Aligning statistical models with inference goals in the neuroscience of language: A dual-dependency taxonomy

Language unfolds over time and across multiple representational levels, from acoustics to meaning. Neural systems must, therefore, integrate temporal with representational structure, linking the evolving input with the hierarchical units it instantiates. These operations give rise to two fundamental statistical dependencies in linguistic and neural data: covariance, the instantaneous shared structure across features or recording sites, and temporal dependence, the influence of past states on the present state. As experiments become more naturalistic, neuroimaging and electrophysiological data increasingly express both forms of structure, producing correlated variables and continuous temporal dependencies that complicate interpretation. Because statistical models handle covariance and temporal dependence differently, they support distinct kinds of inference about language-brain mapping. We introduce a Dual-Dependency Taxonomy that classifies modeling approaches by the dependencies they represent. This framework clarifies the linguistic-neural relationships each model family can reveal, the questions they cannot address, and the methodological implications that follow.