What Does 'It' Mean When There’s Nothing to Point At?
The Pronoun Puzzle
Imagine you hear two sentences: “A man walked in. He sat down.” You instantly know that “he” is the same man. The pronoun looks back to its antecedent — the phrase “a man” — and picks up the same person. This is ordinary anaphora, and it usually works smoothly.
But what about a sentence like this?
Every farmer who owns a donkey feeds it.
There is no single donkey. The phrase “a donkey” does not name a specific animal. If a farmer owns three donkeys, which one does “it” point to? We can’t answer that by looking at the world. Yet the sentence still makes perfect sense — it means that each farmer feeds every donkey they own.
This is the heart of the problem of problematic anaphora. When a pronoun depends on an expression that isn’t about a definite thing, how does the pronoun get its meaning? That question launched a huge debate in philosophy of language starting in the 1980s.
Kamp and Heim’s Variable Files
The problem caught fire when the linguist Irene Heim (born 1954) and the philosopher Hans Kamp (born 1940) created Discourse Representation Theory (DRT). They said that when you hear “a donkey,” your mind does not treat it as a quantifier that says “there exists a donkey.” Instead, it creates a discourse referent — a mental placeholder like an empty file card with a variable (say, x). Later, a pronoun “it” simply points to the same variable.
For a simple story like “A man walked in. He sat down,” DRT says you build a file with a variable x that is a man who walked in, and then you add that x sat down. The whole story is true if there is at least one man for whom both things hold.
Now take the donkey sentence again. DRT says the odd universal force — that every farmer feeds every donkey they own — comes from the way conditionals and words like “every” bind variables. In DRT, the word every can grab the donkey‑variable from inside the phrase “who owns a donkey” and bind it together with the farmer variable. The result works like this: for every pair of a farmer x and a donkey y that x owns, x feeds y. That matches the strong reading.
DRT seemed brilliant. But a problem soon appeared, often called the proportion problem. Imagine we say:
Most women who own a donkey feed it.
Suppose there are ten women. One woman owns ten donkeys and feeds them all. The other nine own one donkey each and don’t feed them. According to DRT’s rule, we are counting farmer‑donkey pairs. That would give ten feeding pairs and nine not‑feeding pairs, so most pairs are feeding — and DRT predicts the sentence is true. But many people feel the sentence should be false: most women (nine out of ten) fail to feed their donkey. DRT’s neat machinery seems to count the wrong things here. The debate is still alive.
Dynamic Semantics: Meanings that Shift
Another approach was developed by Jeroen Groenendijk (born 1949) and Martin Stokhof (born 1950) with their system Dynamic Predicate Logic (DPL). They wanted to keep the insights of DRT but build them into a more traditional logical system. Their core idea is that meaning isn’t just a static truth condition — it’s a potential to change information as a conversation moves along.
In DPL, an existential quantifier like “a man” works like this: it takes all possible input assignments of values to variables, and for each one it resets the value of x to some man. That new assignment then passes as the input to whatever comes next. Conjunction (the word “and”) lets the left half change the assignment that the right half sees. So “A man walked in. He sat down.” ends up meaning: there is an x such that x is a man, x walked in, and x sat down. The quantifier seems to bind the pronoun even though it occurs in a different sentence — without magic, just by letting variables stay “alive” across discourse.
For donkey conditionals, DPL adds a special treatment of “if”: it quantifies over all assignments that satisfy the “if” part, and requires that each such assignment can be extended to one that satisfies the “then” part (with the same x and y). This again gives the strong reading: every donkey‑owning farmer feeds every donkey they own.
But DPL shares a difficulty with DRT. Consider this short story:
A man broke into Sarah’s apartment. Scott believes he came in the window.
The second sentence can mean that Scott has a general belief — something like “a man who broke in came through the window” — without having any particular person in mind. But DPL (like DRT) treats the pronoun as if the quantifier “a man” reaches into Scott’s belief. That forces a reading where Scott has a belief about a specific man. So neither DRT nor early dynamic semantics can easily capture the vague, general belief reading. Many philosophers see this as a serious challenge.
