Why “Most” Isn’t Just a Word—It’s a Mathematical Idea
A Word That Does Math: What Does “Most” Really Mean?
You look out the window and say to your friend, “Most birds are chirping right now.” She nods. But what exactly did “most” add? Did you count every bird? Did you compare the chirping ones to the silent ones? The word feels natural, but it’s actually a clever mathematical shortcut — and it’s been studied by thinkers for over two thousand years.
“Most” is an example of a quantifier. Quantifiers are words or phrases that say how many, how much, or what portion of something have a certain property. They include “all”, “some”, “no”, “at least five”, “infinitely many”, and many more. But for a long time, logicians — people who study the pure rules of reasoning — focused almost entirely on just two. The rest seemed unimportant. Then, in the 20th century, they realized that all these words can be understood as precise mathematical objects, and that this insight could explain how language works, how we reason, and even how fast your brain can decide that “most birds are chirping” is true.
This is the story of generalized quantifiers, the discovery that turned ordinary words into objects of mathematical study.
Aristotle’s Game of All, Some, and None
The first great investigation of quantifiers began around 350 BC with the Greek philosopher Aristotle (384–322 BC). He was fascinated by four expressions that he thought formed the backbone of reasoning: all, no, some, and not all. He studied how these words could be arranged into patterns that always produce true conclusions from true starting points. For example, from “All dogs are mammals” and “All mammals are animals”, you can safely conclude that some dogs are animals. Aristotle also noticed that the truth of these statements depends on how two groups, or sets, relate to each other.
In modern terms, Aristotle was treating quantifiers as relations between sets. “All dogs are mammals” says the set of dogs is contained inside the set of mammals. “Some cats are striped” says the set of cats and the set of striped things overlap — they have at least one common member. Aristotle did not use the language of sets, but the idea is there: quantifier expressions take two arguments that stand for groups of individuals. We now say they are second-order relations — relations between sets, not just between individuals. That was a remarkable insight, though it stayed half-hidden for centuries.
Frege’s Big Leap: Quantifiers Are Second-Level Concepts
The next major step came in the 1870s with the German mathematician and philosopher Gottlob Frege (1848–1925). He invented modern logic, introducing the now-familiar symbols ∀ (“for all”) and ∃ (“there exists”). But Frege also had a deeper idea. He saw that quantifiers are not about individual things; they are about properties of sets.
Imagine you have a collection of toy animals on a rug. The quantifier ∀ looks at a set of animals. It says “yes” to a set only if that set contains absolutely every toy animal on the rug. The quantifier ∃ says “yes” to a set if that set isn’t empty. Frege called these second-level concepts, because they take ordinary concepts (sets) as input and return a truth value. In his view, the universe of all things was fixed — he didn’t yet have the idea of varying the domain, or universe of discourse, which later generations added. Nevertheless, Frege had essentially discovered the notion of a generalized quantifier, though his logic students would not fully appreciate it for another eighty years.
Opening the Quantifier Zoo
In the 1950s, the Polish logician Andrzej Mostowski (1913–1975) made a simple but powerful observation. The truth conditions for ∀ and ∃ are properties of subsets: ∀ is true for a set if that set equals the whole universe, and ∃ is true for a set if that set is non-empty. But many other properties of sets can also serve as quantifiers. For instance, the property “contains at least 5 elements” or “contains infinitely many elements” or “contains an even number of elements”. Mostowski defined a generalized quantifier of type ⟨1⟩ as any rule that, for every universe M of individuals, picks out a collection of subsets of M. Then a formula with that quantifier, applied to a property, is true exactly when the set of things with that property belongs to that collection.
Soon after, the Swedish logician Per Lindström (1936–2009) generalized further. He allowed quantifiers that bind multiple variables in multiple formulas, producing relations between relations. For example, “most” can be seen as a quantifier of type ⟨1,1⟩: it compares two sets, say A (the dogs) and B (the barkers), and says that the intersection A∩B is larger than the part A−B. That is, “Most dogs bark” means: among the dogs, the barking dogs outnumber the non-barking dogs. This fit perfectly with the way linguists were beginning to analyze everyday language.
Why Language Loves Certain Quantifiers
Once researchers looked at real sentences, they noticed striking patterns. Almost all simple determiners — words like “every”, “some”, “no”, “most”, “at least three” — obey two constraints: conservativity and extension.
Conservativity means you only care about the part of the second set that overlaps with the first. “Most dogs bark” is equivalent to “Most dogs are dogs that bark”. The dogs that don’t bark matter; the things that aren’t dogs at all are irrelevant. This is why saying “Most dogs are barking dogs” sounds redundant, not informative — it just makes the hidden structure visible.
Extension means the quantifier doesn’t sneak a peek at things outside the first argument’s set. If I know that “All students are tired” in a classroom, it stays true if a new planet appears elsewhere in the universe, as long as the set of students and the property of being tired don’t change. Together, conservativity and extension mean the quantifier acts as if the noun argument restricts the domain of quantification. That is exactly what the noun in “most dogs” seems to do.
These properties are not logically required — you could easily invent a quantifier that violates them — but every human language studied so far uses determiners that satisfy them. That is a clue that our brains might build quantifiers from a limited set of operations.
Inferences on Autopilot
Another universal feature is monotonicity. A quantifier is right increasing if extending the second set never turns a truth into a falsehood. “Most dogs bark” implies “Most dogs bark or run”, because adding more activities to the barkers can only increase, never shrink, the set of dogs that do something. Similarly, “All dogs bark” is right increasing; “No dogs bark” is right decreasing (shrinking the second set preserves truth).
Linguists noticed that monotone quantifiers are everywhere in language, and psychologists found that they are easier for people to process. In lab experiments, when subjects saw pictures of dots and were asked whether “most dots are yellow”, they answered faster and more accurately than with non-monotone quantifiers like “exactly half”. Even artificial neural networks learn monotone quantifiers more quickly. Monotonicity also powers a special kind of lightning-fast reasoning: you hardly have to think to leap from “Most Americans who know a foreign language speak it at home” to “Most Americans who know a foreign language speak it at home or at work”. The inference just feels obvious, because your brain seems to track the monotonicity pattern without counting words.
Why It Still Matters (Even to You)
Every time you say, “All my friends play soccer,” “Some movies are boring,” or “Most pizzas are delicious,” you are deploying a miniature logical engine. The words you use don’t just name amounts; they define precise relationships between sets. And your brain handles those relationships so effortlessly that you rarely notice you’re doing math.
The study of generalized quantifiers bridges logic, language, and the science of the mind. It shows that the meaning of a simple word like “most” can be pinned down in exact terms, and that the properties that make it useful — conservativity, extension, monotonicity — might be the very properties that make it learnable. When researchers build computer models that learn the meanings of words from examples, they find that quantifiers with these properties get picked up faster, mirroring what human children do.
You don’t need a piece of paper and a pen to make a logical inference; you’re doing it constantly, with words that have been shaped by thousands of years of reasoning. Aristotle started the conversation, Frege gave it a language, and now, every time you look out the window and tell your friend that most birds are chirping, you’re adding your own tiny contribution to a 2,400-year-old puzzle about how thought and language fit together.
Think about it
- If a friend says “Most of my socks are blue,” and you later discover she has exactly six blue socks and six white socks, did she tell the truth? Why does the edge case matter?
- Imagine an alien who can instantly count every object in any situation. Would that alien still need a word like “most”, or could it get by with only exact numbers?
- If you were designing a brand new language, would you include a word for “exactly half of”, or would you leave it out? What might make a community decide one way or the other?
