Plus, any ad hoc experiment like this is likely to produce flawed results, due to factors not taken into account. Even people who study animal behaviour professionally have made (sometimes astonishing, once you know the full picture) leaps of logic, ommissions or let their personal biases influence their conclusions.
(incidentally, my own experience with dogs suggests that they are usually capable of recognising a few dozen words, independent of tone of voice and context. But even if that's true, it varies VASTLY between two dogs. Some dogs are remarkably good at regognizing words, or seem to be, and some dogs either are kinda dumb, or just don't care enough to react in a way that's expected. Which - i.e., the willingness to DO something - by itself is a source of potential bias, because we tend to equate "dog responds" with "dog understands", and it's obviously not a one-to-one correspondence. In either direction.)