1. The Evolution of Color Term Systems in Pama-Nyungan Claire Bowern, Yale University joint work with Hannah Haynie, Colorado State (work under review at PNAS)

2. CHIRILA: Bowern 2016

3. Why Phylogenetics of Color?  Evolution is central to current theories of color term naming:  hierarchical  unidirectional  Testable hypotheses that provide more info about how color term evolution is affected both by perception and language history.  More general example of how cognitive constraints may constrain paths of change.

4. Why phylogenetics of color?  But: there’s no direct study of color term and color system evolution, even though we know that color terms change over time.  cf. Buck 1949, Kay 1975 for PIE: e.g.  English yellow  Greek khlo:rós ‘green’  Old Irish gel ‘white’  Breton glas ‘gray, blue’

5. Berlin & Kay (1969), cf. Kay & Maffi (1999), Kay & Regier (2007), etc  “Languages are frequently observed to gain basic color terms...” but “... are infrequently or never observed to lose basic color terms.”  Languages “gain basic color terms in a partially fixed order”, which proceeds one term at a time. B, W B, W, R B, W, R, Y B, W, R, G B, W, R, Y, G B, W, R, Y, Bl B, W, R, Y, G, Bl IIIIIIIVV

6. Why Color in Pama-Nyungan?  Large, well articulated phylogeny (Bowern & Atkinson 2012)  All possible systems in B&K1969 and K&M1999 typologies attested in this family  Additional, unattested systems in the family  Strong claims in the literature about color naming systems constrained by culture and environment (cf. Wierzbicka 1989)

7. Overview for Today  Use phylogenetic methods to examine the evolution of color terms in Pama-Nyungan  Reconstruct ancestral states at the root and in subgroups  Examine trajectories of change:  order of color term acquisition  color term loss  (transitional) states that don’t conform to B&K’s typology  Use comparative method (not discussed here) and Bayesian methods.

8. Data  190 languages from Comparative Pama-Nyungan Lexical Database  Presence/absence for words glossed with Berlin & Kay’s 11 basic color terms (1969) (focusing on first 6 terms)

9. Deciding on what’s a color term  Dependent on the dictionary (translation) data  Often no information on range.  Inclusion in a wordlist/dictionary implies (but does not guarantee) salience and stability across speakers.  We include derived terms, non-monomorphemic terms, and terms that are polysemous with objects (e.g. red <> blood)  We focus on data for better-described languages and assume that gaps are genuine lexical gaps, not lack of recording (this was evaluated for each language)  Looking at the pattern of gaps  Using foci, not ranges (no evidence for exhaustive partitioning)

10. Binary Matrix

12. Phylogenetic Tree (Tree sample)

13. Methods

14. Bayesian Methods  language as an ‘evolutionary’ system equivalent to biological systems (cf. Messoudi 2012)  Formal modeling approach to historical linguistics: fit evolutionary model to data on a tree and evaluate statistically.  Evaluation of probability of a model, given our data and our knowledge of how things work.

15. Bayesian Phylogenetic Inference  Uses MCMC methods to derive the posterior probability of a model, given:  a set of trees representing the phylogeny  a set of data describing the state of tip nodes (languages)  Estimates likelihood of each character state at ancestral nodes.  Assumes random undirected walk  Estimates transition rates between character states  Here we use BayesTraits 2.0 package (Pagel and Meade 2004)

16. Transition Parameters (q) 012 0--q01q02 1q10--q12 2q20q21--

17. Probabilities of Trait Reconstruction

18. Hypotheses  That “languages are frequently observed to gain basic color terms...” but “Languages are infrequently or never observed to lose basic color terms.”  That languages “gain basic color terms in a partially fixed order”, which proceeds one term at a time.

19. Hypothesis: gain vs loss  Run multiple models of cognate evolution:  Model 1: Unconstrained  Model 2: Constrained to be gain-only [q10 = 0]  Model 3: Constrained to prefer gains (but allow limited loss)  Model 4: Constrained to be loss-only [q01 = 0]  Model 5: Constrained to prefer losses (but allow gains)  Model 6: Constrain losses and gains to be equal [q10 = q01]  Use Bayes Factors to determine support for models.

