When Words Create Blame: Reading the AI171 Preliminary Report Through the Lens of Language

Capt. Amit Singh FRAeS 27th Oct 2025

1. Why Language Matters in Safety Investigations

The Preliminary Report on Air India Flight AI171 (VT-ANB) opens with ICAO’s assurance:

“The sole objective of the investigation shall be the prevention of accidents and incidents and not to apportion blame or liability.”

Yet, as decades of safety research show, even when investigators intend neutrality, the language of a report can quietly construct blame.

A landmark study by Heraghty, Dekker & Rae (2018) demonstrated that when the same accident was described in three styles—“crime-like,” “systems,” and “multi-story”—readers proposed drastically different corrective actions. Readers of the crime-style version were five times more likely to recommend punishment or retraining than those who read the systems narrative. Language alone changed the outcome.

2. The Subtle Grammar of the AI171 Narrative

Between 08:08:33 UTC and 08:09:11 UTC, the preliminary report documents a critical 38-second sequence. But the way it narrates those seconds invites interpretation beyond data.

“The aircraft achieved the maximum recorded airspeed of 180 knots IAS at about 08:08:42 UTC and immediately thereafter, the Engine 1 and Engine 2 fuel cut-off switches transitioned from RUN to CUTOFF one after another with a time gap of 01 sec.”

 “The handle was found to be firmly seated in the 5-degree flap position, consistent with a normal takeoff flap setting. The position was also confirmed from the EAFR data. The landing gear lever was in “DOWN” position.”(No EAFR)

The words “immediately” and “about” give the impression of precision while signalling uncertainty. Placed together, they imply a causal chain—that speed started decreasing due to cut-off—though the data source for this critical event is not identified as EAFR.

When the position of the Flap can be verified by EAFR data, the why not source the same data to verify the position of the landing gear. Instead, the image of the landing gear cockpit panel is referenced showing the lever is down. It leads the reader to assume that like Flaps are extended to 5 position, landing gear is extended.

“In the cockpit voice recording, one of the pilots is heard asking the other why did he cut off.”

This is not a verbatim quotation; it is a paraphrase summarised by investigators. By positioning this human dialogue directly after the unsourced technical line, the narrative rhythm builds a moral sequence: cut-off → question → human intent. The phrase pilot error never appears—yet the suggestion is unmistakable.

3. When Data Attribution Disappears

Notice the alternating patterns of evidence:

• “As per the EAFR data, the aircraft crossed V1 and Vr … liftoff at 08:08:39 UTC.”
• (No EAFR reference) “The aircraft achieved the maximum recorded airspeed … fuel cut-off switches transitioned …”
• “As per the EAFR, the Engine 1 fuel cut-off switch transitioned from CUTOFF to RUN at 08:08:52 UTC.”

The middle, unsourced sentence carries the narrative punch—yet lacks attribution. Heraghty et al. found that this pattern—factual → interpretive gap → factual—encourages readers to fill the silence with human causality.

4. Passive Systems, Active Humans

Mechanical phenomena appear in the passive voice:

“The RAT was observed getting deployed …”
“The APU air inlet door began opening …”

Human actions, by contrast, take active verbs:

“The pilot responded that he did not do so.”

As Vesel (2020) notes, this agentive language makes readers assign responsibility to individuals even when identical facts could be described systemically. It is not deception—it is grammar. English demands an agent; someone who “does.” But aviation failures are rarely singular acts.

5. Silence on System Power and Context

The report carefully explains the Enhanced Airborne Flight Recorder (EAFR) architecture:

“The forward EAFR contains an additional power source from the Recorder Independent Power Supply (RIPS) … allowing it to continue to record after power is lost to other systems.”

However, it never clarifies whether RIPS engaged or which electrical buses powered the EAFR during the crucial 08:08:42–08:08:56 sequence. When critical transitions are narrated without confirming data continuity, adverbs like “immediately” or “about” transform from temporal markers into rhetorical glue. The absence of explicit system context leaves room for readers to assume a deliberate cockpit act.

6. What Other Fields Have Learned About Words

This pattern is not unique to aviation. Cross-domain research shows the same linguistic effects:

• Lindhout (2019) found that ambiguous technical language in reports conceals system factors and steers readers toward human error.
• Goddard & Ralph (2020) showed that traffic headlines using the passive “a pedestrian was hit” reduced perceived driver responsibility by 26 percent.
• von Schneidemesser & Caviola (2025) demonstrated that adding system context increased public support for infrastructure solutions from 33 to 49 percent.

The same linguistic mechanics operate in accident reports: when system context fades, the human actor dominates.

7. From Language to Policy

Dekker (2014) warned that “how we frame human error determines what we fix.” If reports describe what the pilot did more vividly than what the system allowed, organisations will fix people, not design.

In the AI171 report, system cues—RAT deployment, auto-start of APU, earlier MEL for core network—are present but relegated to background, while the cockpit exchange occupies the foreground. That editorial ordering, not the evidence, transforms a learning opportunity into a narrative of error.

8. Building a More Neutral Vocabulary

To uphold Annex 13’s spirit, investigators and technical writers could:

• Attribute every factual statement to its data source (“as per EAFR,” “as per ATC replay,” “as per maintenance log”).
• Replace evaluative verbs (failed to, did not, neglected to) with process verbs (attempted, initiated, observed).
• Present parallel timelines—electrical, mechanical, human—rather than one linear story implying causation.
• Mark paraphrased CVR interpretations clearly as investigator summaries, not direct quotes.

Small linguistic choices preserve fairness, protect reputations, and advance systemic learning.

9. Conclusion

The AI171 Preliminary Report is factual and professional. But its phrasing—adverbs of immediacy, selective sourcing, active humans, passive systems—demonstrates how grammar shapes belief. Without ever writing “pilot error,” the narrative suggests it. Recognising this is not criticism—it is evolution. If India’s investigative community embraces linguistic awareness, our reports can truly serve their purpose: to learn, not to judge.

📚 References

• Dekker, S. W. A. (2014). Safety Differently: Human Factors for a New Era. CRC Press.
• Goddard, T., & Ralph, K. (2020). Words matter, even in traffic. Vision Zero Cities Journal. https://medium.com/vision-zero-cities-journal/words-matter-even-in-traffic-c59f5efb80a5
• Heraghty, D., Dekker, S., & Rae, A. (2018). Accident Report Interpretation: The Power of Language. Griffith University Preprint. https://doi.org/10.20944/preprints201808.0530.v1
• Lindhout, P. (2019). Learning from language problem-related accident reports. Safety Science, 120, 92–101. https://doi.org/10.1016/j.ssci.2019.07.008
• Reason, J. (1990). Human Error. Cambridge University Press.
• Vesel, C. (2020). Agentive language in accident investigation. Journal of Chemical Health and Safety, 27(5), 254–261. https://doi.org/10.1021/acs.chas.0c00002
• von Schneidemesser, D., & Caviola, S. (2025). Language patterns in German road traffic collision reporting. Research Institute for Sustainability (RIFS), Potsdam. https://www.rifs-potsdam.de/en/news/use-agent-focused-language-collision-reporting-boosts-support-road-safety-interventions
• Woods, D. D., & Cook, R. I. (2002). Beyond human error: Cognitive systems in complex settings. Safety Science, 40(4-5), 403–423.