The work of Wittgenstein is apparently characterized by a strong attitude towards an holistic view of language. Even in Tractatus logico-philosophicus - considered as a program for logical atomism - we find reference to Frege's context principle: names have meaning only in the context of a sentence (TLP 3.3). Names do not have meaning in isolation; only when used to refer to something they are meaningful. Wittgenstein subsequent work seems to be a progressive generalization of the context principle. The Fregean principle is quoted as such at the beginning of Philosophical Investigation in order to show that words do not have denotation in isolation, but only when used inside a language game, a social environment of speech and action. If the sentence is the basic move in a language game, like a move in the game of chess, a language (game) itself is taken to be the basic unit in linguistic activity:
«To understand a sentence means to understand a language. To understand a language means to be master of a technique» [Wittgenstein 1953, § 199]
Quine and Davidson interpret the sentence quoted above as an expression of a radical holism. Quine refers to it in World and Object (§16) to stress the radical dependence of any individual sentence from the entire language. Davidson endorses this holistic attitude in his seminal essay "Truth and Meaning" where he says that «we can give the meaning of any sentence (or word) only by giving the meaning of every sentence (and word) in the language. Frege said that only in the context of a sentence does a word have meaning; in the same vein he might have added that only in the context of the language does a sentence (and therefore a word) have meaning» (p.22). Davidson's original theory of meaning gives a picture of language as a fixed set of sentences whose meaning can be systematically given by their truth conditions.
However the extreme holistic viewpoint has a disastrous consequence, clearly denounced by Dummett 1973 in his comment to Quine's work. Assuming a definition of meaning as inferential role, the paradox could be expressed as such: if the meaning of an expression is its inferential role, the meaning is the set of inferences connected with the given expression; however, given that there is no distinction between analytic and synthetic sentences, we cannot define a restricted set of relevant inferences which defines meanings (as e.g. analytical sentences given in the form of meaning postulates). If meanings depends on the entire language and on the entire system of beliefs of a speaker, given that there are no two individuals with the same identical set of beliefs, no two individuals may give the same meaning to an expression. But if two individuals do not give the same meanings to the same expressions, then they can neither agree nor disagree. Hence communication becomes impossible. The apparent paradox is therefore that if meaning is defined holisticaly, .there is no meaning to be used in communication.
Dummett sees clearly that this criticism of holism is a criticism on its consequences, and then it is not a proof of the falsity of semantic holism; however it is an evidence in favour of a methodological stance against holism (Dummett 1993,p.21). Certainly, if holism holds, we could find it difficult to explain language learning and communication. But not being able to explain language learning and communication can be read both as evidence against holism and as evidence against traditional explanations of learning an communication. Dummett follows the first horn of the dilemma; others will follow the second one.