TL;DR: Making claims or demands about/into other people's internal states, rather than about your state or observable external states, predictably ties people in knots—instead: only make claims about your own experience or observables. This lets the other control the copy of them that's in the shared context.^[1]

Non-Violent Communication rapidly grew into a world-spanning movement due to its remarkable ability to avert and dissolve conflict. The NVC book has a whole bunch of examples of communication procedures and practical tips. Here I'll try and convey the core insights more rapidly by focusing on the underlying principles I've extracted.^[2]

Background models

People have self-models. People have models of other people. Conversations form shared context, which syncs with both of those. Some ways of conversing work better than others.

Syncing back an incorrect model from conversational context can disrupt or harm one's self-model, so dissonance between your self-model and others' models of you is a type of prediction error that hits particularly hard.

Common collision in conversational context (Cartoons start here^[3])

Things can get messy from here! Bob and Alice are having a tug of war over a non-load-bearing part of the map–Bob's state–which is distracting them both from resolving the grievance.

This situation is often resolvable by sharing context if there's enough ambient trust, but it sometimes escalates in very harmful ways.

Just talking about yourself is good, actually

2x2 (Purpose x Domain)

NVC makes it harder to make and easier to spot conflict-inducing claims and demands. The core procedure can be summed up as "only use things from this 2x2":

(you're probably going to want to check at lease some of those footnotes for examples)

We've not really covered the lower half yet, but it's pretty straightforward. Making claims or asks about things that are well specified^[12] and verifiable by all parties^[13] is a safe conversational move since these won't be factually contested.

Reader: That was a 2x2 and some text. I was promised cartoons.
plex: ok, fair

Observables (Cartoons #2)

Summary (slightly spaced repetition)

To wrap up:

1. Conflicts often emerge from claims or asks which are directed at things both parties don't agree on. Risky subjects: the other person's internal state, and underspecified or non-verifiable things.
2. Sticking to claims/asks about your own internal state and facts about the world you can both check is remarkably effective at preventing conflict.
3. The language of Feelings^[8]/Needs^[9]/Observations^[10]/Requests^[11] adds guardrails that make it clear to everyone when you're making claims or asks into someone else's internal state, or about things that are vague or unobservable.

Try it out next time you have a doomy feeling about a conversation, and maybe report back how it goes in the comments :)

Appendix: Speculation on sometimes spectacular successes

People tell wild stories of using NVC to switch people aggressively threatening violence into a calm, reasonable mode, or other similarly dramatic effects.

I have a solid guess as to why this happens. Those people are strongly predicting a fight (and hence claims/demands pushed into their self-model). But then the NVC person massively errors the aggressor's prediction by doing a low probability action (communicating non-violently), and they're now out of distribution.

Confused, they look for signs of what kind of conversation they're in, and... there's a person being reasonable and empathetic in front of them? This brings up patterns in them usually associated with a calm and regulated nervous system. That must mean... they're in an open and respectful conversation??^[16] OK, next token let's try going with that and see if it generates less surprisal.

1. ^{^}

    And leaves you a line of retreat

   Also: only make asks about observables or universal/terminal needs. This leaves them room to decide how (and whether) to change their internal state to fulfil your underlying request, rather than pre-defining how to get there.

   yes i totally cheated and made the tl;dr three four paragraphs with footnotes.

2. ^{^}

   Usually it's taught through examples and somewhat more surface level rules than this guide will emphasise. I expect this form to generalize better for LWers.

3. ^{^}

   thank/blame Claude for the diagrams

4. ^{^}

   Professional example: Alice: "I feel disrespected in meetings when I'm interrupted" (interpretation of Bob's state as disrespectful)

   Relationship example: "I feel abandoned when plans fall through" (interpretation of Bob's state as abandoning)

5. ^{^}

   P: Bob was struggling to manage ADHD symptoms and was interrupting to make sure he understood key points before they slipped away.

   R: Bob was dealing with a depressive episode and was afraid of being poor company.

6. ^{^}

   Sometimes your self-model is incorrect and needs updating, but 'in the middle of resolving a different conflict' is rarely a good time for this kind of vulnerable process.

7. ^{^}

   P: "I feel tense and my shoulders tighten when I'm interrupted during presentations" (NVC Feeling, subtype sensation)

   R: "I feel a heaviness in my chest and tears welling up when plans are cancelled" (NVC Feeling, subtype sensation)

8. ^{^}

   NVC True Feelings are internal emotional or sensory states:

   • "I feel sad and heavy in my chest"
   • "I feel excited and energized"
   • "I feel anxious and my stomach is tight"
   • "I feel peaceful and relaxed"
   • "I feel angry and my jaw is clenched"

   NVC False Feelings import claims about things not actually part of your experience, like the other person's state or external things:

   • Abandoned (interpretation that someone left you)
   • Disrespected (interpretation of others' actions)
   • Ignored (interpretation of others' behavior)
   • Manipulated (interpretation of being controlled)
   • Unappreciated (interpretation of others' responses)
   • Misunderstood (interpretation of others' comprehension)
   • Rejected (interpretation of others' actions)
   • Pressured (interpretation of demands)
   • Hurt (assigning blame for your pain, though can be used NVC-compatibly)

   The key is that Feelings describe just your internal emotional and physical experience, while pseudo-feelings include claims on how others are treating you or what they're doing to you.

