There's a controversy in linguistics over whether you can show families related in superfamiles and the latter related in super-superfamilies. I think most linguists accept that, historically speaking, a branching evolutionary pattern has occurred. However, far fewer believe that you can actually show this by analysis of the languages.
The problem is that languages change faster than even the fastest molecular clocks in organisms. You know the limitation involved in that. The faster the clock, the more accurate it is for recent events. Nevertheless, the fast clock's signal is lost in total turnover as you try to go back beyond a certain range limit.
Sure. A military conquest can virtually wipe out the indigenous language. Similarly, a migration can result in the immigrants' adopting the language of their new home. The traces of the old tongue can be so few as to be almost useless in reconstructing the history of what happened. DNA, however, is a more persistent marker. But before DNA, linguistics was at least a clue to the past.
A more reliable analogy to evolution is the gradual change in a lanaguage over centuries, where there's been no conquest. Say ... France or England for the past 1,000 years.