Category Qualifications (w/ exercises)

by elriggs7 min read15th Sep 201922 comments


Philosophy of LanguagePublic Discourse

This is the second post in the Arguing Well sequence. This post is influenced by A Human's Guide to Words and Against Lie Inflation.


In the last post, we discussed a common problem in arguments that Prove Too Much. In this post, we’ll generalize that problem to help determine useful categories. But before we go on, what’s wrong with these arguments?

Ex. 1 [Stolen from slatestarcodex]

“A few months ago, a friend confessed that she had abused her boyfriend. I was shocked, because this friend is one of the kindest and gentlest people I know. I probed for details. She told me that sometimes she needed her boyfriend to do some favor for her, and he wouldn’t, so she would cry – not as an attempt to manipulate him, just because she was sad. She counted this as abuse, because her definition of “abuse” is “something that makes your partner feel bad about setting boundaries”. And when she cried, that made her boyfriend feel guilty about his boundary that he wasn’t going to do the favor.”

By this definition of “abuse”, a majority of people are “abusive”. It would be better to reserve that word for a smaller group of people who are intentionally manipulating people.

Ex. 2: I also had a friend that was “one of the nicest people everyone knows, wow she listens so well!”. She admitted that she was actually “selfish and manipulative” because she did nice things to people so they’d like her.

I honestly wish everyone was as “selfish and manipulative” as this girl; however, it makes those two words nearly useless. It would be better to reserve those words for people who create win-lose situations (You give, I take) as opposed to win-win situations (Oh wow, you make me feel important. I want to be your friend).

General Frame

What is the general frame of the problem in the two scenarios? You have 2 minutes.

The claim relies on a “bad definition” of a word. It’s “bad” because my expectations weren’t met. If you say you’re selfish, then I expect that you create win-lose situations, but if all you’re actually a really nice person, then you misled me.

In another way, I can say if you meet the qualifications a,b,c, then you are a member of that category. The problem arises when you have the “wrong” qualifications, as in, the person you were talking to was expecting different qualifications.

X meets qualifications a,b,c for [word]: -> X is [word]

Now that you have a new frame to fit everything in, let’s dive in to a couple curveballs.

Ex. 3: If God was all-powerful, could he make a rock so big that he couldn’t lift it?

This is about the qualifications of the word “all-powerful”, and it’s implying that one of those qualifications is “can create a situation that unqualifies itself as all-powerful”. You could define all-powerful that way; however, I (and most other people) are expecting a definition that means “has a lot of power/abilities” like miracles, time travel, etc and not something paradoxical.

Ex. 4: Is a hotdog a sandwich?

Considering just preserving expectations, if someone asked me to make them a sandwich, and I went to hand them either (A) Ham sandwich, (B) Tomato w/ mayo sandwich, or (C ) Hotdog, which one would most surprise them?

Ex. 5: If a tree falls in the woods, and no one is around to hear it, would it make a sound?

What qualifies as a sound? If we agree that it’s vibrational waves between 20Hz-20kHz, then it made a sound. If we agree that there has to be someone to hear it, then it didn’t make a sound. Since the purpose is communication/ expectation preservation, we can just agree on a set of qualifications, solve the philosophical problem, and move on.

Ideal Algorithm

What algorithm were you running? What’s an ideal algorithm for correcting these types of arguments? You have 3 minutes (The previous examples should fit in your algorithm)

1. What is the key word?

2. What are their qualifications for that word?

3. What are the desired qualifications given the context?

4. If (2) and (3) disagree, then argue about which set of qualifications provide clearer communication given the context.

Running this algorithm on the previous examples is easy, except for the “tree falling in the woods”. The key word is “sound”, but there is no “(3) desired qualification” for sound in this case. If there is no agreed qualification, then what’s needed is to agree on a qualification. Updating

  1. What is the key word?
  2. What are their qualifications for that word?
  3. What are the desired qualifications given the context?
  4. If 3 doesn’t exist, then agree on a set of qualifications.
  5. Else if (2) and (3) disagree, then argue about which set of qualifications provide clearer communication given the context.

