Clear writing is part of human control.
AI answers often bury the decision under jargon, synonyms, long sentences, and detail that nobody requested. A contributor shared a useful intervention: ask the agent to use ASD-STE100 Simplified Technical English, then make more detail available on demand.
These measurements describe vocabulary variety, sentence length, and output length. They do not establish that the generated text complied with ASD-STE100, became easier to understand, preserved the user's intent, or omitted nothing important. Treat them as a lead for replication, not as a Machine Pidgin benchmark.
A language designed for consequential instructions
ASD-STE100 is a controlled natural language for technical documentation. It grew from an airline request in the late 1970s to make English-language maintenance instructions easier to understand for an international workforce. The current release is Issue 9, dated 15 January 2025.
The standard has 53 writing rules in nine sections and a controlled dictionary with approximately 900 approved general words and approximately 1,200 not-approved words with suggested alternatives. Approved words generally have one approved meaning and one part of speech. Writers can still use technical nouns and technical verbs for their subject.
That structure matters more than the raw word count. The aim is predictable language: fewer synonyms, shorter sentence structures, and clearer instructions. It is a design purpose, not independent proof that an AI prompt will produce compliant or safer text.
The prompt is doing more than one thing
From now on, write in ASD-STE100.
Stay concise and to the point.
Prefer short messages.
Do not say everything you know.
I can always ask for details.This asks for both a controlled language and progressive disclosure. It does not, by itself, verify compliance with the standard.The final sentence is the most interesting design choice: “I can always ask for details.” It makes the first answer a starting layer rather than a complete dump of everything the system can produce. That can preserve the human's attention and the right to decide what deserves expansion.
It also creates a confound. The shorter outputs could come from ASD-STE100, from “prefer short messages,” from “do not say everything,” or from their combination. A safety warning, governing exception, uncertainty, or authority limit must not disappear merely because the response is brief.
Turn the instruction into a habit
The contributor uses Espanso, an open-source text expander, to replace one trigger with the full instruction:
:simpleOne shortcut makes the writing constraint easy to apply consistently. Espanso's official guide explains how to create custom matches.What a stronger test should measure
A reproducible follow-up should freeze the tasks, models, settings, baseline prompt, repetitions, and metric definitions. It should separate the ASD-STE100 instruction from the three brevity instructions, verify actual STE compliance, and test comprehension, factual accuracy, missing qualifications, accessibility, and intent preservation—not only shorter text.
That distinction is central to Machine Pidgin. Compact language helps only when it keeps the information and authority the human needs. Less text is not automatically better communication.
← All field notes Discuss or propose a replication → Report a correction →