We aim to enable the smooth communication of persons physically unable to speak. In our past study, we proposed three Japanese input methods using a portable eye-controlled communication device for users with conditions such as cerebral palsy or amyotrophic lateral sclerosis (ALS). However, these methods require nearly 30 seconds to cycle through one Japanese character. In this paper, we suggest a method to estimate the input word using the clues of nearby characters and accumulated experience. In addition, to raise the precision of the prediction, we use the connection between words based on a thesaurus. We have realized precise word conversion via a few input letters, as proved by the result of the simulation experiment.