Language evolutionary tree

Forum » Broad sound correspondence hypothesis

Computerized comparative linguistics - Evolution

Comparative linguistics - Trees
Offline comment by: Harald on February 7, 2013, 11:00 I make the point about the system, but I don't understand why you skip some clear sound correspondences like S <-> T and C <-> T. This missing link is very obvious in the English to German detail sheet.
Posted by: Vincent on February 14, 2013, 21:00 Yes, you are right. S <-> T is part of the present rules. I just had a mistake in my XML file at the time you tested - it is corrected now. Re. C <-> T: this is part of sound correspondences I have dropped, because they caused a big "statistical noise" - which means it generated too much cognates between remote languages which were due to chance. But to be honest, this was before I adapted my algorithm to a pattern, where consonant matches only count if they appear in the same order within the words. I have tested the tree with C <-> T as a valid related consonant match - and it doesn't produce more "statistical noise". Now let's play farther and make following hypothesis: we don't work with consonants for coding the lexical morphemes, but with numbers which represent consonant clusters. It means, we gather consonants which are more or less remotely related within one cluster like this: Cluster 1: D, T, TH, C, S, Z, ZH, SH, CH Cluster 2: G, K, H, KH Cluster 3: B, P, F Cluster 4: V, W Cluster 5: L, R Cluster 6: Y Cluster 7: M Cluster 8: N The result is very convincing: here is the "new" tree (I have set the limit for statistical noise at 75 - so all branch-connections in the red zone should be considered are more related to chance ("statistical noise") than real genetic proximity! Here is the detail sheet for English to German (better proximity than in the classical sound correspondence tree by the way!) Harald, thank you for this discussion - it is good to try even simpler sound correpondance patterns and it seems it even improve the results. I will try more on these tracks.

Comparative linguistics - Trees

Offline comment by: Harald on February 7, 2013, 11:00

I make the point about the system, but I don't understand why you skip some clear sound correspondences like S <-> T and C <-> T. This missing link is very obvious in the English to German detail sheet.

Posted by: Vincent on February 14, 2013, 21:00

Yes, you are right. S <-> T is part of the present rules. I just had a mistake in my XML file at the time you tested - it is corrected now. Re. C <-> T: this is part of sound correspondences I have dropped, because they caused a big "statistical noise" - which means it generated too much cognates between remote languages which were due to chance. But to be honest, this was before I adapted my algorithm to a pattern, where consonant matches only count if they appear in the same order within the words. I have tested the tree with C <-> T as a valid related consonant match - and it doesn't produce more "statistical noise".

Now let's play farther and make following hypothesis: we don't work with consonants for coding the lexical morphemes, but with numbers which represent consonant clusters. It means, we gather consonants which are more or less remotely related within one cluster like this:

Cluster 1: D, T, TH, C, S, Z, ZH, SH, CH
Cluster 2: G, K, H, KH
Cluster 3: B, P, F
Cluster 4: V, W
Cluster 5: L, R
Cluster 6: Y
Cluster 7: M
Cluster 8: N

The result is very convincing: here is the "new" tree (I have set the limit for statistical noise at 75 - so all branch-connections in the red zone should be considered are more related to chance ("statistical noise") than real genetic proximity!

Language Evolutionary tree

Here is the detail sheet for English to German (better proximity than in the classical sound correspondence tree by the way!)

Language Evolutionary tree

Harald, thank you for this discussion - it is good to try even simpler sound correpondance patterns and it seems it even improve the results. I will try more on these tracks.

Blog author: Vincent Beaufils