Quite a lot of songs DO resemble normal speech - the example I gave is one. Even when there are distortions, they are usually clearly audible as such, so you're learning to do expressive manipulation of the language. Somebody like Purcell setting English or Berlioz setting French knew what they were doing, and even though Wagner would sometimes draw out and heighten syllables, the overall stress pattern of his own German text is preserved.

The screwups are usually people setting languages they don't know that well themselves. Handel's "All WE like SHEEP". I suspect Mozart's Italian was not always very idiomatic either.

A large part of the effectiveness may be due to the way singing shuts off some of your filtering mechanisms. You don't get to revise the sounds in the song into a scansion pattern you're familiar with from your own language - the elements of that Turkish line have very different order and timing from the way you'd organize the same information in an English sentence, but if you sing it your English-based habits can't get in the way.