What are the differences/similarities/relationships between this and Proving Too Much? You have 3 minutes

My framing of Proving Too Much is a subset of this (I did kind of gave that away in the intro). It’s about the category of 100% Truth/accurate predictions/ map-territory correspondances in the context of wanting to find actually true things that reflect reality. I expect a qualification to lead to only true claims; however, if it implies false or inconsistent claims, then my expectations are violated and that qualification is wrong.

Hard exercise: The category of 100% true reasons to believe things should have no members. How would you construct a category for 0-100% beliefs that is actually more useful?

With this algorithm down, let’s tackle a few more problems.

Final Problem Set

Ex. 6: That salad with cucumbers on it should be called a fruit salad, because cucumbers are botanically a fruit

The key word is “fruit salad”. Most people who order a fruit salad at a restaurant and get cucumbers as their “fruit” would not be happy because their expectations were violated. As the saying goes, the customer is always right.

Ex. 7: Christianity is true to me like Islam is true to someone else.

If I told someone that I made them soup and put a bowl of cereal in front of them, then they might laugh or be disappointed. Both implies a violation of expectation.

Oh, I don’t mean true in that way, but I am talking about something close, let’s call it “reflects reality” instead of truth. Like if Christianity reflects reality, then if I built a time machine, I should be able to go back in time, see Jesus die, be buried, and rise again. Then the bible reflects reality. If Islam reflects reality, then I should be able to go back in time and see Jesus ascend to heaven, but not die or resurrect. Then the Quran reflects reality. So the Bible and the Quran can’t both reflect reality since they’re predicting I would see two different things. Does what I’m trying to say make sense?

Ex. 8:

"Let's run further than we ran yesterday"
"You mean *farther"

Using the more grammatically correct word doesn't change our expectations. In either case, I'd expect to run a greater distance than yesterday.

This is interesting because it generalizes to all grammar corrections that don't change expectations when the context is communication. If the context is instead signalling competence (like a resume), then it would be important.

Ex. 9: Is cereal soup?

This one is interesting because it’s saying the qualifications for the word “true” are different than in Proving Too Much. Instead of being abrasive and claiming the word “true” as mine and can only mean one thing, I can instead just Taboo the word and replace it with its meaning.

Ex. 10: Is water wet?

If someone told me the pool water is wet, I'd think they were saying something trivially true to be silly. If they told me, in all seriousness, the water isn't wet, then I might think the water is fake/an illusion or that generally something is wrong with the water. So in the context of two normal people communicating, water is expected to be described as wet. (If the context is chemistry research papers, there may be a different answer)


One of the purposes of arguing well is clear communication. When talking to someone else (or yourself!), knowing what key words means to each person aids in understanding each other and helps in avoiding confusion.

In the next post, we’ll be discussing false dilemmas, how they arise, and how to deal with them. What's wrong with the classic example:

"You're either with me, or against me!"

[Feel free to comment your answer to the hard question and if you got different answers/ generalizations/ algorithms than I did. Same if you feel like you hit on something interesting or that there's a concept I missed. Adding your own examples with the Spoiler tag >! is encouraged]


22 comments, sorted by Highlighting new comments since Today at 10:57 PM
New Comment

Sometimes there are genuinely progress otbe made changing those expectations. For example say that the "abuse" angle happened on a slave-owner context. If the situation is standard slave-owning with no neglects would the conclusion to be to not apply "abuse" to slavery? If that context can be wrong about the appropriateness of their norms how can we be sure that our relationship norms are optimal?

I happen to think that "manipulative" is not inherently bad and the girl used the word correctly. In general the perspective that words have literal meanings is really downplayed. At the limit where words are only their expectations "literally" starts to mean "figuratively". The kind of reasoning that goes "most real numbers are transendental numbers" works because words have intensions and not just a fuzzy cloud of associations.

I could see it very plausible that someone would say "Hey let's ran for two hours unlike the one hour we ran yesterday" and if he doesn't want to spesify the amount he might express it with "further" and not mean "farther". The reply is not an error correction but rather a needed disambiguation.

The viewpoint of connecting words to their qualifications seems powerful. I just think that it should direct the discussion to the qualifications rather than enforce some magically all agreed upon qualication standard.

Thanks for presenting your different answers as encouraged!

