Target words selection for a Persian brain-computer-interface-based speller using language model

Abstract

Utilizing a language model in a brain-computer-interface-based (BCI-based) speller has been proven helpful in improving the performance of the system. Since it is important to evaluate the effect of the language model on the system, it is necessary to choose the words in a way that they can represent different levels of difficulty based on the language model. In this paper, we will give a brief introduction to the Persian BCI speller that we are going to develop. Then, we will explain how we select words with three levels of difficulty based on both a character-level and a word-level language model. We consider three levels of difficulty, named Easy, Medium and Hard, for which we have selected 3, 6 and 3 words, respectively. We then make a sentence for each of the words, so that the words are presented within a context. By doing so, the language model can predict the possibility of occurrence for each word. The novelty of this paper is in the fact that so far, there have not been implemented any Persian BCI spellers utilizing a language model as an aid in character recognition using the brain signals.