The 20/24 bit DACs perform interpolation of the 16 bit data. In very rough terms, it's equivalent to a let's say a 1200 dpi scanner, which uses interpolation software to produce a 9600 dpi image. The software uses the built in algorithms to add bit data that it deems logical. For example, if it sees data 1 3 4 6 8 9 10, it may add 2 and 5 in the appropriate places. That doesn't mean that 2 and 5 belong there, but it makes an assumption that they do.
So, to make a long story short, the 24 bit dacs fill in bits in the 16 bit data stream. The resulting output is a 24 bit interpolated data stream. Hope that made sense.