The viewpoint of connecting words to their qualifications seems powerful. I just think that it should direct the discussion to the qualifications rather than enforce some magically all agreed upon qualication standard.

Discussing what our different qualifications for a key word is very helpful for clear communication. Once those are explicit, you still have to agree on a set of qualifications. The standard I'm arguing is "words shouldn't be misleading". This is slightly difficult since words mean different things to different people in different contexts, so the standard adapts to the situation, and the words used should be the words that cause the least amount of miscommunication.

Regarding the running example, that might be true. I didn't give a lot of context in the exercises, so they could be interpreted in different ways. I intended this one to be purely grammar correction with no clarity added.

I happen to think that "manipulative" is not inherently bad and the girl used the word correctly. In general the perspective that words have literal meanings is really downplayed. At the limit where words are only their expectations "literally" starts to mean "figuratively". The kind of reasoning that goes "most real numbers are transendental numbers" works because words have intensions and not just a fuzzy cloud of associations.

The perspective is downplayed in this post? in every day life?

I'm confused on the expectation part. If I view words as "only their expectations", you think that's bad right? Also, if I view "real numbers" and "transcendental numbers" as only their expectations, can't I still tell a mathematician "most real numbers are transcendental numbers" and they'll know what I mean?

If the situation is standard slave-owning with no neglects

What's no neglects? Regardless, I think this point was important/interesting. You're saying that we should change the meaning of words because they have instrumental value (right?). In a world where bad things are not considered "abuse", we would like to change everyone's minds so that they think their bad activity counts as abuse too. Is this what you're saying?

Well "Proving too much" form: We don't want to hit all slave owners so it would be more handy if the term "abuser" only referred to a small group of people. A related point of view could be that someone that tortures their slaves is a abuser but someone that beats their slaves to bruises in keeping of discipline is just correct normal everyday housekeeping.

I don't mean we should start forming defintions with instrumental goals in mind. The term "abuse" is loaded with attitudes and the text was close to reading "abuse would imply I should oppose it and I do not oppose this behaviour therefore it is not abuse". The default mode of operations comes with a lot of unpercieved instrumentality baked in. If you notice it you can decontruct it and do a more deliberate/explicit decision. It is very different to say "That is not abuse, that is microagression" than "That is not abuse, that is okay".

Literal meaning is downplayed by the post. Instead of answering "what is a fruit salad?" we just bypass by referring that a expectation would be violated in a certain circumstance. With different definitions one could argue either a)"Just because you have a salad made of fruit doesn't make it a fruit salad" or b) "Fruit salad is a salad made of fruit". Some persons might worry that accepting A introduces an inconsistency. For example applying the word "terrorist" to someone that uses threat of violence for poliical ends could definitionally apply to well liked groups. Using the standard "well I would not have expected for terrorist to mean that person" to reject such labelings would radicallly change what retoric like "we oppose terrorists" means (it starts to means something close to "we hate your enemies" instead of a principled stance against spesific tactics).

With the mathmatican example I was more going for the behaviour of "most". Someone that doesn't know it it can mean "if I think up a random real numbers very few of them are non-trancendental". However for a mathematician "most" can have the "almost all" meaning of "all but a set of measure 0" which makes for a claim that can be exactly proven. And the proof is not up to interpretation. For example "most real numbers are not between 5.4 and 5.5" would be false as read by a mathematician but probably interpreted to be true by a more lay person. A mathematician doesn't have the problem of making a judgement call what is or is not "most". The contrasting approach would be to have number grouping A and number grouping B. "A has numbers like 3, 5.46 and 7.222 and B has numbers like pi, e and e^pi. Is the overlap in members of A and B signifcant or insignificant?" This kind of framing doesn't allow for mathematics to have teeth. "well is 6.13 in A? I would probably expect it to be but it could fail to be in it". If you have a intension of "A is real numbers" then you can go "6.13 is real so it is in A". If you have to be suspicious whether the foundation of the word applies to this particular situation it's more like the grouping approach rather than the set approach.

