The Arguments We Don’t Know We’re Having

Good news that sounded like a lie

Through 2023, economists and officials kept delivering what they thought was good news: inflation was coming down. And it was. The rate had peaked at about 9 percent in 2022, and by the end of 2023 it was around 3.

Many people heard that news and concluded they were being lied to. Their groceries weren’t any cheaper. Their rent wasn’t any lower. So when someone on television said inflation was falling, the obvious explanation was that the person was out of touch or not telling the truth.

Both sides were right. They were using the same word for two different things. To an economist, “inflation” means the rate at which prices are rising. When inflation falls, prices are still going up, just more slowly. To most people, “inflation” means that prices are high. If inflation is going down, prices should be going down.

This is not a guess about what people were thinking. In December 2023, a Civiqs poll asked registered voters what “inflation going down” means. Only 21 percent gave the economists’ answer: prices are still rising, but more slowly. Twenty-six percent said prices are going down, and 41 percent said prices stay the same.¹ The economist Alan Blinder, a former vice chair of the Federal Reserve, summed it up as “the public’s failure to differentiate between the price level and the rate of change in the price level.”¹

I’d put it less harshly, and more generally. The economists didn’t notice they were using a technical meaning. The public didn’t notice there was a technical meaning to notice. Neither side saw that the argument was about a word. Each side just concluded the other was clueless or dishonest.

That is what this publication is about.

The claim

Here is the idea, as plainly as I can put it.

A large share of disagreement is not about the thing itself. It’s about which thing the words point to. And most of the time, nobody notices.

By “the words” I don’t just mean single words. The same thing happens with a phrase (“serious possibility”), a slogan (“Defund the police”), a category (a “common” side effect), a number (“unlimited” data), a symbol (a flag), and once, famously, a missing comma. I’ll use “term” to cover all of them.

The part I care most about is the last part: nobody notices. People don’t usually know they’re arguing about definitions. When people knowingly argue about what a word should mean, at least the disagreement is out in the open. The more common case is quieter. Each person assumes their meaning is the meaning, so they talk past each other, sometimes for years. Because they can’t see the real problem, they explain the friction some other way: the other person must be stupid, stubborn or lying.

This happens at every scale. It happens between partners, in project meetings, between companies and their customers, between doctors and patients, and between political camps. The stakes change, but the mechanism is the same.

The name of the publication has two meanings, and both are deliberate. “Arguing semantics” is the problem: arguing about words without realizing it. It’s also what I’m doing here: arguing for semantics, making the case that meaning deserves more attention than it gets. The subtitle works the same way. Said in frustration, “What are we even talking about?” is the moment you realize you’ve been talking past each other. Said calmly, it’s the question that fixes it.

Clever, right?

Two things the claim is not:

It is not “everyone secretly agrees.” Sometimes you clear up the words and find a real disagreement underneath, about facts or about values. That’s fine. A real disagreement is worth having. The goal is to find it, not to dissolve it.

It is not “words don’t matter.” Sometimes the fight over a word is the real fight. If one side gets to define “racism” or “recession,” it gets something valuable: the stigma or the authority the word carries. Philosophers call these fights “metalinguistic negotiation,” arguments about how a word should be used, and they argue such fights are everywhere.² They’re real arguments, and they deserve to be had openly.

Why words fail us

Words are pointers, not containers

We tend to talk as if words contain their meanings, as if “inflation” or “help” or “done” carries a fixed package from the speaker to the listener. It doesn’t. A word is more like a pointer. It directs the listener to something in their own head, and the listener’s head is not the speaker’s.

This idea is old. In the 1930s, the Polish-American scholar Alfred Korzybski built a whole movement, General Semantics, on the slogan “the map is not the territory.”³ Wittgenstein made a related point in a more famous form: for a large class of cases, “the meaning of a word is its use in the language.”⁴ If meaning is use, and different people use a word differently, then the same word really does mean different things in different mouths.

Compression

Ideas are big and language is small. To talk at all, we compress. We squeeze a detailed view into a short label and trust the listener to unpack it.

The trouble is that each listener unpacks it with their own key. The shorter the label, the more is left to the listener, and the more the unpacked versions differ. That’s why slogans are such a reliable source of talking past each other. A three-word slogan on a sign or a flag stands for a whole set of beliefs, but it can’t say which ones. Each person who reads it fills in the rest.

