In their classic 1998 textbook on cognitive neuroscience, Michael Gazzaniga, Richard Ivry, and George Mangun made a sobering observation: there was no clear mapping between how we process language and what was happening in our brains. Fast forward to today, and the landscape has changed dramatically. We now have a much more sophisticated understanding of language’s neural architecture—not just the basic abilities of speaking and understanding, but the intricate computational machinery that makes human communication possible.

Although there is still much to learn, the progress has been astounding. Consider the figure here, for example, which shows the best overall model we had of the neural architecture of language in 1998—a model that hadn’t improved much since the 1800s—compared to what we can assemble today. I am honored to have witnessed these advances firsthand and played a small part in what the field has accomplished.

Let me summarize here what I take to be most important takeaways, focusing on a new integrated framework, the Linguistic-Sensorimotor Model (LSM), which offers a fresh perspective that bridges traditional linguistic theory with our understanding of how the brain controls movement and processes sensory information.

The Core Insight: Language as a “Species” of Sensorimotor Processing

The most provocative claim of the LSM is that language processing shares the same fundamental neural architecture as non-linguistic sensorimotor systems—like reaching for a coffee cup or tracking a moving object with your eyes. This doesn’t mean language is “just” movement or that decades of linguistic research are invalid. Rather, it suggests an evolutionary relationship: the two types of systems are homologous.

Think of it this way: humans are a species of ape. This doesn’t reduce us to chimpanzees or deny our unique qualities—it simply acknowledges our shared evolutionary ancestry and the homologous traits we inherited. Similarly, language processing is a “species” of sensorimotor processing. The brain didn’t invent entirely new computational machinery for language; instead, it repurposed and specialized existing sensorimotor control architectures that had evolved for coordinating perception and action.

This evolutionary perspective helps explain why language systems are organized hierarchically (like other sensorimotor systems), why they involve both “sensory” and “motor” components at each level, and why feedback control mechanisms—essential for adjusting movements in real-time—also operate during speech planning and comprehension.

The principles of sensorimotor-like architecture and hierarchical organization can be seen in the figure above, where like-colors represent connected sensory-related (posterior areas) and motor-related (frontal areas) systems, and which are organized hierarchically: pink-green networks roughly map to higher-level morphosyntactic systems, pink-yellow-black maps to phonological systems, and light green maps to low-level sensorimotor control.

Translation Systems

Processing architectures that involve different kinds of representations, sensory-related and motor-related in our case, need a way to communicate with each other, a means to translate one kind of code to the other. Neuroscientists working on nonlinguistic sensorimotor control systems have proposed the existence of such translation systems, and research on language systems has identified homologues. The case for a phonological translation system involving area Spt has been extensively studied and is on fairly firm evidentiary ground. Recent evidence suggests another homologous network in a region called the posterior supramarginal gyrus that serves a similar translation function at the level of morphosyntax.

One of ways we know such translation systems exist is because when they are damaged by stroke, for example, they give rise to unique language syndromes: fluent speech (because motor-related systems are intact), but with error-prone speech output (because motor-related systems can’t error check their plans against sensory-related targets), and with preserved language perception and understanding (because the ventral stream systems are intact). Classically these are called “conduction aphasias” due to the failure of accurately conducting information between sensory and motor systems.

Neuroscience Essential Reads

The Receptive-Expressive Asymmetry

Conduction aphasias preview another important and more general principle of the architecture of language. This is the existence of a functional asymmetry in the involvement of sensory-related versus motor-related components in language comprehension versus language expression (speaking). Language comprehension, under most circumstances, involves just the ventral networks, the temporal-lobe, sensory-related systems. Expressing language, in contrast, involves the whole network, both ventral and dorsal systems. This fact should not surprise us given that language architecture is homologous to nonlinguistic systems. For example, we don’t need to motorically interact with snakes or birds or any object to understand what it is; the ventral stream is generally enough. However, if we want to interact with them using our frontal lobe action system, we still need to accurately perceive their sensory properties; both dorsal and ventral networks are involved. Language is no different.

Hemispheric Asymmetry

Strong left dominance of language is traditionally held to be a core feature of the neural organization of language. New research has shown, however, that different linguistic abilities are differently lateralized. In the case of perceiving and understanding words, for example, left dominance appears to be atypical, with bilateral organization representing the modal tendency. The details of the lateralization story have yet to be revealed.

Glimmers of Further Complexity

Despite the remarkable advances in understanding the neural architecture of language, there is much more to learn. Already, we have strong evidence for two parallel networks of speech motor coordination, not just the classical inferior frontal one. Research on prosody, a relatively neglected aspect of language processing, is turning up new and exciting implications for normal and disordered language function. Ideas regarding how semantic information is stored and used has advanced significantly as well and is still being refined and reimagined. Further, there are some areas that are emerging as core hubs for language processing but have yet to studied in detail (the inferior frontal sulcus comes to mind for me). And then there’s the question of dynamics. I’ve been talking about architecture here, but the neural dynamics within this broad architecture and on finer scales, will need to be worked out before we have a detailed understanding of how language works. Happily, there is much ongoing work on this topic.

Conclusion

When I started my research program aimed at understanding the neural architecture of language in the 1990s, I would have been overjoyed had I known how much progress the field would make in just two and half decades. The prospect for further advances and for many applications to treatments of language disorders has never been brighter. Advances in diagnosis and treatment of speech and language disorders have steadily accrued, and already there are many active clinical research programs aimed at using neuroprosthetics to enable people to speak again after having lost the ability due to injury or disease. All of this is made possible by hundreds of linguists, psychologists, neuroscientists, engineers and others who asked a simple question: how is it that the brain enables language?

Adapted from Wired for Words: The Neural Architecture of Language by Gregory Hickok, published by MIT Press.