BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:America/New_York BEGIN:DAYLIGHT DTSTART:20260308T030000 TZOFFSETFROM:-0500 TZOFFSETTO:-0400 RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3 TZNAME:EDT END:DAYLIGHT BEGIN:STANDARD DTSTART:20261101T010000 TZOFFSETFROM:-0400 TZOFFSETTO:-0500 RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11 TZNAME:EST END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20260909T042343Z UID:C683F8C2-23C6-4F40-96BC-7E06F0D30CC1 DTSTART;TZID=America/New_York:20260909T153000 DTEND;TZID=America/New_York:20260909T160000 DESCRIPTION:Air Traffic Control (ATC) communication is a safety-critical pr ocess governed by strict procedural and phraseological standards\, yet mos t prior applications of Large Language Models (LLMs) in aviation have focu sed on isolated tasks such as speech recognition or static regulatory ques tion answering. This work proposes a three-stage adaptation pipeline for p redictive\, context-aware ATC communication. First\, open-weight LLMs are domain-adapted on FAA/ICAO regulatory documents and aviation question-answ er pairs to establish procedural alignment. Second\, the models are instru ction-tuned on structured ATC dialogue templates to learn sequencing\, rol e consistency\, and standardized phraseology. Third\, the adapted models a re re-tuned using simulation-grounded Digital Twin (DT) traces from Greens boro Airport (KGSO)\, enabling response generation conditioned on evolving aircraft state\, runway configuration\, and operational context.\nWe eval uate open-weight LLMs across these stages using both semantic and operatio nal metrics.\nResults show that high accuracy can be achieved\, with Phi-4 producing the strongest overall performance and simulation grounding subs tantially improving context-aware ATC generation. These findings suggest t hat combining domain adaptation\, phraseology-aware instruction tuning\, a nd DT-guided re-tuning provides a promising foundation for simulation-driv en controller training\, decision support\, and future human-in-the-loop A TC systems.\n\nCo-sponsored by: XDI Lab (www.xdilab.com)\n\nVirtual: https ://events.vtools.ieee.org/m/576384 LOCATION:Virtual: https://events.vtools.ieee.org/m/576384 ORGANIZER:hmoradi@ncat.edu SEQUENCE:42 SUMMARY:Generative AI for Air Traffic Communication Automation - Fine-Tuned Language Models Using Simulation-Grounded and Real-Trace Data URL;VALUE=URI:https://events.vtools.ieee.org/m/576384 X-ALT-DESC:Description: <br /><p>Air Traffic Control (ATC) communication is a safety-critical process governed by strict procedural and phraseologica l standards\, yet most prior applications of Large Language Models (LLMs) in aviation have focused on isolated tasks such as speech recognition or s tatic regulatory question answering. This work proposes a three-stage adap tation pipeline for predictive\, context-aware ATC communication. First\, open-weight LLMs are domain-adapted on FAA/ICAO regulatory documents and a viation question-answer pairs to establish procedural alignment. Second\, the models are instruction-tuned on structured ATC dialogue templates to l earn sequencing\, role consistency\, and standardized phraseology. Third\, the adapted models are re-tuned using simulation-grounded Digital Twin (D T) traces from Greensboro Airport (KGSO)\, enabling response generation co nditioned on evolving aircraft state\, runway configuration\, and operatio nal context.&nbsp\;<br>We evaluate open-weight LLMs across these stages us ing both semantic and operational metrics.&nbsp\;<br>Results show that hig h accuracy can be achieved\, with Phi-4 producing the strongest overall pe rformance and simulation grounding substantially improving context-aware A TC generation. These findings suggest that combining domain adaptation\, p hraseology-aware instruction tuning\, and DT-guided re-tuning provides a p romising foundation for simulation-driven controller training\, decision s upport\, and future human-in-the-loop ATC systems.</p> END:VEVENT END:VCALENDAR