Not too few words, but too few shared distinctions

My first guess about all this was that English (or pick your language of choice) simply doesn’t have enough words: that we’re stuck using rough labels because the precise ones don’t exist. I no longer think that’s quite right. English has an enormous vocabulary. The shortage is of shared distinctions. We use one word for several different things, and we don’t notice we’re doing it, because inside our own heads the word only ever meant one thing.

“Inflation” already has two meanings: the level of prices and the rate at which they change. Economists have the distinction. Most people don’t. The word is the same either way.

Who notices?

I find it useful to sort these disagreements by one question: who realizes that the words are the problem?

1. Nobody notices

This is the core case, and I think the most common. Both people use the same term, they mean different things by it, and both believe they’re arguing about the thing itself.

This isn’t just a feeling. A good deal of psychology research shows we are built to miss it:

• We underestimate how ambiguous we are. In experiments, people saying ambiguous sentences expected to be understood more often than they actually were. In the researchers’ words, speakers “underestimate the ambiguity of their own utterances.”⁵
• We overestimate how clearly we come across. People sending sarcastic emails expected about 89 percent of their messages to be read the way they meant them. About 63 percent were.⁶
• We start from our own point of view. Listeners first interpret what they hear from their own perspective, and correct for the speaker’s perspective later, if at all.⁷
• Closeness makes it worse, not better. In one study, married couples tried to convey the meaning of ambiguous phrases to their spouse and to a stranger. They were more confident their spouse understood them. They were no more accurate.⁸
• Asking “did you understand?” doesn’t help much. In one study, only about half of the people surveyed understood that “the tumour is progressing” is not good news, even though respondents were confident they understood the terms.⁹

This shows up everywhere once you look for it.

In 1951, US intelligence analysts wrote that a Soviet invasion of Yugoslavia “should be considered a serious possibility.” Sherman Kent, one of the CIA’s senior estimators, later asked his colleagues what odds they’d had in mind when they agreed on that phrase. The answers ranged from 20 percent to 80 percent. Kent himself had meant about 65. A senior State Department reader had assumed the odds were much lower.¹⁰ These were careful professionals who had agreed on the wording together, and they still didn’t notice they meant different things by it.

In medicine, the European Union defines a “common” side effect as one that affects between 1 and 10 percent of people. In one study, patients told that a side effect was “common” guessed, on average, that it affected 34 percent. Patients given the actual figure, 2.5 percent, guessed about 8.¹¹ Nobody in that exchange is arguing. They’re just leaving with very different ideas of the risk, and neither the doctor nor the leaflet nor the patient knows it.

The same thing happens in contracts. In 1960 a federal judge opened his ruling with a sentence that should be better known: “The issue is, what is chicken?” A buyer had ordered chicken expecting young broilers. The seller had shipped older stewing birds, which were also, as far as the seller was concerned, chicken.¹² In 2017 a court opened another case with “For want of a comma, we have this case.” A Maine overtime law without an Oxford comma turned out to be ambiguous, and a dairy ended up settling with its drivers for $5 million.¹³

2. One side notices and uses it

The second case is when someone does see the gap between meanings, and benefits from it.

Sometimes that’s a strategic definition. A word that carries moral weight is valuable. Whoever controls its definition decides who gets the stigma or the approval. In 1938 the philosopher Charles Stevenson called this a “persuasive definition”: giving a familiar word a new meaning while keeping its emotional charge.¹⁴

The 2022 argument over the word “recession” is a clean example. In the US, a common rule of thumb says a recession is two quarters in a row of shrinking GDP. The official referee, the National Bureau of Economic Research, uses a broader definition with no two-quarter rule. In July 2022, a week before the GDP figures were due and with a second negative quarter looking likely, the White House published a post explaining that the two-quarter rule “is neither the official definition nor the way economists evaluate the state of the business cycle.”¹⁵ The argument got heated enough that Wikipedia had to restrict editing on its “Recession” page.

The twist is that the second negative quarter didn’t hold up. Revised government data now shows the economy grew in that quarter.¹⁵ A national argument over what a word means turned out to rest on a number that later changed.

Sometimes it’s deliberate ambiguity. A vague term lets different people read their own meaning into it. In organizations, the communication researcher Eric Eisenberg argued this can be useful, because it “promotes unified diversity”: people who disagree can rally around the same words.¹⁶ Political slogans do this too. So does marketing.