This post is arguing that you individually should use words in order to more effectively communicate to others. It also provides a frame to better understand other's use of words, and also how to convince someone else to use a word differently (because they're being misleading to their audience)

When I think of a word to convey qualifications [a,b,c], I should think about how the person I'm talking to in that moment interprets that word. Is it also [a,b,c]? If I don't think so, I preemptively explain the qualifications. If I do think so, I should save time and just say the word, but also, always notice when the other person "acts funny" or reacts not as expected them too. That points to a miscommunication of qualifications or a misunderstanding of the other person. Either option is good to clarify.

Similarly, when someone else uses a word, I can think "What qualifications [a,b,c...] did they mean?" instead of "They used word X, they must mean qualifications [a,k,z]"

I think this satisfies the mathematician example since I can easily use predefined ideas/jargon to clearly communicate to them, and use expanded versions of those jargon/definitions if I was talking to a layperson.

The abuse example, I think we both agree that being aware of [which qualifications each person in the conversation has in mind] is a good thing, right?

I think the fruit salad terrorist example shows where we disagree. To be clear, which of the following do you disagree with?

1). If I offered "fruit salad" at my restaurant (and really gave cucumber salad), I would have misled my customers.

2). In the reverse, if I ordered "fruit salad" and actually got a cucumber salad, I would complain and argue for the menu to be changed because it's misleading.

3). If I used the word "terrorist" [qualifications: someone that uses threat of violence for political reasons] to describe a well liked group, then my audience might assume I mean "well liked group is a [group that has caused literal injury/death for a political aim, and we should dissolve the group/hate them/etc]".

4). Bob describes a well-liked group (that has only threatened violence politically, but never bombed a church), then I might argue that they're trying to be misleading and we'll all just assume Bob's biased and he's just attacking his out-group.

5). Bob says "Look at these various violence acts that [well-liked group] has committed for their political agenda. I think this qualifies as terrorism and should be taken seriously". Then I'd be more sympathetic towards Bob.

In abuse it is very important to be aware which qualifications are in fact used and they in fact differ and their differences are important. However declaring one set of qualifications as "correct" goes into actual opinion rather than just clearing definitions.

I most disagree with 4) instead fo assuming that a mischaracterization is taking place it would be more apporiate to think the person actually cares about threats. In a real life example when I am playing overwatch and somebody says "pick reinhart or I throw" I might pick reinhart but I will file a report for "inactivity". The threat of throwing not carried out is on the same severity level as actually throwing. I get there might be genuine difference in opinion whether it is as bad but to me it is.

Heinouness of benefitting from violence without comitting it. Consider workplace sexual harassment. If you in a position to fire a person and ask for a sexual favour you don't need to make an explicit ultimatum for a person to be genuinely concerned that they will be fired if they refuse. "It was just talk" isn't going to fly. Further more illegal threats. If you credibly say you are going to beat somebody that is an offence even if you do not lay a finger on them. It has to be credible thought. If it is absurd then it can be counted as just talk. In the same way if a polician starts to talk about how somebody should be hurt that can credibly be construed to be a suggestion/order to commit violence. Hence "incitement against a group of people" being something you can be quilty of. And if you call in a bomb threat with no actual bomb there is a chance it will not be written off as a joke.

There is a genuine discussion how to respond to and whether to utilise credible threats. Assuming what the correct stance should be is avoiding that discussion instead of having it.

With fruit salads I don't know whether I meanigfully disagree. Like if somebody assumes my gender wrong and I don't correct them have I mislead them? And I think there is room that a opinionated cook could try to sell that cucumber is now a fruit. In fashion if you are low status and don't do as the fashion says you are violating the norms but if you are high status and do something original you are a trend setter and define what the fashion is. Thus "those shoes do not go with those pants" is a similar kind of claim to "cucumber doesn't go into fruit salad". If I stylise someone and their close ones decry them as ugly have I lied that I stylised them? If the restuarant list "champagne" as one of drink options but it is not produced in the correct area, it that misrepresentation or does the saving grace that average customer gets the idea what kind of drink it is effectively save it from being blameworthy?

A high status person violating norms for trendsetting/ counter-signalling is violating expectations for a specific purpose very much like a comedian violates them for laughs. I agree that violating expectations helps achieve certain goals.

But, if my goal is to argue well, communicate well, find more accurate beliefs, then I should be focused on not violating expectations for the sake of clarity.

