, John T. Hale3, Jixing Li1; 1City University of Hong Kong, 2National University of Singapore, 3Johns Hopkins University
Introduction. Syntactic movement is a central concept in generative linguistics for explaining non-canonical word order and long-distance dependencies. In movement-based accounts, surface word order is derived from deeper structural representations through displacement operations, which leave traces at the original base positions. Such representations have played a key role in formal analyses of passives, wh-dependencies, and related constructions, yet their psychological and neurobiological reality remains debated. Alternative accounts explain processing difficulty in terms of surface dependency relations, memory demands, or usage-based constraints, without appealing to transformations or multiple levels of syntactic representation. This question is especially important from a cross-linguistic perspective, because English relies more heavily on overt displacement, whereas in Chinese many dependencies are realized in situ or mediated by discourse-pragmatic mechanisms. Here, we test the neural reality of syntactic movement in English and Chinese during naturalistic listening, using syntactic node counts, trace-based regressors, and word embeddings derived from X-bar–style tree annotations. Methods. We analyzed the English and Chinese subsets of a publicly available naturalistic functional magnetic resonance imaging (fMRI) dataset (Li et al., 2022). The dataset included 49 native English speakers (30 females; mean age = 21.3 ± 3.6 years) and 35 native Chinese speakers (15 females; mean age = 19.3 ± 1.6 years), who listened to audiobook versions of The Little Prince in their native languages. We manually annotated all sentences in the English (N = 1,502) and Chinese (N = 1,577) stimuli with X-bar–style syntactic trees. From these annotations, we derived top-down and bottom-up syntactic node counts, as well as a binary trace regressor. As a comparison, we also generated Penn Treebank–style context-free grammar (CFG) trees for all sentences and computed corresponding node-count measures. We aligned these syntactic predictors with fMRI responses using vertex-wise general linear models. In parallel, we extracted deep- and surface-structure word embeddings from each layer of LLaMA 3.1 8B. Deep structure was operationalized as a reordering of the same lexical items according to X-bar dependency relations, whereas surface structure preserved the original word order. We used banded ridge regression to model neural responses from the selected deep- and surface-structure embeddings, together with a combined embedding defined as their average. At the group level, statistical significance was assessed separately for English and Chinese using one-sample, one-tailed t-tests with a cluster-based permutation procedure. Results. GLM analyses showed that traces and all syntactic node-count measures significantly predicted activity in canonical language regions in both languages, with effects centered in the left temporal cortex. In English, trace effects and the contrast favoring X-bar over CFG top-down node counts additionally recruited left frontal regions. In the model-based encoding analysis, combined embeddings outperformed either deep- or surface-order embeddings alone in English, suggesting that neural responses were best explained by representations integrating both underlying dependency structure and surface word order. In Chinese, however, surface-order embeddings outperformed both deep-order and combined embeddings, suggesting that neural responses were more strongly driven by overt sequential structure than by reordered dependency-based representations.
Topic Areas: Syntax and Combinatorial Semantics, Computational Approaches