9. ^{^}

   NVC Needs are ~terminal or human universal needs:

   • Safety
   • Space
   • Harmony
   • Support
   • Order
   • Understanding
   • Rest
   • Purpose
   • Connection
   • Choice
   • Creativity
   • Belonging
   • Growth
   • Trust
   • Learning

   but not ways of fulfilling those needs as these build in a specific way of meeting the need (making a demand into the other's self-model, or an implicit ask for an action as if that's the only way to fulfil your need):

   • Attention from your partner
   • A raise at work
   • Time alone in the morning
   • Regular phone calls
   • Hugs
   • Punctuality
   • Feedback on your work
   • A clean house
   • Response to emails
   • Solutions to problems
   • Agreement from others
   • Peace and quiet after 9pm
   • Professional respect
   • Validation of your feelings
   • Recognition for your efforts
10. ^{^}

    NVC Observations are specific, factual and measurable:

    • "You arrived at 9:20am when we agreed to meet at 9:00am"
    • "This is the third time you've messaged me today"
    • "You looked away and began typing while I was speaking"
    • "There are five unwashed dishes in the sink"
    • "You said 'I don't want to discuss this'"

    Judgments, evaluations, generalizations are not Observations:

    • "You're always late"
    • "You're being needy"
    • "You're not listening to me"
    • "The kitchen is a mess"
    • "You're shutting down the conversation"

    NVC says Observations could be verified by a recording, they state what happened without adding meaning, labels, or patterns to it. This is sufficient but not always necessary, as for some people more subtle things might actually be observable and not under dispute.

11. ^{^}

    NVC Requests are clear, actionable, and outcome focused:

    • "Would you be willing to tell me what you heard me say?"
    • "Would you stop talking when I'm presenting?"
    • "Would you let me know by 3pm tomorrow if you can't make it?"
    • "Would you tell me why you disagree?"
    • "Would you stop scheduling meetings during my lunch break?"

    But not unclear requests, as the status of unactionable/ill defined requests is often disputed, or demands, or asks of internal state, which don't let the receiver have sovereignty over how and whether the request is fulfilled, e.g.

    • "Tell me what you heard me say"
    • "You need to be more professional"
    • "Just listen better"
    • "You have to change your attitude"
    • "I need you to care more"
    • "Be more respectful"

    A demand makes no allowance to not go along with the ask without rejecting the whole demand-statement, an ask explicitly leaves the option for the person to accept the ask-containing statement while turning down the desired outcome.

    NVC also suggests using Positive (do) rather than Negative (don't) requests where possible; this seems helpful but is not always practical.

12. ^{^}

    If a claim or the result of an ask is underspecified in ways that might cause divergence, that's an opening for conflict on whether those conditions are fulfilled. "Is Bob lazy?" Depends what you mean by lazy. Did Bob work less than 2 hours per day last week? That leaves less to the ear of a listener. Similarly, hyperbole like "Bob never works" causes failures, as the literal interpretation is ~always false, and that means you're always running on interpretations.

13. ^{^}

    If a claim or the result of an ask is not observable by all parties, there is an opening for conflict on whether those things happened. Try and cash out your unobservables in observables. NVC says "things that a recording device would see", but this can be adapted to things which all parties genuinely see and don't dispute. Asking Bob to step up and do his share doesn't give clearly defined conditions on whether he's succeeded or not. It might or might not help, but it doesn't give the kind of clarity which will settle the matter and give your future selves criteria to check against.

14. ^{^}

    P: Alice says she's doing most of the project work, while Bob says it's even.

    R: Alice sees their different spending styles as signs they're growing incompatible, while Bob views it as a normal adjustment period.

15. ^{^}

    P: Instead of debating unmeasurable effort, they can look at story points completed and logged hours from last sprint as observable markers of contribution.

    R: They look at their joint savings and proportional spending over time to get ground truth on financial changes.

16. ^{^}

    I'd bet at decent odds that this general effect (knocking people out of distribution, then confidently providing an alternate context) lets some people do things like

    The Hell’s Angels, notorious for crashing hippie parties and then fucking everyone up, came to Kesey’s compound, and Kesey…somehow socially hacked them. They ended up behaving like perfect gentlemen the whole time, then left promising to play support for Kesey if he ever needed it.

    - SSC Book Review: The Electric Kool-Aid Acid Test