📄 Standard and nonstandard lexicon in aviation English: A corpus linguistic study

Summary

The paper looks at a safety problem in pilot–controller radio talk. Some words from ICAO standard phraseology also get used in looser, everyday senses, and the gap between the two can cause misunderstanding. The authors built a small corpus of about 26,421 words from Cockpit Voice Recorder and Quick Access Recorder transcripts. The Philippines' national aviation regulator (CAAP) supplied a few of these, and the rest came from publicly accessible airline transcripts. They searched it for items whose usage departed from the ICAO Radiotelephony Manual. They also ran a convenience-sample survey of pilots and ATCs, adapted from an IATA instrument (Said, 2011), asking in which flight phases such deviations happen. Four items stood out: go ahead, hold short, priority and affirm. All four occur in nonstandard ways now and then, sometimes leading to real confusion on the taxiway. The two groups of respondents disagree on where this happens. The authors frame the variation through World Englishes and Kachru's concentric model. Their practical conclusion is that aviation students need explicit awareness of these sense shifts.

Key points

• The corpus has 26,421 words of CVR/QAR transcripts. CAAP released only three transcripts, with airlines anonymised, because of confidentiality.
• Standard vs nonstandard occurrences:
  • go ahead: 7 standard (87.5%) vs 1 nonstandard.
  • hold short: 6 standard (85.71%) vs 1 nonstandard.
  • priority: its single occurrence was nonstandard.
  • affirm: its single occurrence was nonstandard, realised as "affirmative".
• Standard senses come from the ICAO Radiotelephony Manual. Nonstandard senses come from the surveys and the ICAO Phraseology Reference Guide. Examples:
  • priority used without MAYDAY or PAN-PAN.
  • "affirmative" instead of affirm, since it can be misheard as "negative".
• The ATC survey names Route/En-route Clearance as the most deviation-prone phase (28.57%). Taxi, Landing and Approach Clearances were each rated lowest (3.57%).
• The pilot survey points instead to Takeoff, Altitude, Approach and Landing Clearances (15.79% each). Pilots rate En-route and Taxi as least affected.

Interesting ideas

• I1. The study uses a controlled-language manual as the reference definition of each term and labels every corpus occurrence as on-spec or off-spec, which gives a simple, auditable measure of drift in any regulated register.
• I2. The Asiana 222 case shows that a directive was taken by the wrong aircraft in multi-party, interleaved radio talk, so the failure was about who was addressed, not what the word meant.
• I3. One criterion for nonstandardness is acoustic confusability: "affirmative" sits too close to its opposite "negative", so a polarity word gets penalised for how it sounds, not for its meaning.
• I4. The Ottawa incident shows that a nonstandard opener ("while we wait") led the pilot to expect a specific next step, a takeoff clearance. Misreading can come from predicted event sequences, not just the words themselves.
• I5. The paper backs its corpus counts with a practitioner survey and finds that pilots and controllers disagree about where problems happen, a reminder to slice errors by situation and by viewpoint.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

• I1 (frames): Speech-act rules could carry a "sanctioned sense" for register-specific phrases. Mova could then flag go ahead when it is read as permission to move rather than permission to speak.
• I2 (frames): Coreference and frame building could resolve the addressee of each imperative from call-sign vocatives. Unresolved or ambiguous addressees should be reported rather than guessed.
• I3 (back): When generating instructions or answers, the realiser could avoid polarity-bearing words with near-homophone antonyms. It would prefer short, contrastive forms, as ICAO does with affirm.
• I4 (world): Event ordering could store expected procedure scripts, such as clearance sequences. Mova could then mark when a text's events depart from the script, explaining why a reader might have expected something else.
• I5 (learn): Evaluation could break down parser and frame errors by discourse situation and by annotator group. This would show where disagreements cluster rather than reporting a single average.