Chosen-key attacks are a bit different than other kinds of cryptographic attacks. Usually, they are intended to not just break a cipher but to break the larger system which relies on that cipher.
The attacker should have some knowledge regarding the relationship between various keys that can be used in the cipher. Usually, he knows exactly what keys have been used or he himself can choose the secret key.
An example of a chosen-key attack can be a situation when an intruder tries to compromise a hash function based on a block cipher. If the attacker was able to find two different keys which would produce two block cipher outputs that are somehow related to each other, this would mean that the main property of hash functions (never produce predictable output!) had been broken.