“Defund the police” held together people who wanted to shift some of a police budget to social services and people who wanted to abolish police departments. That breadth helped it spread, and then it broke in public debate. In June 2020, 26 percent of Americans favored “defunding the police,” but 43 percent favored “reallocating police funds to social work and mental health.”¹⁷ Gallup said outright that the ambiguity of “defund” was getting in the way of agreement.¹⁷

On a food package, “natural” has no formal legal definition in the US. The FDA’s long-standing policy only says nothing artificial or synthetic has been added that you wouldn’t expect in the food. In a 2015 Consumer Reports survey, most people said “natural” should mean no artificial ingredients (84 percent), no toxic pesticides (84 percent) and no GMOs (82 percent).¹⁸ The label lets each shopper see what they hope for. And when AT&T sold “unlimited” data plans while slowing down customers after as little as 2 gigabytes, the head of the Federal Trade Commission put it simply: “‘unlimited’ means unlimited.” The company paid $60 million to settle.¹⁹

The motte-and-bailey move belongs here too. Someone asserts a bold claim, and when challenged retreats to a much narrower, easily defended version of the same words, then goes back to the bold one once the challenge has passed.²⁰

3. Everyone notices

The third case is the healthy one. Both people realize they mean different things by a term, and they argue openly about what it should mean, or set it aside and argue about the substance.

That doesn’t mean they’ll agree. But the argument is now about the right thing.

Moving disagreements from the first kind to the third is the whole point of this publication.

What makes it worse

A few forces keep feeding the problem. Each will get its own piece eventually. Here’s a quick map.

Lossy compression. Headlines, slogans, posts and flags are compressed by design, and they’re how most ideas now travel.

Context collapse. The researchers Alice Marwick and danah boyd described how “social media technologies collapse multiple audiences into single contexts.”²¹ A sentence written for one audience, with one shared meaning in mind, is read by every other audience too, each with its own meaning.

Meanings drift. The psychologist Nick Haslam documented how words like “trauma,” “abuse,” “bullying” and “prejudice” have stretched to cover both new kinds of cases and milder ones. He calls it “concept creep.”²² Steven Pinker described a related cycle, the “euphemism treadmill”: a polite new word gradually picks up the stigma of the old one, so it gets replaced in turn.²³ When a word’s meaning moves, people who learned it at different times are using different words without knowing it.

Framing. The word chosen can shape how an issue feels. In a 2014 study, Americans were more likely to say they were “very worried” about “global warming” (15 percent) than about “climate change” (9 percent).²⁴ Framing is real, but it’s sometimes weaker than its reputation. A famous campaign to rename the estate tax the “death tax” barely moved opinion: about 70 percent favored repeal under either name.²⁴ That’s worth keeping in mind. Not every word choice is a magic trick.

What the evidence shows

It’s hard to measure talking past each other directly. But the traces are everywhere.

Change one word, and the answers swing. For decades, a major US survey, the General Social Survey, has asked half its respondents about spending on “welfare” and the other half about spending on “assistance to the poor.” In the mid-1980s, support for more spending was about 39 points higher for “assistance to the poor.”²⁵ By my own tabulation of the 2024 survey data, 33 percent say we spend too little on “welfare,” while 71 percent say we spend too little on “assistance to the poor.” In forty years, the gap has barely moved. One caution: some researchers argue people picture different programs under each label, so part of the gap may be real disagreement about which programs are meant.²⁵ Even so, that’s still a gap in what the terms point to.

Slogans poll differently from the policies inside them. “Defund the police” is the clearest recent case.¹⁷

Ask people what a term means and you get very different answers. In 2018, Gallup asked Americans what they understand “socialism” to mean. The most common answers were equality (23 percent), government ownership or control (17 percent), and benefits and services (10 percent). Twenty-three percent had no opinion.²⁶ In 2023, a USA Today/Ipsos poll asked what “woke” means. Fifty-six percent chose “informed, educated, and aware of social injustices”; 39 percent chose “overly politically correct and policing others’ words.”²⁷ When a country argues about socialism or wokeness, it is having several different arguments under one name.

