A novel unsupervised non-iterative approach to word segmentation

Word segmentation is the crucial first step of natural language understanding (NLU) for Chinese corpora. Our early paper presented "Evaluation Selection and Adjustment" (ESA), an unsupervised approach to word segmentation. In this article, we present a novel non-iterative variation of ESA, comparing it with other similar methods and we get better performance. Besides that, we analyze "Balancing" and compare it with "Standardizing", another algorithm, to solve the problem of "how to evaluate the words of different lengths and compare them with each other for statistical methods of word segmentation". The experimental results show that "Balancing" is more effective than "Standardizing" for the task.