Semantic Representation And Verification Of Aviation Radiotelephony Communication Based On RNN

In the civil aviation transportation,many accidents are caused by Aviation Radiotelephony Communication(ARC).To solve this problem,International Civil Aviation Organization(ICAO)has improved the communication standard of air traffic.However,the accidents caused by ARC problem still happen unavoidably,and the incorrect readback between pilots and air traffic controllers is one of the mainly factors leading to accidents.Thus,verifying the semantic consistency of ARC has important practical significance for the safety of civil aviation transportation.The latest studies indicate that Recurrent Neural Network(RNN)is excellent in representing the semantics of sentences.RNN is a natural feed-forward neural network,its hidden layer can capture the information of sequences.Long Short-Term Memory Recurrent Neural Network(LSTM-RNN)has been successfully applied to address the vanishing gradient problem of RNN.So RNN and LSTM-RNN are adopted to propose semantic representation and verification models for civil aviation communications,respectively.First,the corpus is constituted by the sentence pairs of readbacks based on the actual radiotelephony recordings.Single words after sentence pairs segmentation are represented using one-hot vector and word2 vec.Then,RNN and LSTM-RNN model are applied to represent semantic vectors of sentence pairs,respectively.Third,the cosine similarity of semantic vectors is employed to measure the semantic similarity of sentence pairs.Finally,some common classifiers are utilized to verify the semantic consistency of sentence pairs.The experimental results show that both RNN and LSTM-RNN are suitable to represent semantic vectors of sentence pairs,and the consistency between semantic vectors can be verified effectively.The results of experiments also prove that the performance of LSTM-RNN model outperforms RNN model,which achieve a higher accuracy in our task.