The research team encountered challenges when attempting to decode more intricate phrases.
Exploring New Horizons
After establishing mental privacy protections, the researchers began their evaluation of an inner speech decoding system using a set of predefined words. Patients were instructed to imagine verbally articulating sentences displayed on a screen. The results showed variability in effectiveness, with one patient achieving accuracy rates of up to 86 percent using a limited vocabulary of 50 words. However, the accuracy dipped to 74 percent when the vocabulary expanded dramatically to 125,000 words.
Transitioning to unstructured inner speech, the limitations of the brain-computer interface (BCI) quickly became evident. Their initial test involved patients observing arrows that pointed in various directions on a screen. Following a short interval, participants were tasked with recalling and reproducing the sequence, such as “up, right, up,” using a joystick. While the system made some attempts at decoding this internal thought process, it only performed slightly better than random chance.
In a subsequent experiment, Krasa and his team explored the decoding of more complex, cue-free phrases. Participants were asked to think of the names of their favorite foods or to recall memorable quotes from movies. Krasa remarked, “This didn’t work. What came out of the decoder was kind of gibberish.”
Currently, Krasa views the inner speech neural prosthesis as a proof-of-concept. “We didn’t anticipate this would be feasible, but we accomplished it, which is thrilling! However, the error rates are too high for practical daily use,” he acknowledged. He suggested the limitations might lie in the hardware, specifically the number of electrodes used and the precision of neuron signal recordings. Additionally, he noted that representations of inner speech could be stronger in different brain regions rather than primarily in the motor cortex.
Krasa’s team is pursuing two related projects stemming from their work on inner speech decoding. “The first project examines how much quicker an inner speech BCI could operate compared to traditional speech attempts,” he explained. The second project focuses on individuals with aphasia, a condition where patients have the ability to control their mouths but struggle to produce spoken words. “We aim to evaluate whether inner speech decoding could provide assistance for them,” Krasa stated.
Cell, 2025. DOI: 10.1016/j.cell.2025.06.015
