The video above does a nice job of demonstrating the phenomenon of audio Pareidolia in which our brain tries to impose meaning on what would appear to be random noise or stimuli. In the case of these examples, unlike more traditional examples of pareidolia, there are actual words and sentences being sung, but the way they are articulated and modified in order to be "musical" makes them ripe for misinterpretation.
This video is also a nice example of another phenomenon, which is the way our brain processes stimuli from different senses in order to create a unified picture of the world. I talked about this before in my discussion of the McGurk Effect. In the case of this video, the visual cues from the subtitles prime our brain to interpret the lyrics to fit the subtitles. This can create a bit of a disconnect in the viewer, particularly when confronted by a song with which one is very familiar. In these cases, it become much harder to "mishear" the lyrics. By contrast, for songs with which I am unfamiliar, all I can hear are the incorrect lyrics. In either case this is a nice demonstration of the fact that humans are not passive receivers of perceptual stimuli, but rather actively modify that stimuli in order to create a coherent and unified picture of the external world.