An Artificial Immune Network approach for Pinyin-to-character conversion

This paper proposes a novel approach based on Artificial Immune Network for dealing with the task of Pinyin-to-character (PTC) conversion. The researches in recent years have nearly indicated that the sparse data problem and the independent identical distribution (iid.) assumption are two main difficulties of improving the PTC performance, and these two problems widely exist in the supervised learning methods. This paper presents an online learning approach to overcome the above problems. This model has a kind of ability of adaptively adjustment by using the feedback information, and in this model, the discriminative function gives the partial ordering relation of each immune chain so as to implement the partial perception online learning. The experiments show that our PTC conversion method based on the online learning technology can achieve a better performance than the n-gram language model, and this kind of improvement is hardly acquired by the classical supervised learning methods.