Neural Syntactic Brain Region Maps And Complex Network Construction Based On Interpretable Machine Learning Methods

The brain is the most complex organ in humans,controlling processes such as cognition,behavior,and consciousness.With the rapid development of various neuroimaging techniques,the ability to measure human brain activity has been improving,which has led to great advances in the understanding of human brain function.Among them,functional magnetic resonance imaging(fMRI)is one of the main imaging techniques to study the complex cognitive neural activity of the human brain,as it can capture the functional activity of the human brain with high spatial resolution.Language is a hallmark of human civilization and a tool for human understanding of the world.The analysis of language mechanisms in the human brain has been a hot research topic in neuroscience.Based on the fMRI data collected by listening to natural language,the language processing mechanism of the human brain can be explored by whole-brain modeling analysis.This approach is more consistent with the response of the human brain under natural conditions and the results obtained are more systematic.Therefore,in recent years it has become a common research method to parse the language mechanisms of the human brain.This thesis investigates the syntactic processing mechanism of the human brain under natural conditions based on fMRI data collected while subjects listen to natural language.The semantic and syntactic effects that activate human brain functions in natural language are separated based on interpretable machine learning methods;the syntactic processing mechanisms of the human brain are systematically investigated for different syntactic structures.The main research work of this thesis is as follows.1.A framework for parsing the language functions of the human brain based on interpretable machine learning methods is constructed,and a brain region map of the neural syntactic functional strength of the human brain is shown.The existing research results on the syntactic mechanism of the human brain are often difficult to separate specific syntactic phenomena or to integrate experimental results under different conditions,so there is no systematic and in-depth analysis on the syntactic mechanism of the human brain.In this thesis,we take the words of natural language text as input features and the activation signals of human brain as output values,and model each voxel and interpret the model output based on the construction of corresponding sample sets.Subsequently,the interpretation results are statistically classified by syntactic structure type(e.g.,noun subject structure)with the help of modern grammar theory knowledge to obtain the functional strength of each voxel in processing each syntactic structure.Finally,the functional strengths of all syntactic structures are superimposed in terms of voxels,which is the overall syntactic functional strength of the corresponding voxels.Compared with previous studies,which only compare the syntactic functions of voxels by the performance of the model,this thesis quantifies the syntactic functional strength of voxels in order to explore the syntactic processing mechanism of human brain more accurately.2.a distributed processing model of the human brain for processing syntactic structures was discovered,demonstrating the selectivity of brain regions in processing different syntactic structures.In studies of global language processing mechanisms based on natural language,although relevant studies exist to separate semantics and syntax through various types of algorithms,most of them usually cannot continue to refine syntax to a more detailed level,resulting in subsequent processing of syntax only as a whole.However,the human brain is a very fine-grained operating system and the processing of syntax is also very complex,and treating syntax as a whole usually results in the loss of much-hidden information about the syntactic processing function.Therefore,this thesis explores each syntactic structure individually based on the strength of each voxel’s processing function for different syntactic structures.Unlike many studies of the human brain’s linguistic mechanisms that consider syntax as a whole,this thesis further investigates the human brain’s syntactic processing mechanisms at a more detailed level of syntactic structure and discovers the deeper operating mechanisms of the human brain during syntactic processing.3.Syntactic complex networks were constructed for each voxel in the human brain,refining the syntactic processing mechanisms down to the word pair level.By removing the generalization restrictions imposed by the syntactic structure,it is possible to study the differences in the processing of specific word pairs by different voxels,and static and dynamic syntactic networks of syntactic processing mechanisms are constructed for each voxel,which corresponds to the syntactic processing mechanism of the voxel as a whole and the syntactic processing mechanism of a specific moment,respectively.The smallest unit of information in the syntactic complex network is the syntactic relationship between words and words.Therefore,the syntactic complex network can precisely capture the linguistic processing mechanisms of voxels at the word level,showing the specificity of each voxel at the word level;meanwhile,the syntactic processing differences of different voxels at the word level are investigated according to the different network structures.Based on the syntactic differences of different voxels,a new method of dividing human brain regions is further proposed,in which voxels with similar syntactic processing mechanisms are grouped into one category of regions,and the syntactic functional connectivity relationships between regions are subsequently calculated,thus providing new insights into understanding the synergistic mechanisms of different brain regions in processing syntax.