Chinese Prosodic Phrase Prediction Based on Shallow Semantic Features

摘要

Syntactic structure features can improve the performance of prosodic phrase prediction. But only using the syntactic structure features, the performance of the algorithm is worse than the traditional text features. In this paper, we use the statistical machine learning method (CART, Adaboost and CRF) for prosodic phrase prediction based on the shallow semantic features. Experiments show that the shallow semantic features can effectively improve the performance of prosodic prediction model. And we also optimized the features, and divided them into the global and local semantic structures. The optimized experiments show that the optimized features can improve the performance of the model.