Thread by @ndyjroo

2 great papers on IE phylogenetics. Read both!

G = Goldstein, “Divergence time estimation in IE”, in press @Diachronica2 ()

H+ = Heggarty, …, & Gray, “A hybrid model for the origin of IE”, @ScienceMagazine (https://t.co/rDHOpxmFYf)

A thread ... 1/11shorturl.at/iuITW
science.org/doi/full/10.11…

@Diachronica2 @ScienceMagazine G examines Bayesian divergence-time inference & Proto-Romance. Results include: Romance lg formation “begins after 300 CE” & “Classical Latin is a sampled ancestor of the Romance languages.” This connects with H+: some of their assertions & predictions are unsupported. 2/11

@Diachronica2 @ScienceMagazine H+ introduce a wonderful new database, IE-CoR (), co-edited by Paul Heggarty, Cormac Anderson, & @mattitiahu. IE-CoR is precise, thorough, well-documented & easy to use. Congratulations to the whole team on an exceptional resource. 3/11iecor.clld.org

@Diachronica2 @ScienceMagazine @mattitiahu Like other studies also led by Russell Gray (Gray & Atkinson 2003, Nicholls & Gray 2008, Bouckaert et al. 2012 w/ 2013 correction), H+ argues that PIE was spoken earlier than most IE-ists have thought: ~8k BP (similar to the Bouckaert et al. result). 4/11

@Diachronica2 @ScienceMagazine @mattitiahu Caveat! Much of H+ responds to Chang et al. 2015 (), itself a response to Bouckaert et al. That said, a striking new result with IE-CoR is that H+ do not replicate our finding that ancestry constraints distinguish early vs. late IE chronologies. 5/11shorturl.at/kKLQU

@Diachronica2 @ScienceMagazine @mattitiahu But H+ has problems, eg: 1 The dataset is full of homoplasy as we discussed in detail in 2015. Example: independent causatives of *mer- “die” in the meaning “kill” are coded as cognate. A root-derivative database like IE-CoR routinely treats analogous traits as homologous. 6/11

@Diachronica2 @ScienceMagazine @mattitiahu 2 The cousin problem: As shown here, early languages belonging with clade X are regularly inferred to be “cousins” rather than “siblings” to X (let alone in X). The Novgorod dialect is East Slavic but is here a sibling to East+West Slavic. 7/11

@Diachronica2 @ScienceMagazine @mattitiahu The cousin problem is systematic, eg in Gallo-Romance, Indo-Aryan, Scandinavian, West Germanic & West Iranian. It results from homplasy (advergence + drift) in closely related languages, inferred to be ancestral. The same causes lead to ancestors being placed as siblings. 8/11

@Diachronica2 @ScienceMagazine @mattitiahu 3 Extremely important in H+ is their comparison of binary vs multistate coding. Binary coding gives the early (“farming”) chronology, multistate coding turns out to give the late (“steppe”) chronology. H+ critique multistate coding results, but … 9/11

@Diachronica2 @ScienceMagazine @mattitiahu Against the binary analysis that H+ prefer, the multistate analysis infers a Nuclear IE clade, puts Novgorod in East Slavic, OCS in South Slavic, Middle Frisian & Old Dutch near Dutch and Frisian, and OHG with German. Other problems remain, to be sure. 10/11

@Diachronica2 @ScienceMagazine @mattitiahu The multistate analysis puts Tocharian in a weird place, but only at a low probability, and the H+ binary-coding analysis also puts Tocharian in a low-probability weird place (a subgroup with Anatolian). It’s not obvious that the multistate analysis overall is worse. 11/11

@Diachronica2 @ScienceMagazine @mattitiahu @threadreaderapp unroll