Davidson claims that the building block theory is wrong because direct contact between linguistic theory and events, actions and objects described in non-linguistic terms must be made at the level of sentences, not at the level of individual words, and must be made via the concept of truth, rather than that of reference:
Words have no function save as they play a role in sentences: their semantic features are abstracted from the semantic features of sentences, just as the semantic features of sentences are abstracted from their part in helping people achieve goals or realize intentions. If the name `Kilimanjaro' refers to Kilimanjaro, then no doubt there is some relation between English- (or Swahili-) speakers, the word, and the mountain. But it is inconceivable that one should be able to explain this relation without first explaining the role of the word in sentences; and if this is so, there is no chance of explaining reference directly in nonlinguistic terms....
When it comes to interpreting [a Tarski-style theory of truth] as a whole, it is the notion of truth, as applied to closed sentences, which must be connected with human ends and activities.
The thought here comes in two parts. First, the meaning of words is to be explained by their connection with human actions and the beliefs, desires and intentions that motivate them. Second, this means that the connection must be made at the level of sentences. I'll suppose that Davidson's position would allow that the connection could be made at the level of imperatives as well as declaratives, however. Corresponding to truth-conditions for declaratives we will have compliance-conditions for imperatives.
Why sentences? Perhaps the simplest answer is because it is sentences that express thoughts. This is not just something we learn in elementary school; it is shown by the structure of propositional attitude constructions:
In (1), we characterize Harold's belief by a that-clause, and what does the descriptive work in this clause is the sentence ``Russia is in turmoil.'' This is the sentence we use to describe Harold's belief, and it is the sentence we would expect him to use to express it. In (2), the tense-less sentence ``Elwood to close the door'' characterizes Gretchen's desire, and if she is in a suitable position of authority over Elwood, we would expect her to issue a request or a command using these very words.
The close connection between thoughts and sentences makes sentences a natural place to see meaning flowing from thought to language, according principles something like the following:
to issue a command to
X when they want X to do A, then such utterances are complied
with iff X does A.
To elaborate on Davidson's example, suppose we have a corpus of sentences containing the word ``Kilimanjaro,'' which are used to assert various things and request or command various things:
Our linguist discovers the beliefs that motivate sincere speakers to utter such statements as (3), (4) and (5). She discovers the actions that will be deemed to comply with such imperatives as (6), (7), and (8). So there is a direct contact, in the linguistic theory, between these sentences and various intentions, goals and beliefs with various propositional contents (or involving the acceptance of various sentences, or whatever one wants); we list some of the facts that correspond to (3)-(8):
From this, we get the semantic theory, of which we list some of the postulated facts:
Given the semantic facts about sentences, at least as our theorist has postulated them, a certain pattern emerges:
From this, our theorist derives the following.
(S9) is not an additional fact about the language. It is a
way of ``summing up'' a pattern
that emerges in these facts.
The rejected alternative is to suppose that the direct contact between language and the world is made at the level of reference. On this view, (S9) would not be derived from (S3)-(S8), but would be a fact that was part of the explanation of (S3)-(S8). But this would mean that (S9) would have to be based on some principles that link reference to non-linguistic facts in the way that (A) and (B) link sentences to human goals and intentions. What would these principles be?
We might suppose that, say, the words are directly
associated with images, so that ``Kilimanjaro'' was
associated with an image of that mountain. This is the idea
that is Wittgenstein's target, and Davidson also seems to
associate the building-block theory with this
idea. Wittgenstein would
argue that it wouldn't mean anything that the word was
associated with an image of Kilimanjaro if the it didn't
have the right role in the relevant language games, and if
it did have the right role in the language games, then
``Kilimanjaro'' would stand for Kilimanjaro, no matter what
connections there were between that word and images in ones
mind.
So the argument comes to this.
(10) What makes (S9) true is that people use the word ``Kilimanjaro'' when they want to say something about Kilimanjaro, or when they want to request or command that something be done to, at or with Kilimanjaro.
(11) But saying something about Kilimanjaro, or commanding that something be done to, at or with Kilimanjaro, is done by uttering a sentence; it is the utterance of the sentence that has the property of being a statement about or command relating to Kilimanjaro.
(12) But that is to say that the semantic facts about ``Kilimanjaro'' derive from the semantic facts about the sentences of which it is a part.
I think, however, that this argument is wrong, and that Wittgenstein's example of the builder's language game can show us why.