Pronouns as Hidden Descriptions
A different family of answers says that problematically anaphoric pronouns are really definite descriptions in disguise. These are called D‑type theories (the “D” sometimes stands for “description”). One influential version comes from Stephen Neale (born 1958). He argues that anaphoric pronouns are like the word “the” plus descriptive material recovered from the earlier sentence.
So in “If a farmer owns a donkey, he feeds it,” the pronoun “it” on one reading stands for a numberless description — something like “all the donkeys he owns.” Neale uses a special determiner, whe, which means roughly “the (non‑empty) things that F are also G.” That turns the sentence into a claim that every donkey‑owning farmer feeds every donkey they own.
But a puzzle lurks here too. If the pronoun can mean “all the donkeys she owns,” why doesn’t that numberless reading appear in simple discourse anaphora? Consider:
Sarah owns a donkey. She feeds it.
No matter how hard you try, this never means “Sarah feeds every donkey she owns.” It only means she feeds at least one. So the numberless‑description idea predicts a reading that never actually shows up in normal speech. That mismatch makes many philosophers skeptical.
A more recent D‑type theory from Paul Elbourne (born 1971) takes a different path. He says pronouns have the same structure as a definite description but with an unpronounced noun — so “it” is secretly “it donkey.” And he uses situation semantics with a silent “always” to force universal readings in donkey conditionals. But even Elbourne’s theory must explain why sentences like “If a bishop meets a bishop, he blesses him” work, even though “the bishop who meets a bishop” doesn’t pick a unique person. Uniqueness is hard to satisfy when two bishops are indistinguishable. D‑type theorists have to find clever ways around that.
Context‑Sensitive Quantifiers
A fourth approach, the Context Dependent Quantifier (CDQ) theory, was developed by Jeffrey King (building on work by George Wilson). King’s idea is that the pronoun “it” in problematic anaphora is itself a tiny quantifier — a word like “every” or “some” — whose force and the set of things it ranges over are filled in by the surrounding sentences.
In a simple discourse like “A man walked in. He sat down,” the pronoun acts as an existential quantifier: “some man who walked in sat down.” In donkey conditionals, though, the pronoun still has an existential force (“some donkey she owns”) but the conditional’s own semantics plus a familiarity condition (the pronoun is definite) forces the pairing so that every donkey she owns must be fed. So the weak reading of “If a farmer owns a donkey, he feeds it” gets turned into a strong reading by the interplay of context.
King’s theory also predicts that a sentence like “Every woman who owns a donkey feeds it” can have a weak reading — it might mean every woman feeds at least one donkey she owns, not necessarily all of them. And in fact, some donkey sentences seem to work that way:
Every person who had a credit card paid his bill with it.
Nobody thinks this means they used every credit card; just some card. But other donkey sentences feel strongly universal. The evidence is muddy. Some philosophers argue there is a real ambiguity; others think context and world knowledge simply nudge us toward one reading. This is an active research front, and CDQ helps keep the conversation messy and interesting.
Why “It” Still Puzzles Us Today
So what’s the big deal? A tiny word like “it” touches everything from ordinary conversation to the logic that computer programs run on. If we can’t explain exactly how pronouns work, we don’t fully understand how humans track people, objects, and ideas through a story. And if we want to build machines that read or translate naturally, we need a precise theory.
The debate that started with donkey sentences in the 1980s is still simmering. DRT and dynamic semantics gave us powerful new tools for thinking about how meaning flows through discourse. D‑type theories showed that pronouns can behave like hidden descriptions, even when that creates new puzzles. CDQ brought context and quantification together. No single theory has convinced everyone, and each one runs into trouble with examples like general belief readings, indistinguishable bishops, or the correct reading of “most” sentences.
The mystery remains open. Which means that next time you say “If I have a pet, I name it,” you’re using a sentence that philosophy of language still hasn’t completely tamed.
Think about it
- If you say “If I have a pet, I name it,” what does “it” refer to before you even have a pet? Can a word point to something that doesn’t exist yet?
- A computer following strict rules like DRT might count donkey‑farmer pairs and miss the proportion problem. Do you think a machine could ever understand a story the way you do, or is something else needed beyond logical rules?
- Why might it matter whether “Every student who read a book returned it” means they returned every book they read, or just some book? How would the two meanings affect a library fine policy?
