The accuracy of readback plays an important role in flight safety.However,readback semantic inconsistency may occurs during radiotelephony communication.Thus,a method based on deep CNN(convolutional neural network)is proposed to verify semantic consistency of readback content.Words in radiotelephony communication corpus are transformed into word vectors.On top of the word vectors,CNN model is used to obtain the semantic vectors of readback sentence pairs, then cosine measurement is applied to measure the semantic similarity of readback sentence pairs. Finally, consistency verification is implemented by classifier. Experiment results show that semantic consistency of radiotelephony communication can be verified by CNN model effectively. The average accuracy of the experiments is up to 82.5%.%陆空通话复诵的准确性对飞行安全有着重要作用.由于飞行员和管制员在复诵过程中会无意识地产生一些问题,提出利用深度卷积神经网络(deep CNN,deep convolutional neural network)方法对陆空通话内容的语义进行一致性校验.首先将陆空通话语料库中的词语进行词向量转化,在此基础上,利用CNN得到飞行员和管制员复诵对的语义向量,然后利用余弦相似度对其语义相似性进行度量,最后通过分类器完成一致性校验.通过实验可证明,CNN可有效地实现陆空通话语义一致性校验,且实验精度结果能够达到82.5%.
展开▼
