Language is an important part of our lives, allowing us to share ideas and tell stories through speech. But how does your brain make sense of all the sounds we hear when someone is talking? This question has fascinated scientists for decades. One powerful idea is that the brain does not just sit back and listen, it behaves more like a fortune teller, trying to predict what the person is about to say based on the things they recently said. For instance, if someone says “liar liar, pants on ...”, your brain might already predict the next word: “fire” (Figure 1). That is because you have heard that phrase so many times before that your brain has learned the pattern and can guess what comes next, even before the word is actually spoken! This process of guessing is called “predictive coding” [1].

How Does Your Brain Predict What You Will Hear Next?

Scientists have used special methods to measure the activity of the brain as it makes its fortune teller-like predictions. They found that the brain can predict 2 seconds into the future—guessing what a person will say before the sound reaches the listener’s ears! [2].

One of the most famous studies of language and the brain discovered something very interesting. The researchers designed an experiment where people read surprising sentences like “He spread the warm bread with socks” and unsurprising sentences like “He spread the warm bread with butter” [3]. When the final word was not what the person predicted, that is when a prediction violation happens, the brain responded differently compared to when the word was expected. Specifically, scientists found that the brain’s response gets much stronger when the word that appears does not match the one it was expecting. So, when your brain guesses “butter” but sees “socks” instead, it gets a little jolt of surprise. The original finding was with written sentences, but this effect also occurs when you hear spoken sentences [4].

Interestingly, scientists also found that the brain responds to prediction violations not just for word meanings, but also when the words are put together in the wrong way. For example, when you hear a sentence like “The shirt was on ironed”, the brain’s response is much stronger than when you hear a sentence put together correctly, like “The shirt was ironed” [5]. What is especially fascinating is that different types of prediction violations, like a surprising word vs. a grammar mix-up, lead to different kinds of brain responses. So the brain is not just sensitive to something going unexpectedly; it also reacts differently depending on what kind of prediction was violated. Together, these findings show that the brain is constantly guessing ahead, and when those guesses are wrong, it notices.

In the decades since these discoveries, scientists have found that the brain does not just predict upcoming words. Our brains appear to make lots of different predictions at the same time. Some parts of the brain predict words, but other regions predict sentence meanings, and others predict speech sounds (scientists call these speech sounds phonemes) [6]. For example, scientists used a brain imaging method called Magnetoencephalography (MEG) to study brain activity while people listened to stories (you can read this Frontiers for Young Minds paper to learn more about how MEG works). Just like a fortune teller looks for signs to guess what will happen, the brain keeps track of the last few sounds in a word as clues to help it guess what sound is likely to come next. If the next sound is easy to predict, the brain moves on quickly. But if it is harder to tell what is coming, the brain holds on to the earlier sounds a little longer, using them to make a better guess. This shows that the brain does not treat all sounds the same, it adjusts how long it keeps information based on how confident it is about what is coming up next.

Finally, scientists have also found evidence that different parts of the brain share their predictions with one another. For example, when hearing the sentence “liar liar, pants on ...”, one part of the brain may predict that the next word will be “fire”, and share this prediction with a different part of the brain, which processes speech sounds, so that it can predict hearing an “f” sound (the first sound in the word “fire”). Moreover, researchers think that the brain makes predictions all the time and not only in certain situations such as when what you hear has rhymes [6].

How Does the Future Affect the Past?

We talked about how your brain can predict what is coming next, but sometimes your brain does something even more amazing. Just as a fortuneteller may sometimes make guesses about things that have already happened to someone, the brain also uses incoming information to predict the past. In one study, Gwilliams and colleagues [7] looked at what happens when the first sound of a word is unclear, like a mix between “p” and “b”. If your brain is not sure what it just heard, it does not decide right away. Instead, it listens to the rest of the word before responding.

For example, if the word ends with an “EE” sound, like parakEEt, the brain decides the unclear sound at the beginning of the word was probably a “p”. If the word ends with an “AY” sound, like barricAYde, the brain chooses “b” instead. It is kind of like autocorrect on your phone, it might be unsure at first, but once you finish typing, it uses the full word to guess what you meant (Figure 2). Even though our ears are hearing each sound one by one, we perceive these corrected words as a whole. The brain is constantly using every clue, even ones that come after the fact, to make sense of what we hear.

