A \$\Sigma\Delta\$-ADC basically works deliberately adding "noise" (not real noise though) to the input signal. By doing so, the ADC can average out signals that are between two quantization levels.
For example, let's say that we have a 0-noise signal at 10.3, then the ADC would measure
10 10 10 10 10 10 10 10 10 10 ...
The ADC would round it down to the nearest quantization level. However, if we were to deliberately add some noise, we might get something like
10 11 10 10 11 10 10 10 11 10 ...
We have on average 3 x 11 and 7 x 10, which yields an average of \$10.3\$. We managed to capture a signal with a higher resolution, while the core ADC only detects 10's and 11's.
A \$\Sigma\Delta\$-DAC will do the same thing, though in the other direction. It will deliberately generate a signal with a lot of "noise" - now digitally, such that the average value is what it is supposed to be. For example, if we want to reproduce the 10.3 of earlier, while we only have a DAC that produces only integer values, then we can digitally generate a sequence
11 10 10 10 10 11 11 10 10 10 ...
We finally low-pass filter this signal to get a nice signal. This is also sometimes called dithering. Also note that this is basically a glorified digitally-controlled Pulse-Width Modulator with some better properties.
So the main advantage is that you can turn a low-resolution DAC into a higher resolution one by sacrificing the maximum bandwidth.