Automatic Korean word spacing using Pegasos algorithm

In this paper, we describe an automatic Korean word spacing approach using structural SVMs to relax the independence assumptions required by HMMs. We use a Pegasos algorithm for fast training of structural SVMs. We show the Pegasos algorithm for structural SVMs outperforms significantly HMMs and traditional binary SVMs, and it is much faster than CRFs and structural SVMs without loss of performance.

► We propose Korean word spacing models using the Pegasos algorithm for structural SVMs. ► We show that the Pegasos algorithm outperforms HMMs and traditional binary SVMs. ► We show that the Pegasos algorithm is faster than CRFs and structural SVMs. ► We show that the second-order Markov model has the best performance.