Partisans increasingly use the same words differently. One team of researchers analyzed tens of millions of YouTube comments on the channels of different US news networks and found the audiences were effectively speaking different languages. As they put it, “what is black lives matter to one can be all lives matter to the other.”²⁸ Another study of a decade of CNN and Fox News captions found the same keywords appearing in increasingly different contexts, peaking in 2020.²⁸

Philosophers have a test. The philosopher David Chalmers proposed a simple way to tell whether a dispute is about words. Ban the disputed term, and ask: “can you state what you are disagreeing about without using (or mentioning)” it?²⁹ If you can, there’s a real disagreement underneath. If you can’t, the dispute was about the word.

Here’s what’s missing. Chalmers says verbal disputes are “familiar and ubiquitous.”²⁹ Others agree. But as far as I can find, nobody has actually measured how much disagreement is semantic, or how often the people involved fail to notice. It’s a claim everybody seems to accept and nobody has counted. That’s one of the things I want to try to do here. More on that below.

It’s also worth stating the counterweight honestly. The couples researcher John Gottman’s institute reports that most of the conflicts couples have, about 69 percent, are “perpetual problems” rooted in real differences in personality and values, not misunderstandings.³⁰ Not every fight is about words. Part of this project is learning to tell the difference.

This is not a new idea

People have been worrying about this for a long time.

General Semantics, Korzybski’s movement, was popularized in the 1940s by S. I. Hayakawa, whose bestselling book Language in Thought and Action offered a memorable rule: “Cow1 is not cow2; Jew1 is not Jew2; politician1 is not politician2, and so on.”³¹ Things that share a name are not the same thing.

In 1946, George Orwell complained that “the word Fascism has now no meaning except in so far as it signifies ‘something not desirable’,” and that with a word like democracy, “not only is there no agreed definition, but the attempt to make one is resisted from all sides.”³²

In 1956, the philosopher W. B. Gallie described “essentially contested concepts”: ideas like art, democracy and social justice whose proper use “inevitably involves endless disputes about their proper uses on the part of their users.”³³ For Gallie, some of these arguments never end because they can’t. That’s not a failure. It’s what those concepts are like.

So what’s new? Mostly speed and scale. Compressed phrases now reach every audience at once, instantly, stripped of the context that once told you what the speaker meant. The old problem has a much bigger stage.

What we can do about it

The good news is that the fixes are mostly simple. They’re just not habits yet. The toolkit will get its own series. Here’s the preview.

Ask “what do you mean by that?” and mean it. Not as a debating trick, but because you genuinely don’t know yet.

Taboo the term. Restate your position without using the word, phrase or slogan in question. This is the same idea as Chalmers’s test. The writer Eliezer Yudkowsky put it well: “the first line of defense is not to define your problematic terms, but to see whether you can think without using those terms at all.”³⁴

Number the meanings. Following Hayakawa, if a word is doing two jobs, give each job its own label, like racism-1 for personal prejudice and racism-2 for unequal outcomes built into systems, and then discuss each one separately. Often you’ll find you agree about one and disagree about the other.³¹

Sort the disagreement. Is it about what a term means, about what’s true, or about what matters? Each needs a different kind of conversation. Evidence can settle a factual disagreement. It can’t settle a disagreement about values, and it can’t settle a definition either.

Restate the other side first. The philosopher Daniel Dennett popularized a set of rules, credited to Anatol Rapoport, for criticizing a view fairly. The first one: re-express the other person’s position “so clearly, vividly, and fairly that your target says, ‘Thanks, I wish I’d thought of putting it that way.’”³⁵ If you can’t do that, you may not yet know what they mean.

At the level of whole societies, there’s real evidence that structured, face-to-face conversation helps. In 2019, an event called America in One Room brought together more than 500 registered voters from across the country for a weekend of moderated discussion with balanced briefing materials. Views moved on both sides, often away from the extremes, and participants grew warmer toward the other party, while a control group didn’t change.³⁶ There’s some evidence those effects fade over time, which is worth knowing. Ireland’s Citizens’ Assembly, a group of randomly chosen citizens who deliberated over several weekends, made recommendations on abortion that preceded a 2018 referendum, which passed with 66 percent of the vote.³⁷

Institutions can help too. Intelligence agencies now attach number ranges to words like “likely.” Software teams agree a written “Definition of Done” before work starts.³⁸ Regulators have started saying what “unlimited,” “eco-friendly” and “healthy” have to mean.¹⁹³⁹⁴⁰ Each is a group deciding, deliberately, that a term will mean one thing.

What this publication will do

Here’s the plan.

