分类先于发现：为什么给东西起名字是最早的权力
We assume we see first and classify afterwards, but the classificatory frame decides in advance what can be seen at all — and once a classification is written into institutions it acquires the standing of fact. The real question is not whether a classification is true but who pays when it is wrong, and whether it can be revised
I. A Bestiary That Makes You Laugh
In a short essay on invented languages, Borges describes a "certain Chinese encyclopedia" in which animals are divided into categories: those belonging to the Emperor, embalmed ones, tame ones, suckling pigs, sirens, fabulous ones, stray dogs, those included in the present classification, those that tremble as if they were mad, innumerable ones, those drawn with a very fine camelhair brush, et cetera, those that have just broken a water pitcher, and those that from a long way off look like flies.
Foucault quotes this list in the preface to The Order of Things. He reports laughing for a long time, and then being left with a persistent unease. The source of that unease is worth naming: it is not that the list is bizarre, but that it makes our own order of classification look equally arbitrary. A foreign list can unsettle us only if classification was never merely a tool for describing a world that was already there. It also settles what counts as "the same thing," and what may be placed side by side for comparison.
The question this essay takes up is therefore concrete: if classification were only the labelling of pre-existing things, why should a different set of labels disturb the way we see?
II. The Grid Comes Before the Thing
Foucault's answer is a layer that precedes any particular body of knowledge. He calls it the episteme: in any period, a set of unargued rules determines what counts as knowledge, what counts as resemblance, and what may be arranged on the same table. This layer produces no conclusions of its own; it decides which conclusions can be proposed at all.
This is easily misread as "everything is subjective." It says something more specific: observation requires prior coordinates. Without a set of distinctions already in place, what is present is a continuous blur; one cannot even say "this is a thing."
Kuhn gives another version of the same mechanism in The Structure of Scientific Revolutions. Normal science is not the accumulation of every available fact; it is puzzle-solving inside a range marked out by a paradigm. The paradigm decides what counts as a problem, and what counts as an anomaly. Anomalies do not walk into view on their own; they are recognised as anomalies inside a frame — a result that departs from expectation is a crack demanding explanation under one paradigm, and may not be data at all under another.
The naming apparatus has a history of its own. The most durable legacy of Linnaeus's Systema Naturae (1735) is not any single species but binomial nomenclature: a device that lets anyone, anywhere, identify the same organism stably. It did not discover species; it made species stably speakable.
III. Once Classification Enters Institutions, It Becomes Fact
So long as a classification stays in the study, it remains arguable. What makes it hard to move is its entry into institutions.
The Shang Jun Shu places the fixing of names and shares first among the tasks of government: "when names and shares are fixed, the realm is ordered," and "law is the people's life, the root of governance." The "share" in question is name and entitlement: who falls into which category, what belongs to whom, and which transgression draws which penalty. Shang Yang did not ask the population to grasp legal reasoning; he required that the statutes be "clear and easy to understand," so that "the foolish and the wise alike can know them," with judges and legal clerks explaining and enforcing them at every level. The core of the design is not the severity of punishment but the uniformity of classification: once everyone registers everyone else under the same categories, the ambiguous margins left by kinship and custom shrink to a minimum. Classification is no longer an instrument of knowledge; it is an instrument of administration.
Medicine offers an example in the other direction. In The Man Who Mistook His Wife for a Hat, Sacks records a distinguished musician who could parse musical detail but could not recognise his students' faces, and who once mistook his wife for a hat. Give the condition a diagnostic name and colleagues can take over, insurers can process the claim, treatment has a route. The gateway function of naming is real. But the same book shows the other side: once the category is written on the chart, everything else — who he is, what he can still do, what he wants — recedes behind the category. The label is both the net that catches a person and the press that flattens one.
Geertz, borrowing Weber's image in The Interpretation of Cultures, describes human beings as animals suspended in webs of significance they themselves have spun. The web is a shared classification: what counts as rudeness, as courage, as kin. Individuals do not choose categories from outside the web; they grow up inside it — which is the everyday form of classification preceding judgement.
IV. The Counter-Case: Not Every Classification Is Arbitrary
Here the argument must stop, because it has a common over-extension: since classification precedes discovery, all classification is merely an arbitrary arrangement of power. That inference is false, and harmfully so.
When Mendeleev arranged the periodic table in 1869 he left gaps and predicted the properties of elements not yet found; those elements were subsequently discovered, and the predictions held broadly. This shows that one kind of classification is constrained by the phenomena themselves — get the grid wrong and the phenomena will punish you. Common ancestry in biology is the same sort of case: not a convention, but a claim that evidence can support or refute.
The point, then, is not to announce that all classification is constructed, but to separate two kinds:
- Discovery classifications are constrained by phenomena. Their test is predictive power; error is punished by reality. The periodic table and evolutionary trees belong here.
- Governance classifications are constrained by use. They need not predict anything; they need only let an administrative process run. Criminal categories, diagnostic categories, credit grades, and status registers belong here.
Confusing the two leads to two opposite errors. Treating a governance classification as a natural fact disguises an institutional choice as the structure of the world, so that questioning it looks unprofessional. Treating a discovery classification as pure power abandons the judgement that some classifications are simply wrong — and that judgement is precisely what the classified most need.
V. The Test Is Not "Is It True" but "Can It Be Changed"
Once the two are separated, evaluating a governance classification should not ask whether it "reflects reality" but three more workable questions:
- Do the costs and the benefits fall on the same people? If the convenience accrues to the institution and the cost of misclassification to the individual, the classification has no built-in motive to correct itself. Diagnostic labels, risk scores, and status registers are all cases needing a dedicated channel of appeal.
- Is there an exit? Can one be reclassified, appeal, or have the category abolished? A category with an entrance and no exit accumulates error over time, however reasonable it was at the start.
- Is it publicly checkable? A classification that cannot be inspected cannot be argued with, and therefore cannot be corrected. The trouble with secret scoring algorithms and unpublished lists is not that they use categories but that they cut the only feedback loop a classification has.
The limits deserve stating plainly.
First, not every governance classification should be exitable. If infectious-disease reporting, aviation safety defects, or food recalls allowed voluntary opt-out, the category would lose its only function. In those domains, compulsory classification is itself the protection.
Second, anti-classification also injures people. "We have no such category" is one of the most common forms of exoneration in history: an injury that cannot be named cannot be counted, appealed, or compensated. To demand that every category be revocable would revoke the vocabulary of the injured along with it.
Third, stability has value. Legal certainty depends on categories being predictable over time; letting them shift at will imposes costs on those who relied on them.
The conclusion is therefore not "fewer categories" but treating naming as a public matter that has to be interrogated:
Every classification does two things at once: it decides what can be seen, and it decides who bears the cost of being sorted wrongly. The first is an epistemic question; the second is a question of power. Most disputes over classification deadlock precisely because the two are mistaken for one.
Seen this way, naming something is not the end of knowing. It is where knowing and answering both begin.
