In Hempel's raven paradox, we start with looking at a generalization,
(1) All ravens are black.
It is standardly held that a single black raven is evidence for (1).
(2) This is a black raven
would then be evidence for (1). But (1) is, by contraposition, logically equivalent to:
(3) All nonblack things are nonravens.
Then, by parallel, the evidence for (3) would be
(4) This is a non-black non-raven.
For instance, a white shoe would be evidence for (3). But since (3) is equivalent to (1), a white shoe would then be evidence that all ravens are black, which seems absurd.
There are, of course, many solutions proposed to this paradox -- the whole combinatorial gamut, in fact. My own preferred solution is to deny that (2) is, on its own, evidence for (1), and likewise that (4) is directly evidence for (3). There is a common error that assumes that evidence is just out there to be identified, but in reality nothing is evidence until you know what to do it. 'This is a black raven' tells us nothing about the truth of 'All ravens are black' unless we make a further assumption, namely, that there is nothing preventing other ravens from being relevantly similar to this one.
Consider this case. I ask a hoaxer whether all ravens are black; he shows me this black raven. Is this black raven evidence for 'all ravens are black'? Given the circumstances it seems not, because the fact that I'm getting it from someone known to make hoaxes raises the question of whether what I can learn from this black raven anything about other ravens -- the conditions suggest that even if this raven is black, it is perhaps being used to fool me into thinking that all ravens are black. This is an unusual situation, but it suffices, I think, to show that 'this is a black raven' only serves as evidence for 'all ravens are black' with further assumptions, and it is those assumptions that allow me to generalize. These assumptions are assumptions that let me say that this is not a freak case, or a case that problematizes the generalization.
We see something analogous to this in testimonial evidence. John says X happened; this is evidence that X happened if I can rule out, or at least assume to be ruled out, that John has incentive to lie, that John is delusional, etc. Testimonial evidence, and I would suggest other evidence, is evidence because we eliminate defective causes, causes that introduce a defect in the adequate causal chain that lets us infer one thing from another.
In the case of the ravens, 'This is a black raven' is evidence for 'All ravens are black' with the assumption that nothing prevents other ravens from being like this raven in color. Likewise, 'This is a non-black non-raven' is evidence for 'All nonblack things are nonravens' with the assumption that nothing prevents other non-ravens from being like this non-raven in color. But while the assumptions are structurally similar, they are not equivalent. Thus even when 'This is a non-black non-raven' is evidence for 'All nonblack things are nonravens', it is not thereby evidence for 'All ravens are black'; for that to be the case, in fact, we need to assume in addition that nothing prevents there being black things (since otherwise, it could instead be evidence that everything is non-black, whether raven or otherwise), or that nothing prevents there being ravens. (Likewise, 'This is a black raven' is not evidence for 'All nonblack things are nonravens' without the assumption that nothing prevents there being nonblack things, or that nothing prevents there being nonravens.)