[Note: I am finding a lot of value in our conversation thread so far, and I appreciate your input. It's really forcing me to figure out when and why and how this concept is useful or when it's not.]

What if the audiences expectations are based on faulty beliefs? In particular some given topic might have a bunch of entrenched assumptions so that there are positions that can't be expressed without violating expectations. In the very limit if the communication doesn't violate expectations then it can't convey information, the Shannon entropy is zero. There are probably multiple kinds of surprise here. The "easy" kind would be if nobody expected anybody to say "the sky is red". The "hard" kind would be if one means the lowest wavelength kind of light with "the sky is blue". Exhausting the easy kind can be done relatively effectively and straighforwardly. But when there are conceptual problems then the hard kind of thing is the main source of progress. If you encounter evidence that can't be handled with your current conceptual palette you must come up with new concepts in order to accomodate reality. Those updates tend to be laboursome but they tend to be the valuable ones.

If you're claiming "Threats should be taken seriously and punished", then we agree. If you're claiming "we should punish groups that threaten violence for political reasons as 'terrorist'", then we might agree, but it's not a big deal and not the point of this post.

If you're claiming that

If 100 random US citizens are told "X is a group of terrorists" and are told to ask what actions the speaker is trying to imply X engages in, the majority of the group will write "only threatens violence for political reasons"

, then we disagree. I predict they will write a mix of "threatens and commits acts of terror", but never "only threatens".

I think you would not be misleading if you said "X is a group of terrorists, but only the kind that threatens violence but hasn't actually injured/murdered people, but that's still bad and I think we should take it much more seriously than we are right now", and that this statement would pass the "100 random people" test above.

If you disagree with my prediction, then that's just a difference in our priors on how other people qualify that word. This isn't the point of the post.

If you disagree with "100 random people" test as being a good test, then this is relevant to the post.

While the policy suggestion is indeed outside the scope of the discussion I feel it woud be important to process it differently. "Groups that threaten violence for political reasons are terrorist" and "We should punish terrorists". Calling someone a terrorist is not itself a punishment (unless again the label triggers unstated mechanisms that are beyond deliberate, concious or official control). In the topic area it is not unheard of to be issues where "terrorist" is a special position that warrants different procedure. There the issues would be "punish as criminals" or "punish as terrorists" (or POW or combatant etc). If we connect the long definition straight to treatment reference to a one word concept is unneccesary.

I was refrring to the threat portion becuase that is the difference that is sometimes included and sometimes not included. "Only threathens" doesn't really occur.

Exhange that is likely to happen or happens frequently is:

A: "This is a group of terrorists."
B: "You lied to me. I did research and group has not killed anyone"

Sure if you give long form this kind of misunderstading doesn't happen that much.But consider this:

A: "This murderer will be held in prison for life"
B: "You lied to me. This guy only killed criminals, that is not murder"

You could avoid this by going

A:"This person committed a lot of murder on criminals but didn't kill any innocents. He will spend his life in prison.

Now what is or is not murder might be beside the point of the communication. But accomodating such a weird conception of the crime is not exactly neutral. In choosing such a phrasing one could be normalising that criminals have a weakened right to life.

I guess the differences are slight as I don't really advocate to only use the definitions or conceptions of words you have but I think there is a risk of being too conceptual network pandering and persons should have some share of having some sensibility in their concepts. In particular I think in this instance "definition of murder" would get a population majority behind it yet people would in similar representative way fail to apply the label to these circumstances. Thus the "meaning of words" is more strongly established / can be emphasised more rather than the ad hoc associations.

In terrorist there is a pattern that when evaluating I/me the threat component tends to be weak but when applying to others it tends to be strong. The issue is whether it establishes a principle or whether it is a summation or overview of the attitudial landscape.

Ex. 3

This reminds me of Russel's paradox.

Using the more grammatically correct word doesn't change our expectations.

Using the grammatically incorrect word violates our expectations of how words are uzed.

This reminds me of Russel's paradox.

Cool! I think they are very similar, too. If you define "all-powerful" in a certain way, you can get inconsistencies just like if you define set theory in a certain way (For any property Y, you can build a set containing only members with property Y), you can get an inconsistency (property Y = "This set contains itself").

