Generated by Codex with GPT 5.6 Sol XHigh

Thought Is More Than Inner Speech

Thinking often feels verbal. People silently rehearse an argument, name the pieces of a problem or search for the sentence that captures an idea. That experience makes it tempting to assume that language is the machinery of thought itself. In “Thinking without Words,” Scientific American senior editor Gary Stix interviews neuroscientist Evelina Fedorenko about evidence that points in a different direction: language is extraordinarily useful for communicating thought, but the brain does not need words to produce complex cognition.

Fedorenko did not begin with that conclusion. Her early training emphasized the idea that language made sophisticated human thought possible, perhaps because both language and many forms of reasoning depend on hierarchical structure. Words form phrases and sentences; mathematical or logical operations also combine smaller parts into larger structures. If one general-purpose brain system handled hierarchy, language might provide the scaffold for complex thought.

When Fedorenko began testing that idea, however, she repeatedly found separate systems. Brain regions specialized for language did not also perform the comparable structural work required by mathematics or music. The result led her to replace the language-as-thought hypothesis with a sharper distinction: the brain constructs thoughts through several cognitive systems, then uses language to convey them.

Two Ways to Separate Language from Cognition

One line of evidence comes from people with severe global aphasia after extensive damage to the brain’s left hemisphere. Some can neither understand nor produce ordinary language, yet they remain able to solve mathematical problems, complete logic puzzles, reason about the physical world and infer what another person believes. Researchers can explain the tasks with demonstrations and other nonverbal instructions, much as they do in studies of infants or nonhuman animals. These cases show that losing language does not necessarily erase the capacities usually grouped under thinking.

The second line comes from functional brain imaging. Fedorenko’s team first developed methods for locating each participant’s language network rather than relying on an average brain map. That distinction matters because functional regions do not sit in exactly the same place in every person; averaging across many brains can make neighboring but separate systems appear to overlap.

Once the researchers had identified an individual’s language regions, they could observe those regions while the person solved logic problems, used working memory, planned decisions or completed other nonlinguistic tasks. Across dozens of studies, the language network was largely quiet during these activities. Other networks handled the work. The repeated dissociation supports a strong claim: even high-level cognition, including social reasoning and novel problem-solving, can proceed without linguistic representations doing the computation.

That does not make language unimportant. Fedorenko describes it as a flexible channel for transmitting inner states—a practical substitute for telepathy. It allows one person to hand another an idea without requiring the second person to discover it independently. Much of what humans know arrives through this channel rather than through direct experience, and language lets communities coordinate, preserve social knowledge, teach skills and pass accumulated understanding to later generations.

This communicative role may also explain why the world’s roughly 7,000 spoken and signed languages share features that make information easier to produce, perceive, understand and learn. On this account, language was shaped less as a private engine for thought than as an efficient interface between minds.

No Single “Golden Ticket” for Human Intelligence

If language alone does not explain human cognition, no other single capacity is likely to do so either. Fedorenko points to several brain systems that expanded during human evolution: one supports language, another reasons about other minds, another tackles unfamiliar problems, and another integrates information across extended sequences of events. What makes human thought distinctive may be the combined sophistication of these systems rather than one uniquely powerful module.

The article’s argument is therefore about separation, not complete isolation. Existing methods are not very good at tracing how the language and thinking systems interact. The aphasia evidence shows that many cognitive operations can survive the loss of language, but it does not settle how learning a language affects cognitive development or how words may guide attention, memory and reasoning in everyday life. Likewise, finding that language regions are quiet during a task shows that they are not performing the core computation; it does not show that linguistic knowledge never influences the task.

Artificial intelligence offers a new way to investigate those unanswered questions. Large language models provide the first experimental systems outside a biological brain that display substantial linguistic competence. Researchers can vary their training data, compare speech and text exposure, and connect language models to symbolic tools for mathematics or problem-solving—manipulations that would be impossible or unethical in children. Fedorenko sees the models’ fluent language and uneven reasoning as consistent with the distinction found in humans, although the analogy is a research opportunity rather than proof that artificial and biological minds work alike.

The central takeaway is not that words are disposable. Language is one of humanity’s most consequential inventions of evolution because it moves thoughts between people and across generations. But the thoughts themselves arise from a broader cognitive architecture. Inner speech may accompany thinking, organize it or make it shareable; it is not the only form thinking can take.