Its a term borrowed from the "operant conditioning" part of learning theory.
Basically OC states that there are four ways that learning takes place, represented by four "quadrants". The terms positive and negative here are used the same way they are in math when dealing in integers. Positive means
to add and negative means
to take away.
So, one quadrant is
positive reinforcement. This states that to increase the frequency of a desired behaviour, you should add a reinforcer after the desired behaviour is performed, upping the odds of its re-occurance. So, giving a dog a treat when they sit is an example of positive reinforcement. You're "adding" something. (the treat)
The quadrant that goes along with that one most often in terms of dog training is
negative punishment. Negative punishment states that in order to decrease the frequency of an undesired behaviour, one should take away a reinforcer. So an example of this would be putting the puppy in a "time out" in the next room when they bite. undesired behavior = biting, consequence = the removal of something "good", being with people/play, etc.
The other two are used in corrective based training more.
Positive punishment is when you add a punisher to the equation to decrease the frequency of an undesired behaviour. An example would be hitting a dog when they bite you, you're "adding" the hit.
The last is
negative reinforcement. This is my least favorite, as it basically states that in order to teach a new behaviour, you add punishment until the behaviour is performed and then take it away only when said behaviour has occured. An example that Skinner used with pigeons was to electrify the floor they stood on in their little metal cages, so they were being constantly shocked, and the shock would turn off only when they pecked food from a certain dish, teaching them that the way to turn off the shock was to eat from that dish.
Little long winded, but I hope that helped!
