Detecting Attribute Information in Notice to Airman

A NOTAM is a notice that contains critical information for all aviation stakeholders and is especially important for flight crews. NOTAMs notify abnormal conditions of National Aviation System components such as a facility, service, procedure, or hazard that is not known far enough in advance to be publicized through other means. It is an essential component of pre-flight planning and briefing. To facilitate communications, NOTAM uses a distinct language with acronyms, abbreviations, and constructs. Going through infrequent, multi-source abbreviated, cryptic NOTAMs becomes tedious and time-consuming. In this work, we work towards segmenting NOTAMs and defining boundaries for each Notam attribute to acquire a better understanding of the NOTAM structure and easier comprehension of the conveyed critical message. This will help the different stakeholders related to air traffic management, such as pilots, crew, dispatchers etc. As NOTAM does not adhere to English grammar rules, the first stage of automation could be to partition NOTAM into attributes as defined by FAA specifications. In this paper, we pre-train a BERT model for NOTAM data. We then use this pre-trained BERT model for segmentation and identification of attributes in input NOTAMs.