Case studies. Most pieces will take one term, whether a word, a phrase, a slogan or a symbol, and work through it in detail:

1. A real moment where it caused a fight.
2. A map of its meanings: who uses which, and since when.
3. Who noticed: nobody, one side, or everyone.
4. The taboo round: each side’s position restated without the term.
5. What’s left: the real disagreement, if there is one.
6. One tool you can use.

The examples will come from every scale: relationships, work, marketing, medicine, law and politics. Every piece will link back here.

The Atlas. Each term we cover gets a living entry on the site: its competing meanings, who uses which, its history, and the evidence. It will grow over time, and it will be the place to send someone mid-argument.

Original research. I want to try to measure what nobody seems to have measured: how much disagreement is really about meaning, and how often the people involved don’t notice. Modern language tools make it possible to look at disagreements at scale, from two-person conversations to public debate. I’ll publish the method, the results and the mistakes openly as I go.

Your examples. If a word, phrase or slogan has started a fight in your life, at home, at work or anywhere else, send it in. Some of the best case studies will come from readers.

So: what are we even talking about? Let’s find out.

Sources

1. Civiqs poll (2023); Stantcheva, Brookings (2024). Full reference ↩ ↩²

2. Plunkett & Sundell, Philosophers’ Imprint (2013). Full reference ↩

3. Korzybski, Science and Sanity (1933). Full reference ↩

4. Wittgenstein, Philosophical Investigations §43 (1953). Full reference ↩

5. Keysar & Henly, Psychological Science (2002). Full reference ↩

6. Kruger, Epley, Parker & Ng, JPSP (2005). Full reference ↩

7. Keysar et al., Psychological Science (2000). Full reference ↩

8. Savitsky et al., JESP (2011). Full reference ↩

9. Chapman et al., Psycho-Oncology (2003). Full reference ↩

10. Kent, “Words of Estimative Probability” (CIA, 1964). Full reference ↩

11. Knapp, Raynor & Berry (2004); EU frequency definitions. Full reference ↩

12. Frigaliment v. B.N.S. (1960). Full reference ↩

13. O’Connor v. Oakhurst Dairy (2017). Full reference ↩

14. Stevenson, “Persuasive Definitions” (1938). Full reference ↩

15. White House CEA (2022); NBER; BEA revised GDP. Full reference ↩ ↩²

16. Eisenberg, “Ambiguity as Strategy” (1984). Full reference ↩

17. Economist/YouGov and Gallup polls (2020). Full reference ↩ ↩² ↩³

18. FDA policy on “natural”; Consumer Reports survey (2015). Full reference ↩

19. FTC v. AT&T (settled 2019). Full reference ↩ ↩²

20. Shackel, “The Vacuity of Postmodernist Methodology” (2005). Full reference ↩

21. Marwick & boyd, New Media & Society (2011). Full reference ↩

22. Haslam, “Concept Creep” (2016). Full reference ↩

23. Pinker, “The Game of the Name” (1994). Full reference ↩

24. Yale/George Mason (2014); Bartels, Brookings (2003). Full reference ↩ ↩²

25. Smith, Public Opinion Quarterly (1987); GSS 2024. Full reference ↩ ↩²

26. Newport, Gallup (2018). Full reference ↩

27. USA Today/Ipsos poll (2023). Full reference ↩

28. KhudaBukhsh et al. (2021); Ding, Horning & Rho (2023). Full reference ↩ ↩²

29. Chalmers, “Verbal Disputes” (2011). Full reference ↩ ↩²

30. The Gottman Institute (not a peer-reviewed study). Full reference ↩

31. Hayakawa, Language in Thought and Action (1949). Full reference ↩ ↩²

32. Orwell, “Politics and the English Language” (1946). Full reference ↩

33. Gallie, “Essentially Contested Concepts” (1956). Full reference ↩

34. Yudkowsky, “Taboo Your Words” (2008). Full reference ↩

35. Dennett, Intuition Pumps (2013). Full reference ↩

36. Fishkin, Siu, Diamond & Bradburn, APSR (2021). Full reference ↩

37. Citizens’ Assembly report on the Eighth Amendment (2017). Full reference ↩

38. The Scrum Guide (2020), “Definition of Done”. Full reference ↩

39. FTC Green Guides; EU Directive 2024/825. Full reference ↩

40. FDA “healthy” final rule (2024). Full reference ↩