At the simplest level, a key-value pair is just two values, one of which you have designated to be a “key” and the other you have designated to be the “value”.
However, it is more common to talk about key-value pairs in the context of a mapping, i.e. a (mathematical) function which maps from a key to the corresponding value or values. Depending on the properties of this mapping, you may constrain the set of key-value pairs. For example, for a 1-to-1 mapping, you need the keys in the set to be unique.
Follow-up questions:
Is this the same as an array?
Well … an array could be considered as a mapping from a set of indexes (integers) to values. But a mapping is more general. And in Java, arrays have other properties that distinguish them from Maps … and they have a much simpler, faster, and less memory-hungry implementation.
(Note that in some languages, there is no “array” data type per-se. Instead, the primitive is a “hash” or an “associative array” … which is a more general map.)
And is a key always a string?
No. A key can be any type. (It is generally a bad idea to use a mutable type as a key, especially if your mapping is implemented using one of the standard Map types in Java. However, even that can work in some circumstances.)
Say a variable of type String points to an object or int variable points to int value. Isn’t that variable a “key”, and the object a “value”?
No. Or at least, not in a static language like Java. The thing that distinguishes a key-value pair from a variable binding is that the “key” object is data, and hence can take different values. By contrast, a variable’s name is hard-wired in the source code of your program: you can’t change it at runtime.
(In some dynamic languages, you can create new variables dynamically (at runtime), and for such languages you could argue that variables are key-value pairs in a mapping that represents the scope … at some point in the program’s execution. But Java isn’t like that …)