Linguistics and causal models

Linguistics is the shining torch of science and rationality in the Empire of Darkness that is the humanities. They make testable predictions, they understand statistics and hold beliefs with reasonable levels of confidence, they don’t overextend their methods to produce ridiculous dogma.

Something interesting I noted while watching this video: How Can Linguists Reconstruct Ancient Languages is a close analogy between linguistic models and causal models. 

Here’s a eustion: how do we know that Sanskrit and Greek descend from a common ancestral language? As opposed to, e.g. one descending from the other, or both languages being independent? What testable difference would you expect to see in these languages as a result?

Linguistically, what it means for Language B to descend from Language A is that all “features” of Language B are explained by Language A – i.e. if every feature in Language B has a cognate feature in Language A (and I say feature not term because a feature could also be, e.g. a relationship or rule between some number of terms). Well, realistically Language B would have independently developed features as well.