Have you ever noticed that when someone says, “We are going to the beach this afternoon. Do not forget to bring your ____”. You might predict that they will mention something like “swimsuit” or “sunscreen” even before they finish. 

This ability to actively anticipate upcoming semantic content is one of the ways that the human brain achieves remarkable efficiency in language comprehension. 

This predictive processing relies upon two inextricably related processes: the anticipatory processing of future semantic information based on available context and retrieved knowledge, and the integration of new inputs with a representation of preceding context, which includes top-down predictions generated from this context. 

This complex process relies on a hierarchical brain network that is widely distributed across the frontal, temporal, and parietal lobes, as well as sensory brain regions (Figure 1A). However, it remains unclear how they interact to achieve efficient language comprehension. 

To investigate this, a research team led by Dr. LI Xiaoqing from Institute of Psychology, Chinese Academy of Sciences conducted a functional magnetic resonance imaging (fMRI) experiment. Participants were asked to comprehend sentences presented in a part-by-part fashion (Contextual part; Anticipation phase; Integration phase, for details please see Figure 1C), with the semantic constraints of sentential contexts being manipulated to influence the predictivity of upcoming semantic content (Strong vs. Weak). 

Using this paradigm, they examined the functional connectivity patterns of the temporoparietal junction (TPJ) and inferior frontal gyrus (IFG) situated at the top of the predictive processing hierarchy during anticipation and integration phases. 

Figure 1. (A) Organisation of hierarchical predictive processing revealed by Caucheteux et al. (2023); (B) The TPJ and IFG seeds; (C) Task illustration. Image by Dr. LI Xiaoqing's team.

They found that when upcoming semantic content was highly predictable in the strong-constraint contexts during the anticipatory phase, both left TPJ and bilateral IFG showed stronger visual coupling, while right TPJ showed stronger connectivity with the regions within control, default mode, and visual networks, including IFG, parahippocampal gyrus, posterior cingulate, and fusiform gyrus (Figure 2 and 3). These connectivity patterns were weaker when the predicted semantic content appeared during the integration phase, in line with predictive coding theory. 

This pattern is reversed when upcoming word is unpredictable in the low-constraint context; connectivity was stronger after this lowly predictable word has been presented, allowing semantic integration with preceding low-constraint context. 

Figure 2. Functional connectivity seeding from (A) right TPJ and (B) left TPJ seeds. Image by Dr. LI Xiaoqing’s research team.

 Figure 3. Functional connectivity seeding from (A) left IFG and (B) right IFG seeds. Image by Dr. LI Xiaoqing’s research team.

These results suggest that both top-down semantic prediction and bottom-up integration during predictive processing are supported by flexible coupling of frontoparietal regions with control, memory, and sensory systems. This flexible connectivity of higher-order frontoparietal cortex with other brain systems contributes to our efficient language comprehension.

This work was supported by the National Natural Science Foundation of China, and Institute of Psychology, Chinese Academy of Sciences. 

This paper is now accessible online in the Journal of Neuroscience. 

He, Y., Shao, X., Liu, C., Fan, C., Jefferies, E., Zhang, M.*, & Li, X.* (2024). Diverse frontoparietal connectivity supports semantic prediction and integration in sentence comprehension. Journal of Neuroscience. e1404242024. 

Reference: 

Caucheteux, C., Gramfort, A., & King, J. R. (2023). Evidence of a predictive coding hierarchy in the human brain listening to speech. Nature human behaviour, 7(3), 430-441.