Williams (2000)

Davidson's radical interpreter takes an observer perspective. Occassion sentences play 
an evidential role. What matters for speaker and hearer is mutual recognition.
For Wittgeinstein understanding requires participants, who publicly
manifest their understanding to constitute a subject's understanding.
It can be summarised as an I-thou relation versus an I-we relation.

Davidson talks about interpreting natives from scratch, whereas Wittgenstein talks about
the learning situation as a starting point, a language game in which linguistic
moves are obvious. The learning situation is asymmetric, which is what makes normativity
possible, instead of having merely physical reactions.

Davidson claims in the Epitaphs paper that language is neither stable nor shared.
The only thing that would be required is for each utterance to be interpretable.
Because of mistakes and malaproprisms new "passing theories" are continually needed.
But how can there still be normativity, if all mistakes can be accommodated in 
interpretation?

Davidson's candidate norm is the intention of the speaker. When an interpretation
matches the speaker's intention, then it is correct. This is problematic for two
reasons. The first objection is that it is circular: intentions are linguistic,
because beliefs and meanings are interdependent in Davidson's theory.
The second is about the instability of language in Davidson's account. With
each utterance and every change of belief a new passing theory is created,
a new language. This makes it difficult to have intentions that anticipate the
future. An intention will have no connection to a later utterance.

Davidson has a narrow notion of language. He isolates the language part from language
games. Given charity, the language part could then be interpreted. However, if charity
is seen as agreement in judgment we get understanding without needing interpretation.
Interpretation is an idle wheel on which nothing turns, according to Wittgenstein.

Williams argues that Davidson's claim that interpretation can go all the way down is an
illusion. We need a bedrock of shared judgments, and confirmity of behavior.