In the second half of the dialogue the interlocutor is Cratylus. He presses the nature view hard: every name has been set correctly, and there is no such thing as a “false name.” From here Socrates sets about dismantling the nature view he had been leaning toward.
First, convention comes back
An image (εἰκών) must not reproduce everything about its object. Reproduce all of it and you no longer have an image but a second original. So a name, too, need not carry every “correct” letter. An imperfect name can still be a name — and with that step, convention is back in the argument.
The decisive case is the word “hard (σκληρόν).” It contains λ. On the sound-symbolism account, λ means “smoothness” — the opposite. And yet we understand each other. Cratylus says it is “because of habit (ἔθος).” Socrates turns it back:
ἔθος δὲ λέγων οἴει τι διάφορον λέγειν συνθήκης;
When you say “habit,” do you think you are saying something different from convention? (Greek quotations throughout; Japanese and English are L/LAB working translations.)
Numbers are the same. Giving each number a name “like” it is impossible without borrowing the authority of convention. Socrates settles the matter: make names as like their objects as you can, but “this vulgar thing, convention,” must be used as well.
The core — the first name-giver had no model
Cratylus holds that names are the road to knowledge. Socrates’s reply is very simple. The one who first set names had not a single name to consult. If “things can only be learned from names,” that person could not have known the things he was to name.
ἔστιν ἄρα, ὡς ἔοικεν, δυνατὸν μαθεῖν ἄνευ ὀνομάτων τὰ ὄντα
It is possible, then, it seems, to learn the things that are without names. (438e)
And it is better to learn from the truth itself than from an image. A name is only an image of its object, and whether the image is a good one can in the end be judged only by looking at the object. So knowledge sits prior to names.
And then, toward what does not change
At the end of the dialogue Socrates asks, as “something I often dream about”: is there a “beautiful itself,” a “good itself” — each one thing?
αὐτό … τὸ καλὸν οὐ τοιοῦτον ἀεί ἐστιν οἷόν ἐστιν;
Is not the beautiful itself always such as it is? (439d)
If the beautiful slips away even as we speak, we can neither call it “that” nor say it is “such.” And if the knower and the known are both changing without pause, the object becomes something else at the moment of knowing, and knowledge does not come off. So “the beautiful,” “the good,” “each thing that is” are not like a Heraclitean flow. Here Plato steps toward the unchanging Forms.
The dialogue does not resolve
The Cratylus closes without ruling for either side. Whether the correctness of names is natural or conventional is left open. But one thing is decisively blocked: the road that would analyse names to extract the truth about the world. To the still-young Cratylus, Socrates leaves a parting word — inquire bravely and well, and do not accept it easily.
Setting out from the analysis of language, this dialogue arrives at “knowledge is prior to language,” and from there bridges to Plato’s theory of Forms and dialectic. A name mirrors the world, but a name is not the world — 2,300 years later, Saussure and Wittgenstein both return, in changed form, to this question.