Language comprehension is mainly carried out by specialized cognitive mechanisms that are language-specific and are not engaged in other high-level cognitive functions. Additionally, much research has found that domain-general resources—e.g., executive function, working memory, or inhibition—also contribute comprehension. However, prior studies are overwhelmingly based on cleverly designed artificial tasks, which do not mimic comprehension “in the wild”, where many processes must dynamically interact. What is the division of labor between language-specific and domain-general resources in real-world comprehension? In this talk, I will describe studies that employ naturalistic paradigms to probe the contributions of these cognitive mechanisms, and use fMRI to directly contrast the domain-specific “core language network” and the domain-general “multiple-demand (MD) network”. The studies examine the functional profiles of these networks based on psycholinguistic constructs (prediction of word sequences and hierarchical structure), behavioral measures (reading times), and data-driven analyses. The findings challenge two decades of studies about the role of executive resources in comprehension. Therefore, I will end by describing a large-scale meta-analysis of previous task-based studies from ~500 participants that, when cast in light of the new results, corroborates them.