Aha ...

I've just spotted a difficulty, which I need to scope out:

'I can talk' looks empirical, as does 'You can talk'. For Moorean reasons, they may be slightly different.

Of course, 'You can talk' looks like 'You can walk'. But there are empirical tests which we can agree on for 'You can walk' (without which the statement would begin to lose its meaning ... ).

But while there are some similar empirical tests for 'You can talk', they aren't relevant to the fundamental case that I want to depend on. We might check whether a recovering stroke victim can talk, but we can't 'check' whether 'You can talk' is in general true: in the second sense, there is a normative factor associated with the radical translation argument. I decide (for me) that you can talk - even if only tacitly, by talking to you.

We may, of course decide what 'counts' as 'walking' as well, of course. (And we can probably completely articulate this within general tacit constraints).

But we can still talk even if we can't agree on what counts as walking ...