Combining forward and backward processing for Korean BaseNP identification
Sheen-Mok Lee, In-Ho Kang, Gil Chang Kim · 2004
BaseNP identification is an important NLP task. We propose a Korean baseNP identification method that uses a state-based model focusing on directionality of the state transition. In Korean baseNP identification, the degree of difficulty in identifying a start position is different from that of identifying an end position. This fact implies that the property of the model can be changed if the state transition is processed in a right-to-left manner. We propose the backward processing model that processes the state transition in a right-to-left manner. In addition, we combine it with the forward processing model that processes the state transition in a left-to-right manner. The results of both models are better than those of the previous ones in Korean baseNP identification. We can achieve further improvement by combining the two models.