Handwriting-to-text (imagined handwriting) decoding
Decoding attempted handwriting instead of speech: the user imagines or attempts to write letters by hand, and a decoder reads the intended pen trajectories or letter identities from motor cortex. Willett et al. (2021) decoded attempted handwriting from a single intracortical array in the hand knob of motor cortex at about 90 characters per minute (roughly 18 words per minute), with over 99 percent accuracy after autocorrect, the first BCI to match able-bodied smartphone typing speed.
The temporally rich, distinctive pen-stroke trajectories of different letters make them highly separable, more so than the near-simultaneous key presses of an imagined keyboard.
The letter g has a downstroke and a loop that unfold over time as a distinct neural trajectory, so the decoder separates it from q more easily than two keys pressed a few milliseconds apart.
Temporally extended strokes make letters more separable than simultaneous key presses.
This is a language BCI that bypasses speech entirely; it is slower than the best speech BCIs but was a landmark demonstration that attempted fine-motor sequences are richly and reliably decodable.