关键词: 句子边界检测；语素边界检测；黏着语；深度学习

Abstract:

Sentence and token boundaries detection is one of the important tasks in natural language processing. In order to avoid task-specific feature engineering, we have proposed character-level based neural network model for token and sentence segmentation (CNN-TSS). In order to share the information from these tasks, we have treated them as a combined task. The experimental results show that CNN-TSS can achieve high-accuracy without using any external features.

Key words: sentence boundaries detection;token boundaries detection;agglutinative language;deep learning