A new approach for Korean word spacing incorporating confidence value of user's input

One of the problems of automatic word spacers is that the spacer mis-corrects the correctly spaced user's input. In this paper, we propose a new approach of estimating the confidence of user's input and incorporating the confidence value into the spacer for Korean word spacing. The confidence is defined as the probability that the input space is correct and is calculated by maximum likelihood estimation in the training data. The experimental results showed that the proposed method is more reliable than the previous method especially when the user's input is correct, and that the performance can be much improved when the confidence value is more precisely estimated.