采用最大熵模型实现维吾尔语句子边界识别,该模型的训练过程不需要提供手工收集规则、词性标注及形态分析,仅使用较容易获取的单词长度和音节等特征。为确定最佳特征模板,在特征空间上组合出不同特征模板进行测试。实验结果表明,最佳特征模板具有较强的鲁棒性,召回率可达97.72%。 The maximum entropy model is used to realize UWL sentence boundary recognition. The training process of the model does not need to provide manual collection rules, part of speech tagging and morphological analysis, and only uses the easy-to-obtain word length and syllables. To determine the best feature template, different feature templates are combined in the feature space for testing. Experimental results show that the best feature template has strong robustness and the recall rate can reach 97.72%.