How do Scientists Use Language Models to Study the Brain?

Just like scientists use special machines to record brain activity, they also use computational tools to study how language works. One of the most powerful new tools is something called a Large Language Model (LLM). LLMs are computer based systems that learn patterns in language. They are trained by reading huge amounts of text—like websites, books, news stories, and even conversations—and learning which words often appear together. Using this knowledge, they can guess what word is most likely to come next in a sentence. For example, given the sentence “It is a sunny day, I want to go to the …”, the model might guess “beach”, “park”, or “ice-cream shop”.

These models are already used in many of the tools we interact with every day. When your phone auto-corrects a typo, suggests the next word while you are texting, or helps translate a sentence into another language, it is likely using an LLM behind the scenes. But LLMs are not just helpful in everyday technology, they are also powerful tools for science.

You might already see the connection! Since these models work by making predictions, they are a great tool to test the fortune telling powers of the brain. Now, this does not mean they understand or process language exactly the way human brains do. Yet, because they are designed to be really good at learning patterns and making predictions, scientists wondered if they could help us explore how the brain processes speech. So, they started using language models in neuroscience studies. Some scientists are now comparing these artificial fortune tellers (the language models) with the biological one (the brain).

One experiment, led by researcher Charlotte Caucheteux [8], tested whether language models that guess words might be using a strategy similar to our brains. Here is what they did: First, they had people listen to stories while recording their brain activity. At the same time, they gave the same stories to an LLM. Then, they compared the model’s predictions with the brain’s responses. Surprisingly, they found that the brain and the model were often in sync! And what made the match even stronger was when the model was not just trying to guess the next word, but was also trying to predict many words ahead. That is when the model’s predictions lined up even more closely with what the brain was doing. Interestingly, they found that some parts of the brain seemed to predict farther into the future than others, suggesting that different regions focus on predicting different aspects of speech—some parts predict the overall message of the sentence, while others focus on predicting just the next few words.

This work tells us that the brain probably uses all kinds of clues, not just the last word we heard but also the overall message of the sentence, to guess what is coming next. And it shows that using models to investigate brains could help us understand how we make sense of language in everyday life.

Mysteries to be Solved

In this article, you learned about how scientists have found clues that the brain makes predictions to help us understand speech. For example, the brain might use the meaning of a sentence to guess what smaller parts of speech, like words or sounds, will come next. You also saw that scientists are now using LLMs to better understand how the brain processes speech. Since these models are designed to make predictions, comparing them with brain activity is opening up exciting new ways to explore how we understand language.

But there are still mysteries about language and the brain waiting to be solved. For instance, we know that clever engineers designed LLMs and tools like auto-correct, so we understand how they work. But we do not yet have a clear idea of exactly how the brain pulls off its similarly remarkable feat of using future incoming information to make sense of the past. After all, the observation that both the models and our brains rely on predictions does not mean that they work the same way under the hood. Whether there is a specific group of cells in the brain dedicated to making predictions also remains unknown [9]. Who knows, maybe you will be the scientist who figures it out!

Glossary

Language: ↑ A system of spoken, manual (signed), and/or written symbols (words and expressions) that are used and understood by a large group of people to communicate (from this FYM paper).

Phonemes: ↑ The individual sounds made when trying to pronounce letters to create words (from this FYM paper).

Magnetoencephalography (MEG): ↑ A brain imaging machine that detects the magnetic field changes in the brain, resulting from activities of brain cells called neurons (from this FYM paper).

Conflict of Interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Acknowledgment

Funding from the BRAIN Foundation (A-0741551370), the Whitehall Foundation (2024-08-043), and the Esther A. & Joseph Klingenstein Fund, awarded to LG, was used to support researcher compensation.

AI Tool Statement

The author(s) declared that Generative AI was used in the creation of this manuscript. During the preparation of this manuscript, AI-assisted technologies were used only for minor editing purposes (e.g., suggestions for grammar correction). All content was prepared and reviewed by the authors.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.