Researchers report that what someone is about to say can be read, at least to some extent, from neuronal signatures in the brain: With the help of a special imaging method, they were able to recognize two speech sounds – several seconds before they formulated the subject. The results thus shed light on how the content and motor production of language are processed in the brain. This could benefit research into language disorders, for example. As the researchers emphasize, the method is not suitable for spying on thoughts.
We emit special sound sequences through which we convey information to our fellow human beings: The complex ability to communicate through language is a key element of human success. Interestingly enough, the system also has a “mute” component, as is well known: we also think in a form of language – an inner voice “babbles” inside us without us moving our mouths. Apparently, therefore, linguistic thinking is independent of the motor process of speaking. However, little research has been done on how our brain processes language content and production. One problem is that the neural basis of human language ability cannot be examined through animal experiments.
Brain activity reflected in neuronal magnetic fields
In order to gain new insights into the system in a non-invasive way, the researchers led by senior author Markus Siegel from the University of Tübingen have now used a special imaging method: magnetoencephalography makes it possible to record a person's brain activity from the outside by detecting the electrical nerve impulses caused by them Identify and localize magnetic fields. The process is comparatively powerful, but complex and is only available in a few facilities. However, there is a magnetoencephalography center in Tübingen that the researchers were able to use for their study.
They used the highly sensitive magnetic field sensors to look into the subjects' brains while they followed the instructions of the experiment: "We recorded the neuronal activity while they performed a simple vocalization task," explains Siegel. The study participants had to imagine one of two vowel sounds and sometimes also express it aloud. How they should behave was shown to them on a screen during the tests. The researchers then subjected the information obtained about brain activity to a so-called statistical pattern analysis.
Vowels are already “formulated” silently
As they report, characteristic signatures actually emerged: "We managed to identify the vowel in the brain signals that the test subjects were supposed to vocalize - a few seconds before the execution," says the first author Vera Voigtländer from the University of Tubingen. "This was independent of whether they later said it out loud or just imagined it." As the researchers explain, this reflects how the language content is represented abstractly in the brain. According to them, this proves that there is a general pattern underlying both the inner voice and loud verbal utterances.
The study results thus provide basic information on the processing of language in the human brain. Because this ability can be impaired in various diseases, this type of research also has clinical potential, the scientists say. "In the long term, the results of the study could also help to develop language support systems or make them more efficient," says co-author Steffen Hage from the University Hospital in Tübingen.
The team now wants to stay on the ball: "In the long term, we want to investigate different building blocks of language. We have currently only analyzed individual vowels," says Siegel. "The next step is to look at how the brain processes more complex sounds." However, this could raise concerns that the concept could one day be misused to spy on thoughts. But the researchers sound the all-clear: “The measurements are very time-consuming. The MEG requires a lot of space, the signals are very weak and no complex thoughts will be reflected in the sounds examined,” says Siegel.
Source: Hertie Institute for Clinical Brain Research (HIH), specialist article: PNAS, doi: 10.1073/pnas.2219310120