A framework for evaluating harmful language
About
The language evaluation framework is a guidance document developed by the Kubernetes Naming Working Group . The version of this documentation found on the Inclusive Naming website is a mirror.
It outlines a structured framework for evaluating language and terminology for harm to the community. This enables the community to navigate divisive conversations with a measure of clarity.
The framework was created for an open source technology project. The framework may be applicable to other fields as well.
If you are using the framework for another open source project, it might be helpful to use it in conjunction with the approval workflow the Naming Working Group defined alongside it.
Using the framework
The framework is divided into three sections: first-, second-, and third-order concerns, ranked in order of potential harm to the community.
First-order concerns are language where harm is egregious, overt, and clearly problematic. Second-order concerns are language which is problematic but with a less definite impact. Third-order concerns indicate language that could use improvement but does no demonstrable harm.
Answer all questions for each term evaluated.
When complete, consider questions answered in the affirmative: in general, the more questions answered “yes” or “possibly”, the more likely it is that the language in question needs to be replaced.
If any first-order concerns are a “yes”, replace the language.
If a significant number of second- or third- order concerns are a “yes”, strongly consider replacing the language.
This framework is intentionally non-prescriptive. The intention in this work is to reduce harm for the community; let harm reduction guide your decisions.
First-order concerns
First-order concerns are characterized by:
- Overtness: regardless of its use in the context of code or technology, there is little to no ambiguity outside of technology as to whether the language in question indicates harm
- Identity-specificity: language in question specifically unambiguously identifies a group of people
Is the term overtly racist?
Examples include “master/slave”.
Is the term overtly sexist, transphobic, or pejorative about a gender identity?
Examples do not include “transclusion” of dependencies, or “binary” operators.
Is the term overtly ableist, or pejorative to neurodiverse or disabled people
Examples include performing “sanity checks”.
Is the term overtly homophobic?
Examples do not include “homogenizing” or “homogenous” data.
Second-order concerns
Second-order concerns are characterized by:
- Ambiguity: outside the context of code or technology, language might have connotations related to harmful scenarios like war, militarization, or policing, but the actual etymology of the term is not related to harm of a specific identity
- Lack of specific identity: concerns in this category do not target specific identities, or do so in a non-overt way
Is the term violent?
Examples include “KILL” commands in Unix systems.
Is the term militaristic?
Examples include “marshal/unmarshal”.
Third-order concerns
Third-order concerns are characterized by:
Clarity: is the language in use a metaphor that could be described more precisely using different words?
Anthropomorphism: does language unnecessarily humanize components or processes?
Idiomatic: Is language unclear to someone outside a specific culture?
Is the term evocative instead of descriptive?
Examples include “PetSet” (evocative) versus “StatefulSet” (descriptive).
Is the term ambiguous?
Examples include the use of ABORT/STOP/KILL in Unix-like systems, where they map to specific behaviors, versus general usage in programming languages, where they map to different behaviors or are used interchangeably.
Footnotes
Changes over time
In general, strong democratic societies become more progressive and accepting as time passes. This is a feature, not a bug.
As a result, terms that were once deemed acceptable may, at some future point, be deemed unacceptable.
We recommend:
- Placing a date at the top of any documents/recommendations related to naming, language inclusivity, or harm reduction
- Expecting that some of your work will need re-evaluation at a later date
- Openness to updating language as readers and cultures change
Dealing with trolls
In the handful of months since this work began, both Kubernetes as a whole and WG Naming have dealt with a number of issues and comments from trolls. We anticipate that anyone using this document to guide their own work will receive the same kind of attention.
In Kubernetes we mostly encounter sea lions (concern trolls), who seek to legitimize debate over false concerns in order to use up contributors' energy and time.
We work with our GitHub and other moderation teams to shut down trolling behavior at the source and remove trolling content.
In cases where it’s unclear whether the poster is a legitimate user or a troll, we direct the work back to them: because they’re clearly “legitimately interested” in this topic, we ask them to join us in the WG Naming mailing list, drafting a formal suggestion (attached to an email address and identity we can track) and suggesting replacement terminology. Most trolls do not want to put in the effort.
Rather than be discouraged by trolls, consider it a heartening sign that you are engaged in meaningful work.
Kudos
This work would not have come into shape without referencing the following resources freely available online. We thank the authors of these original documents for helping guide our thoughts on the topic: