The First Brain-Computer Interface Utilizing a Turkish Language Model

One of the widely studied electroencephalography (EEG) based Brain-Computer Interface (BCI) set ups involves having subjects type letters based on so-called P300 signals generated by their brains in response to unpredictable stimuli. Due to the low signal-to-noise ratio (SNR) of EEG signals, current BCI typing systems need several stimulus repetitions to obtain acceptable accuracy, resulting in low typing speed. However, in the context of typing letters within words in a particular language, neighboring letters would provide information about the current letter as well. Based on this observation, we propose an approach for incorporation of such information into a BCI-based speller through a Hidden Markov Model (HMM) trained by a Turkish language model. We describe smoothing and Viterbi algorithms for inference over such a model. Experiments on real EEG data collected in our laboratory demonstrate that incorporation of the language model in this manner leads to significant improvements in classification accuracy and bit rate.