The goal in this context is to guarantee that given a signature, or a set of signatures, you can confidently determine the public key that made them.
For a trivial example of how this might break down, consider an Ed25519 signature under a public key $A$, namely is a pair $(R, s)$ of curve point $R$ and scalar $s$ such that $[8 s] B = [8] R + [8 H(R, A, m)] A$, where $B$ is the standard base point and $m$ is the message. If we didn't hash $A$ in with $R$ and $m$, the verification equation would be $[8 s] B = [8] R + [8 H(R, m)] A$. There are several other public keys under which the same signature satisfies that verification equation: $A' = A + P$ where $P$ is a point of order dividing 8, since $[8] P = \mathcal O$.
That doesn't mean, given a target public key for which you don't know the corresponding scalar, you have any hope of finding a message/signature pair you didn't already have—i.e., it does not violate existential unforgeability. It just means that you can find another public key under which a message/signature pair you already have is also valid.
One of the basic ideas of Chaum's ecash was to anonymize transactions except fraudulent double-spending transactions, so given one signature neither you nor the bank could determine who made it (but you could still confirm its legitimacy), whereas given two signatures the bank can deanonymize the fraudster. However, rather than drop spoilers in your lap, I will leave you now to learn the thrilling details from the paper you are no doubt enjoying!