基于GAN的民航陆空通话文本生成方法
Text Generation in Civil Aviation Radiotelephony Communication Using Generative Adversarial Network
DOI: 10.12677/CSA.2018.812208, PDF,  被引量    科研立项经费支持
作者: 邱 意*, 赵子豪, 李 丹, 程方圆：中国民航大学电子信息与自动化学院，天津
关键词: 陆空通话文本；生成式对抗网络；文本生成；Land and Air Call Text； Generative Adversarial Network； Text Generation
摘要: 民航陆空通话是飞行员与管制员进行话音通信的主要方式，是管制员指令发送与飞行员指令回复的信息载体，飞行员正确理解管制员所发出的指令对于民航运输安全有着重要意义。采用深度学习方法对陆空通话进行实时语义校验可以及时发现飞行员错误的复诵内容。考虑到训练一个有效的差错校验网络模型需要大量的文本数据，本文提出一种基于生成对抗网络GAN的陆空通话文本生成方法。首先对现有真实的陆空通话文本进行筛选和分类，并将其转换成one-hot词向量。其次，使用one-hot训练GAN模型，并使用其生成新的陆空通话文本。最后，根据陆空通话语法规则筛选出符合要求的生成文本。本文比较了不同参数选择下网络训练的效果，通过实验验证，使用GAN模型能生成符合民航陆空通话语言规则的文本。
Abstract: Civil aviation radiotelephony communication is the main way for pilots and controllers to conduct voice communication. It is the information carrier that the controller commands to send and respond to the pilot's instructions. Pilots’ understanding the order correctly by controllers is significant for the safety of civil aviation transportation. Using the deep learning method to perform automatic semantic verification in civil aviation radiotelephony communication can timely discover the contents of the pilot's read-back errors. Considering that training an effect automatic semantic verification model needs a lot of text data, this paper proposed a method of text generation in civil aviation radiotelephony communication using Generative Adversarial Network. First, the existing texts in civil aviation radiotelephony communication are classified and converted into one-hot vectors. Second, the new texts in civil aviation radiotelephony communication are generated based on the seq GAN model. And we compare the training effect of network based on different parameters. Finally, the generated texts are removed, which do not meet the grammar rules in civil aviation radiotelephony communication. This paper compared the effects of network training under different parameter selections. Experiment shows that GAN can be used to generate texts that comply with the civil aviation radiotelephony communication’s rules.
文章引用：邱意, 赵子豪, 李丹, 程方圆. 基于GAN的民航陆空通话文本生成方法[J]. 计算机科学与应用, 2018, 8(12): 1870-1877. https://doi.org/10.12677/CSA.2018.812208

参考文献

[1] 王万乐, 赵琦. 空中交通管制员无线电陆空通话[M]. 北京: 北京交通大学出版社, 2016.
[2] 董昱廷. 陆空通话对飞行安全影响的分析[J]. 移动信息, 2016(6): 00162-00163.
[3] Cui, Q. and Li, Y. (2015) The Change Trend and Influencing Factors of Civil Aviation Safety Efficiency: The Case of Chinese Airline Companies. Safety Science, 75, 56-63.
[4] Jia, G.M., Lu, Y.J., Lu, W.B., et al. (2017) Verification Method for Chinese Aviation Radiotelephony Readbacks Based on LSTM-RNN. Electronics Letters, 53, 401-403. [Google Scholar] [CrossRef] 
[5] Goodfellow, I., Pouget-Abadie, J., Mirza, M., et al. (2014) Generative Adversarial Nets. Advances in Neural Information Processing Systems, 2672-2680.
[6] Isola, P., Zhu, J.-Y., Zhou, T.H. and Efros, A.A. (2017) Image-to-Image Translation with Conditional Adversarial Networks. arXiv:1611.07004 [cs.CV]
[7] Taigman, Y., Polyak, A. and Wolf, L. (2017) Unsupervised Cross-Domain Image Generation. arXiv:1611.02200 [cs.CV]
[8] Che, T., Li, Y., Zhang, R., et al. (2017) Maximum-Likelihood Augmented Discrete Generative Adversarial Networks. arXiv:1702.07983 [cs.AI]
[9] Lin, K., Li, D., He, X., et al. (2017) Adversarial Ranking for Language Generation. arXiv:1705.11001 [cs.CL]
[10] Rajeswar, S., Subramanian, S., Dutil, F., et al. (2017) Adversarial Generation of Natural Language. arXiv:1705.10929 [cs.CL]
[11] Kusner, M.J. and Hernández-Lobato, J.M. (2016) GANS for Sequences of Discrete Elements with the Gumbel-softmax Distribution. arXiv:1611.04051 [stat.ML]
[12] Zhang, Y., Gan, Z. and Carin, L. (2016) Generating Text via Adversarial Training.
https://zhegan27.github.io/Papers/textGAN_nips2016_workshop.pdf
[13] Zhang, Y., Gan, Z., Fan, K., et al. (2017) Adversarial Feature Matching for Text Generation. arXiv:1706.03850 [stat.ML]
[14] Hochreiter, S. and Schmidhuber, J. (1997) Long Short-Term Memory. Neural Com-putation, 9, 1735-1780. [Google Scholar] [CrossRef] [PubMed]
[15] Yu, L., Zhang, W., Wang, J., et al. (2016) GAN: Sequence Gen-erative Adversarial Nets with Policy Gradient. arXiv:1609.05473 [cs.LG]
[16] Zhu, Y., Lu, S., Zheng, L., et al. (2018) Texygen: A Benchmarking Platform for Text Generation Models. arXiv:1802.01886 [cs.CL]
[17] Papineni, K., Roukos, S., Ward, T., et al. (2002) Bleu: A Method for Automatic Evalua-tion of Machine Translation. ACL’02 Proceedings of the 40th Annual Meeting on Association for Computational Lin-guistics, Philadelphia, 7-12 July 2002, 311-318.