If someone would ask me this question I'd say:
"Nothing but their charges. The thing is so because we don't know what exactly happens in the virtual part of the process (say a scattering). Physicists don't know (and, to some extent, they don't really care) if the W being exchanged is going from vertex "a" to vertex "b" or if it's the other way around, so they admit if it goes from a to b, then it is, say, a $W^-$ and if it goes from b to a, then it is a $W^+$. Now, in either case the initial and final states are the same, so it doesn't make any difference wether the boson is positive or negative. Assuming so scientists avoid the problem while it is not essential and can reach very important conclusions."
Though this answer may sound I know what I'm saying I'm not positive about it, so I'd like to know if it's right or not, and how it could be improved.