Using the grammatically incorrect word violates our expectations of how words are uzed.

Clever counterexample, it made me think.

Interpreting that quote to mean "most people expect everyone to use correct grammar 100% of the time" seems immediately wrong. Do you interpret it differently?

Interpreting that quote to mean "most people expect everyone to use correct grammar 100% of the time" seems immediately wrong. Do you interpret it differently?

Yes. 2 things:

1. Gesturing: a) I expect something if I am surprised if it doesn't happen. I don't expect something if I am surprised if it happens. You could say I am "surprised" when I see errors because they are rare. (I might lose this surprise w.r.t a specific text which is a solid block of errors, if I tried to read such a thing.) To get more specific, some errors are "smaller" or more common, say, using "it's" possessively instead of "its".

b) More to the point: I expect people to use correct grammar and spelling most of the time.

For example, the original post has... 1,480 words in the form of 8,502 characters (according to, starting at "This is" ending at "encouraged"). Whether or not it has any spelling errors, I'd say the error rate is low, however you look at it.

2. It can break flow. This can also be seen as, when people are reading, they have a prediction of what comes next. On a word level, "further/farther" isn't a severe case, but surprise, and having the right amount of it, especially at certain levels, can affect a work greatly. Some authors may intentionally use (technically) incorrect spelling (and sometimes grammar) for the purpose of puns, and similar effects, and reception may vary. When things appear incorrect, particularly a lot of things, it starts to add up, and working out what was (meant to be) said is more difficult.

I think this is a very interesting point, thanks! It’s related to chunking and does indeed impede communication!

Fuor eggsample, eye cold mes up sevral wurds ina row.

Even though you understand the meaning of that sentence, it takes much longer to get the same point across because it forces you to read one word at a time as opposed to normal, chunked reading.

Relatedly, we use words as chunks of qualifications/ideas.

This post is about framing words as qualifications to maximize communication. Correct grammar lends itself to more efficient communication, but it’s not about qualifications.

I thinking making use of previous chunks/shorthand/ideas the other person has is powerful for conveying information. It informs not just what grammar and words to use, but which metaphors, allusions, and stories to use.

[Actually, if someone consistently uses incorrect grammar, you could more effectively communicate by using the same incorrect grammar. The same if they consistently use a word that is clearly wrong like using “indigenous” instead of “ingenious”]

This might be a site thing, but all of the content in the spoiler's blocks are visible to me.

Here too. as long as i hover over the article (not just the black boxes), but not on the sides

I see this too.

However, this doesn’t seem to be a bug, per se, but simply a consequence of how the spoiler blocks are implemented on The CSS is, in fact, set up in such a way as to reveal ALL spoiler blocks when you hover over the article. (Presumably, this is a deliberate design decision, though I can’t speak for the LW team.)

(GreaterWrong implements spoiler blocks differently, so this shouldn’t happen there.)

Working fine here (both on Less Wrong and GreaterWrong). What is your browser/version/platform?

This has happened to myself too(just like how yoav ravid explained it). Both on Chrome and Mozilla.

I thought I might have just implemented them incorrectly, but if you see it correctly, then I don't know.

I messaged the moderators yesterday about it.

What versions of Chrome and Firefox? What platform?

Folks, saying “Chrome” or “Firefox”, with no additional information, is not helpful! When reporting website problems, always specify all of:

  • Browser
  • Browser version
  • Operating system (and version)
  • Type of device (smartphone? tablet? laptop? etc.)

EDIT: However in this case it seems to be intentional design, not a browser bug—see my other comment.

Sorry for not specifying, and thanks for the bug reporting tip!

In this comment, there are multiple spoiler blocks that reveal successively. (first one, then first and second, then ...) and only when you hover directly over the block. In this article, it's all of them whenever you hover over the article.

That’s right; on LessWrong, spoilers in comments are implemented very differently from spoilers in posts. (I don’t know why that’s so; perhaps someone from the LW team can comment on this.)

Was definitely intentional, though I think mostly for relatively bad reasons that I should get around to fixing. I could explain, but I think the original intention was mostly just wrong, so let's treat it as a bug for now.