Research on Construction and Classification of Civil Aviation Cabin Corpus Based on Linguistics

Abstract

For a long time, flight accidents caused by errors in land and air communications have not been uncommon. In order to reduce errors in land and air calls, the International Civil Aviation Organization and national regulatory agencies continue to improve the language standards for Chinese language communication in land and air calls. However, incidents of misunderstanding between the two parties on the ground and air communication occur from time to time. Among them, the problem of the ground and air communication caused by the pilot's command repeat error is the most common. Therefore, strengthening the semantic automation consistency measurement of the ground-air Chinese communication command has great and urgent practical significance for the pilot to correctly execute the command issued by the controller. First, use land-air call recordings to construct a repetitive land-air call corpus, and use word vectors to represent the words in the land-air call instructions; then use two RNN or LSTM-RNN models to describe the air traffic control instructions in each group of land-air calls. Repeat the instructions with the pilot and generate the semantic vectors of the two sentences; third, use the cosine similarity to measure the similarity of the two semantic vectors; finally, use common classifiers to determine the semantic consistency of the two parties in the land-air Chinese communication. The paper compares the algorithm models of two word vector generation methods, one-hot and word2vec in Chinese communication. Experimental results show that it is feasible to use RNN and LSTM-RNN models to describe the semantic vector of land-air call instructions and use classification methods to determine the semantic consistency of land-air calls, and the LSTM-RNN model is more suitable for describing land-air calls than the RNN model The semantics of the call Chinese language command has achieved a higher accuracy rate in the experiment.