Rudolf Carnap, answered from the texts and cited to the page.
The verifiability principle, in its original form, holds that a sentence is meaningful if and only if it is verifiable, and that its meaning just is the method of its verification.1 The historical merit of this formulation was real: it drew attention to the close connection between what a sentence means and the conditions under which we would confirm it, and it served as a tool for exposing the sentences of trans-empirical metaphysics as cognitively empty.1
But taken strictly, the formulation is too narrow — it excludes not only metaphysical pseudo-propositions but also certain scientific sentences that plainly do carry factual meaning.1 The core insight survives the criticism, even where the original formulation does not. If we know what it would be for a given sentence to be found true, we know what it means; and if two sentences have identical conditions for being taken as true, they have the same meaning.2
A sentence has cognitive content only if such a determination is possible at all.2 The decisive refinement is the shift from verification to confirmation. No synthetic sentence admits of definitive and final establishment of truth — universal laws of nature cannot be conclusively verified, only progressively confirmed through their observable consequences.2
I therefore replaced the verification criterion with a requirement of confirmability: a sentence is confirmable if we know under what conditions it would be confirmed or disconfirmed, and testable if we know a method of actually carrying out such a procedure.2 A sentence may be confirmable without being testable — we may know what course of events would confirm it without knowing how to bring about that course of observation.2
This is a modification of the verifiability requirement, not an abandonment of it.3 Among empiricists there is full agreement that some close relation holds between a sentence's meaning and the way in which we may come to confirm it; the question is only how to state that relation precisely.3 The choice between different formulations — stricter or more permissive requirements of confirmability or testability — is itself a practical decision about which language to construct for scientific purposes, not a factual dispute with a single correct answer.4
My own earlier position, influenced by Mach and Russell, accepted a molecular language in which physical laws were interpreted as records of present and remembered past experiences — the most radical positivistic form of the testability requirement.5 I no longer hold that view, though I do not think it was simply false; it was one coherent choice among several, with its own consequences.5
The construction of a language-system is always a matter of choosing rules for purposes, and there is no question of right or wrong in the absolute sense — only the question of whether the resulting system is suitable for the work we intend it to do.4
The connection between meaning and confirmation has sometimes been formulated by the thesis that a sentence is meaningful if and only if it is verifiable, and that its meaning is the method of its verification. The historical merit of this thesis was that it called attention to the close connection between the meaning of a sentence and the way it is confirmed.Testability and Meaning, pp. 4–5
If we knew what it would be for a given sentence to be found true then we would know what its meaning is... a sentence has meaning only if such a determination is possible.Testability and Meaning, pp. 3–4
It is a question of a modification, not of an entire rejection of that requirement. For among empiricists there seems to be full agreement that at least some more or less close relation exists between the meaning of a sentence and the way in which we may come to a verification or at least a confirmation of it.Testability and Meaning, pp. 4–5
There is no question of right or wrong, but only a practical question of convenience or inconvenience of a system form, i.e. of its suitability for certain purposes.Testability and Meaning—Continued, pp. 5–6
Previously, influenced also by Mach and Russell, I too accepted a molecular language... At present I no longer hold this view. But I do not think — as Lewis and Schlick do — that it was false.Testability and Meaning—Continued, pp. 19–20