20. Ancestral State Reconstruction  Reconstruct probabilities of presence/absence of each color at well-supported nodes in the tree.

21. Hypothesis: Ordering of Gains  Test dependent <> independent models for each pair of colors in sequence.  Run Bayes Factor comparisons for independent vs dependent models

22. Results

23. Loss vs Gain  Gain-only and Loss-only models fail to converge  Gain/Loss predominant models converge to null (unrestricted) model  Gain/loss equal performs very poorly.

25. Trait Reconstruction  Overall: evidence for sequential color term gain  Northern Pama-Nyungan (Pama-Maric+) and P-Pny: BWR  That system also reconstructed from Proto-Pama-Nyungan, small support for green.  Evidence for green (and yellow) in Western Pama-Nyungan.  Higher nodes consistent with gradual term addition.  Subsequent further elaboration and reduction.  Eastern Pama-Nyungan shows more variability and recent elaboration.  NB, higher nodes modulated by uncertainty in the node itself.

26. Further discussion

27. Decreasing systems  Reconstructions for ‘green’ well supported in Western Pama-Nyungan, but Kartu and Kanyara-Mantharta languages tend not to have it.  Other examples of loss between proto-subgroup forms and modern languages, in particular.  No good evidence for rapid loss.

28. Rapid elaboration?  Mostly, evidence for step-wise elaboration, as per B&K.  But some examples of rapid gain: cf. Dhurga vs Darkinyung (Yuin-Kuri):  Proto-Yuin-Kuri (Black, White, Red)  Darkinyung: Black, White  Dharawal: Black, White, Red  Dhurga: Black, White, Red, Yellow, Green, Blue  Other examples: Yugambeh, Thura-Yura, Central NSW

29. Exceptional systems  Languages with brown but not blue (or yellow or green)  Languages where brown takes the place of yellow.  Languages with blue but not yellow or green  Red:  Some Paman languages without a true term for red?  Dha ŋ u (Yol ŋu) : no red, but green & yellow. (but cf miku ‘colored’, = red in other languages)  Some languages without white as color term?

30. Conclusions  Model comparison supports Kay and Maffi (1999)  But with caveats:  Evidence for term decreases in systems  Evidence for other systems  Evidence for fairly ‘rapid’ elaboration

31. Summary: System Changes  Elaboration (gaining terms)  Reduction (losing terms)  => Most according to B&K, but not entirely so:  gaining blue before yellow  losing red  Replacement  replacing yellow with brown

32. Summary: system changes  System constraints interacting with semantic and lexical change  shifts in (co)lexicalization of color  Semantic shift  alternations between items of salient colors and the color terms, leading to rapid lexical replacement.  hypernym <> hyponym shifts (particularly red <> color)  NB: no evidence for semantic shift within color systems (*green <> yellow, etc)

33. Conclusions  Consistent with a view where universals of perception constrain the types of change that occur, but lexical/change processes may lead to exceptions.

34. Acknowledgments  Chirila Database (Bowern 2016)  NSF-BCS-0844550 and BCS-1423711

35. Comparative Method Results  Problematic: too much lexical replacement, even among colors that we would reconstruct  Many languages recorded with more than one term; hard to know how to disambiguate.

36. ‘Black’ in Pama-Nyungan languages

37. *kara ‘black’

38. *maru ‘black’

39. Semantic change  *tyimpa ‘black’ <> ‘ashes’ [Karnic/Thura-Yura]  *gaywaraŋu ‘white’ <> ‘ashes’ [Yolŋu]  ‘white’ <> ‘shining’ <> ‘clean’ [various]  ‘green’ <> ‘raw’, ‘leafy’ [various]  ‘red’ <> ‘blood’, ‘red ochre’, ‘colored’ [Yolŋu, Paman]