(a) First, it must be asked why Condition C should hold in the first place. In
the classic theories of Chomsky and Lasnik , Condition C is stipulated.
Reinhart  tried to argue that Condition C derives from a general preference
for binding over 'accidental coreference', but this preference itself does not
follow directly from standard pragmatic reasoning: binding yields the same truth
conditions as 'accidental coreference', and thus one cannot simply argue that
the former yields a more 'specific' (and hence 'preferable') meaning than the
latter. The same observation applies to Safir 2004 (), whose 'Form to
Interpretation Principle' must stipulate that, under c-command, the most
'dependent' form available must be picked to express a given meaning.
Purely pragmatic attempts have either failed to derive the c-command condition
(Bolinger ) or have done so by enlisting a version of Reinhart's theory
(Levinson ). In previous work, the PI  sought to derive Condition C from
a more basic principle of 'Non-Redundancy', but as we discuss below the theory
faces empirical difficulties.
On an empirical level, there are a host of exceptions to Condition C, some of
which have been discussed in great detail by Reinhart and her followers
(). Thus (3), which is entirely acceptable with coreference, is
generally analyzed by observing that Condition C only constrains presupposed
coreference, or by making semantic values more fine-grained, replacing
individuals with 'guises' ():