Reading Leaves a Dramatic Imprint on the Brain

A good book recreates the world so vividly that it activates some of the same brain regions that “everyday life” does, according to a recent PLOS One study.

In the study, innovative fMRI analyses of the brains of people reading a richly imaginative book showed that, when they read passages describing characters' movements, the same brain regions lit up that do when we observe motions of people around us. Characters’ intentions were likewise processed in a brain region where intentions of people in the real world are processed.

This occurred so reliably that, with 74 percent accuracy, researchers could tell which passages people were reading by looking at their fMRI scans. 

“This study adds tremendous knowledge to our understanding of the brain-bases for reading,” Georgetown University neuroscientist Guinevere Eden told Bioscience Technology. Eden, who studies reading, was uninvolved with the project. “It provides a picture of the integrated process of reading, not just the separate pieces of the process, and our brain’s full set of reactions to what we are reading—like tasting a delicious vegetable stew instead of just a piece of raw carrot. It is surprising to think about the multitude of things that happen in our brain when we pick up a book for a nice ‘quiet’ read.”

Visualizing multiple brain processes

For the study, Carnegie Mellon University researchers performed fMRI scans on eight people reading chapter nine of Harry Potter and the Sorcerer's Stone about a flying lesson. Cubic millimeter by cubic millimeter, they studied the scans in four-word segments, creating the first “integrated computational model of reading.” For each two-second scan, readers saw four words. For each word, the researchers found 195 features the brain would process. An algorithm calculated the activation of each millimeter of brain for each two-second scan, linking different word features with different brain areas.

In this way, the team predicted which of two passages were being read with 74 percent accuracy.

The model let the team analyze many brain sub-processes at the same time. It may be useful for visualizing, and combatting, many brain disorders, they reported.

Leila Wehbe, the Ph.D. student leading the study with Carnegie machine learning specialist Tom Mitchell, told Bioscience Technology the work clearly established that “some everyday life regions are also involved in processing stories.”

She was surprised that “when you run an experiment with subjects reading at a fast pace, using a slow and noisy imaging tool like fMRI, and having no repetitions to average to reduce the effect of noise, you can build a classifier which can detect which of two passages is being read—from brain activity. This was pleasantly encouraging, along with the fact that you can distinguish regions on the basis of the type of feature they are representing.” 

The team has big plans. “In the future, after we characterize how similar different people with the same reading characteristics are (normal readers), we can compare different populations of reader, normal readers and dyslexics, to see if we can find systematic differences in brain regions that explain the different reading profiles.”

This might let them diagnose different types of dyslexia by brain region, she said, “and suggest an educational strategy.”

Richer view of reading

Eden said she was not surprised certain “everyday” brain areas were lit up during reading. “Previous studies have shown that people activate brain areas of the visual motion system when they see objects associated with a movement (hammer); and they activate the motor system if they think about motor movements (playing tennis),” she said.

What she found most exciting was the way the model offered a richer view of reading. “We tend to constrain our brain imaging studies of reading by using simple single words that are processed in isolation, and also narrow these words to a limited focus,” she said. The words are chosen “for their semantic (meaning) or phonological (sound) properties.”

In the new study, “reading was examined for what reading really is in everyday life: the continuous reading of a popular book, rich in action and meaning, and giving rise to a continuous stream of visual imagery and emotions. These are the aspects of reading we enjoy and that we encourage beginning readers to appreciate. Now, through the investigators’ clever analysis approach, we know how the brain does this.”

Massachusetts General Hospital psychiatrist Evaline Fedorenco was less impressed with aspects of the study linking “everyday life” brain regions and reading.

New avenues for language research

“I am not a huge fan of the so-called ‘embodiment” framework,’” Fedorenco, also uninvolved with the paper, told Bioscience Technology.  “The general idea of this framework is that thinking about various activities relies on--- at least partially--- simulating those activities (hence the activation in the relevant sensory and/or motor cortices).”

But much evidence does not fit. “People with vastly different sensory experiences (congenitally blind individuals) end up with cognitive and neural architecture that’s essentially the same as that of seeing people.  So although sensory/motor activations may well accompany thinking about some concepts, I very much doubt that such simulations form the core part of those concepts,” said Feorenco.

Still, she sees a great future for the work. “The new paper takes a novel approach to understanding how language is implemented in brain tissue. Traditional experiments typically manipulate a single feature of a language stimulus (whether a word is frequent, or a sentence contains a non-local dependency) and see if it affects the response in any part of the language-responsive cortex. Instead, Wehbe and colleagues used fully naturalistic materials (a novel chapter) and examined a whole set of linguistic and meaning features at once.”

She was “excited” because the paper opens up new avenues for language research. “The time is now ripe for tackling the many open questions about language architecture using this method as a complement to traditional, controlled-experiment-based, methods.”

Very promising, cool method

For example, she said, the method could suss out “whether different components of language processing are localized within the fronto-temporal language network, or distributed across it. Using naturalistic materials can enable us to ask this question for many aspects of language simultaneously.”

The approach might improve evaluation of different models of syntactic complexity. “There are two main classes of models. One class--memory-based accounts--focuses on integrating incoming elements to earlier parts of the input, and the difficulty is quantified by the distance between an incoming element and the element earlier, which is structurally co-dependent with the incoming element. Another class--experience-based accounts--focuses on how predictable incoming information is from the preceding context.”

Naturalistic linguistic materials could be matched to these different complexity metrics, letting researchers see which “best predicts the BOLD response.”

She concluded, “This is a very promising, cool method.”

Follow Cynthia Fox at https://twitter.com/@cynthfox.