Currying in a functional language (as others have pointed out above) is automatic within the language. This is because in most functional languages, a function is only allowed to accept one argument. Syntactic sugar under the hood will convert a multi-parameter function like this (using pseudo-functional-perl):

sub foo $x $y { $x * $y }
[download]

into something like this:

sub foo $x { sub $y { $x * $y } }
[download]

and then function application like this:

foo(3, 5)
[download]

into this:

foo(3)(5)
[download]

This is even more obvious in the type signatures of functions in languages like OCaml or Haskell. For instance, the above foo function would have the type signature in OCaml:

val foo :: int -> int -> int
[download]

Which basically can be read as "a function which takes an int, which returns a function, which also takes an int, which then returns an int".

So, to answer your question more directly, using currying in a language like Perl has (at best) only obscure usages. But in a functional language like Standard ML, OCaml or Haskell it is a fundemental part of how the language actually works.