Language-independent call routing using the large margin estimation principle

There has been an increasing research interest in natural language call routing (NLCR) applications. One of the challenges often encountered in NLCR applications is the difficulty of performing language-dependent tasks such as morphological analysis of words and stop-word filtering. In this paper, we propose a novel NLCR system which does not depend on languagespecific information and thus, it can be ported easily to many languages. The proposed system is based on a combination of character c-gram terms and discriminative training using large margin estimation principle. Compared to traditional vectorbased NLCR methods, the proposed NLCR system does not need language-dependent processing and achieves around 1% increase in the classification accuracy.