3 ms·

So, suppose we want to know what 'bark' sounded like in the common language 15,000 years ago. Call that common language X. Well, take languages A and B now w

by graycat 14y ago

So, suppose we want to know what 'bark' sounded like in the common language 15,000 years ago. Call that common language X. Well, take languages A and B now where 'bark' is shared as a cognate. Now, start with language A now and realize that as we go back in time 15,000 years ago to language X how 'bark' was pronounced changed year by year. Then realize that as we go forward in time from language X 15,000 years ago to language B now, how 'bark' was pronounced has changed year by year. Then the change in how 'bark' is pronounced in languages now is smaller than either the change from language A now to language X 15,000 years ago or from language B now to language X 15,000 years ago. So, how close is how 'bark' is pronounced now to how it was pronounced 15,000 years ago? Closer than to how 'bark' is pronounced between A and B now. So, if 'bark' is pronounced nearly the same in languages A and B now, then it was pronounced nearly the same 15,000 years ago. Thus we have an approximation to how 'bark' sounded 15,000 years ago. It may be that languages A and B have a common ancestor more recently than 15,000 years ago. So, for a better approximation want to pick for languages A and B languages for which we have some hope that their most recent common ancestor was X 15,000 years ago. How to do this? For candidate pairs A and B, look at geographical and genetic distances and pick the most distant pair.

Retric 14y ago
Individual words often migrate between languages even if they don't share a common ancestor.
- graycat 14y ago
  Cute. And there are other issues with what I wrote. For a more careful analysis, would have to do, say, a 'probabilistic' analysis, e.g., where in principle can walk back to the beginning and start over but actually, e.g., with some common assumptions, in a space of dimension 3 or greater, probabilistically keep getting farther from the starting point. Your point should come out in the wash in estimating where there is a common language and how far back if use more than one word, etc. My guess, especially from many of the other comments on this thread, is that the basic data is so bad, noisy, etc. and pronunciation changes so large that the math I outlined, while basically correct, would mostly give such large distances that we wouldn't learn much. Maybe a better solution would be just to assume that the tree of languages would follow the tree of genetic inheritance and, then, via genetic analysis trace that tree. But the article wanted to ask what one of those 'common' words would 'sound' like 15,000 years ago, so I gave a way to get an approximation. The cute part is while 'father' in English sounds a bit different from 'vater' in German, the common ancestor should sound closer to either 'father' or 'vater' than those to two each other, that is, closer rather than farther away. That is, our intuition would be that the sound from 15,000 years ago should be much different from either the current 'father' or 'vater', but with the math model I gave that should be wrong.

by graycat 14y ago

Retric 14y ago

Individual words often migrate between languages even if they don't share a common ancestor.

graycat 14y ago
Cute. And there are other issues with what I wrote. For a more careful analysis, would have to do, say, a 'probabilistic' analysis, e.g., where in principle can walk back to the beginning and start over but actually, e.g., with some common assumptions, in a space of dimension 3 or greater, probabilistically keep getting farther from the starting point. Your point should come out in the wash in estimating where there is a common language and how far back if use more than one word, etc. My guess, especially from many of the other comments on this thread, is that the basic data is so bad, noisy, etc. and pronunciation changes so large that the math I outlined, while basically correct, would mostly give such large distances that we wouldn't learn much. Maybe a better solution would be just to assume that the tree of languages would follow the tree of genetic inheritance and, then, via genetic analysis trace that tree. But the article wanted to ask what one of those 'common' words would 'sound' like 15,000 years ago, so I gave a way to get an approximation. The cute part is while 'father' in English sounds a bit different from 'vater' in German, the common ancestor should sound closer to either 'father' or 'vater' than those to two each other, that is, closer rather than farther away. That is, our intuition would be that the sound from 15,000 years ago should be much different from either the current 'father' or 'vater', but with the math model I gave that should be wrong.