Shuguang Yang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Shuguang Yang has a total of 8 co-authors. Here are the top co-authors by number of shared publications:
- Feipeng Chen
School of Linguistic Sciences and Arts, Jiangsu Normal University, China
1 shared publication - Gangyi Feng
Department of Linguistics and Modern Languages, The Chinese University of Hong Kong
1 shared publication - Peilun Song
School of Electrical and Information Engineering, Zhengzhou University, Zhengzhou, China
1 shared publication Suiping Wang
School of Psychology, South China Normal University
1 shared publicationXiujuan Geng
Department of Linguistics and Modern Languages, The Chinese University of Hong Kong
1 shared publication- Yiming Yang
School of Linguistic Sciences and Arts, Jiangsu Normal University, China
1 shared publication
Latest Publications
- 15 June 2026
- Published by Society for Neuroscience in Journal of Neuroscience
- Vol. 46 (27)
Multi-network Topology Underlying Individual Language Learning Success
Adult language learning varies widely among individuals, with some learners quickly acquiring knowledge and skills while others struggle with specific components or overall proficiency despite similar exposure. This variability, once linked to frontotemporal language regions, is increasingly seen as originating from distributed networks involved in attention, control, and memory. The role and organization of these networks in explaining these differences remain unclear. We hypothesized that intrinsic multi-network connectivity underpins these variations, revealing potential neuromarkers of interactions among systems beyond language regions. We tested this in 101 healthy adults (72 females and 29 males) using multimodal neuroimaging before 7 d of artificial language training across six tasks targeting auditory and speech categories, words, morphosyntax, and sentence structures. We identified one general component shared across tasks and five task-specific ones. Using cross-validated predictive modeling and graph-theoretic metrics, we found that the general component’s learning outcome and rate were primarily driven by the dorsal attention and frontoparietal networks. Their local efficiency was a strong predictor, highlighting local resilience and mesoscale segregation. Local connectivity dominated in association cortical networks, while global integration occurred in subcortical regions, reflecting a balance between segregation and integration that influences learning. Only task-specific word learning was predictable, relying on default-mode and frontoparietal hubs. Single-modality predictions were weaker, emphasizing the value of multimodal approaches. These findings advance network-level understanding by showing that intrinsic network topology underlies individual success in language learning, supporting a multiple-system model in which control, attention, default, and subcortical networks work together to shape learning trajectories.
- 15 June 2026
- Published by Society for Neuroscience in Journal of Neuroscience
- Vol. 46 (27)
Multi-network Topology Underlying Individual Language Learning Success
Adult language learning varies widely among individuals, with some learners quickly acquiring knowledge and skills while others struggle with specific components or overall proficiency despite similar exposure. This variability, once linked to frontotemporal language regions, is increasingly seen as originating from distributed networks involved in attention, control, and memory. The role and organization of these networks in explaining these differences remain unclear. We hypothesized that intrinsic multi-network connectivity underpins these variations, revealing potential neuromarkers of interactions among systems beyond language regions. We tested this in 101 healthy adults (72 females and 29 males) using multimodal neuroimaging before 7 d of artificial language training across six tasks targeting auditory and speech categories, words, morphosyntax, and sentence structures. We identified one general component shared across tasks and five task-specific ones. Using cross-validated predictive modeling and graph-theoretic metrics, we found that the general component’s learning outcome and rate were primarily driven by the dorsal attention and frontoparietal networks. Their local efficiency was a strong predictor, highlighting local resilience and mesoscale segregation. Local connectivity dominated in association cortical networks, while global integration occurred in subcortical regions, reflecting a balance between segregation and integration that influences learning. Only task-specific word learning was predictable, relying on default-mode and frontoparietal hubs. Single-modality predictions were weaker, emphasizing the value of multimodal approaches. These findings advance network-level understanding by showing that intrinsic network topology underlies individual success in language learning, supporting a multiple-system model in which control, attention, default, and subcortical networks work together to shape learning trajectories.
- Open Access
- 7 July 2023
- Published by Association for Computing Machinery (ACM)
- p. 165-170
A Study on Semantic Understanding of Large Language Models from the Perspective of Ambiguity Resolution
Generative large language models (LLMs) have demonstrated outstanding performance in language understanding and generation tasks. Whether current LLMs can truly understand meaning or not is an open question. Semantic disambiguation is an important indicator to test machine language ability. To this end, three experiments were conducted to test the ambiguity resolution ability. In Experiment 1, the LLMs were asked to make a semantic discrimination on ambiguity phrases. The ChatGPT (May 3 Version) and ChatGPT Plus (May 3 Version) showed a random level performance of discrimination. To make the task easier for the LLMs, ambiguity sentences were presented with context in the Experiment 2. While the LLMs unified the context to fill the position for ambiguity word, the performance of semantic discrimination on these sentences remained at random level. In addition to test the performance of LMMs, prediction is one of the key ability of LLMs. The prediction by unifying the context is one of the key skill for LLMs. Therefore, the Experiment 3 further test the similarity between the human prediction mechanism and the strong generation capabilities of LLMs. Weak similarities were only found under the condition of local constraint features gradually accumulating, specifically, the cosine similarity of predicted sentence semantic vectors and ending word semantic vectors negatively correlated with close probability only. These series of experiments demonstrate that, it is far away to claim that LLMs truly understand human sentences.
- 7 July 2023
- Published by Association for Computing Machinery (ACM)
- p. 165-170
Open AccessA Study on Semantic Understanding of Large Language Models from the Perspective of Ambiguity Resolution
Generative large language models (LLMs) have demonstrated outstanding performance in language understanding and generation tasks. Whether current LLMs can truly understand meaning or not is an open question. Semantic disambiguation is an important indicator to test machine language ability. To this end, three experiments were conducted to test the ambiguity resolution ability. In Experiment 1, the LLMs were asked to make a semantic discrimination on ambiguity phrases. The ChatGPT (May 3 Version) and ChatGPT Plus (May 3 Version) showed a random level performance of discrimination. To make the task easier for the LLMs, ambiguity sentences were presented with context in the Experiment 2. While the LLMs unified the context to fill the position for ambiguity word, the performance of semantic discrimination on these sentences remained at random level. In addition to test the performance of LMMs, prediction is one of the key ability of LLMs. The prediction by unifying the context is one of the key skill for LLMs. Therefore, the Experiment 3 further test the similarity between the human prediction mechanism and the strong generation capabilities of LLMs. Weak similarities were only found under the condition of local constraint features gradually accumulating, specifically, the cosine similarity of predicted sentence semantic vectors and ending word semantic vectors negatively correlated with close probability only. These series of experiments demonstrate that, it is far away to claim that LLMs truly understand human sentences.