Analysis of Controller-Pilot Communication Messages with Natural Language Processing

View Video Presentation: https://doi.org/10.2514/6.2022-3832.vid This study analyzes transcribed controller-pilot communication messages using a Natural Language Processing (NLP) method by exploring similarities and differences between miscommunication and regular communication messages. Transcribed messages are derived from 42 thirty-minute voice communication recordings from 33 sectors across five Air Route Traffic Control Centers. NLP functions such as sentiment analysis, topic recognition, and part-of-speech tagging are used to analyze a large database of communication messages. The initial findings indicate no significant difference in the sentiment scores between the two groups of messages. Miscommunication and regular communication messages are overwhelmingly neutral and trend positive rather than negative. In addition, the main topic modelling produced similar results: a set of main topics was the same for miscommunication and regular communication messages. These findings are in line with the nature of the ATC language, which is well-structured, non-personal, and limited to well-defined phraseology that uses strict communication protocols. Analysis of audio recordings of miscommunications indicates that the volume and pitch of controller’s speech have very little variation, suggesting that controllers’ emotions are stable. This emotional stability is a very important trait in the human-centered world of air traffic control, since controllers must be in full control of their emotions, especially during critical situations.