Environment

Bayesian phylogenetic analysis of Semitic languages identifies an Early Bronze Age origin of Semitic in the Near East

Description
Bayesian phylogenetic analysis of Semitic languages identifies an Early Bronze Age origin of Semitic in the Near East
Categories
Published
of 8
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Related Documents
Share
Transcript
  Bayesian phylogenetic analysis of Semiticlanguages identifies an Early Bronze Age srcin ofSemitic in the Near East Andrew Kitchen 1, * , Christopher Ehret 2 , Shiferaw Assefa 2 and Connie J. Mulligan 1 1 Department of Anthropology, PO Box 103610, University of Florida, Gainesville, FL 32610-3610, USA 2 Department of History, PO Box 951473, University of California—Los Angeles, Los Angeles, CA 90095-1473, USA Theevolutionoflanguagesprovidesauniqueopportunitytostudyhumanpopulationhistory.Theoriginof Semitic and the nature of dispersals by Semitic-speaking populations are of great importance to ourunderstanding of the ancient history of the Middle East and Horn of Africa. Semitic populations areassociatedwiththeoldestwrittenlanguagesandurbancivilizationsintheregion,whichgaverisetosomeof the world’s first major religious and literary traditions. In this study, we employ Bayesian computationalphylogenetic techniques recently developed in evolutionary biology to analyse Semitic lexical data bymodellinglanguageevolutionandexplicitlytestingalternativehypothesesofSemitichistory.Weimplementarelaxed linguistic clock to date language divergences and use epigraphic evidence for the sampling dates of extinctSemiticlanguagestocalibratetherateoflanguageevolution.OurstatisticaltestsofalternativeSemitichistoriessupportaninitialdivergenceofAkkadianfromancestralSemiticovercompetinghypotheses(e.g.anAfricanoriginofSemitic).WeestimateanEarlyBronzeAgeoriginforSemiticapproximately5750yearsagoin the Levant, and further propose that contemporary Ethiosemitic languages of Africa reflect a singleintroduction of early Ethiosemitic from southern Arabia approximately 2800 years ago. Keywords:  Semitic; language evolution; Middle East; Horn of Africa; Bayesian phylogenetics;population history 1. INTRODUCTION Semitic languages comprise one of the most studiedlanguage families in the world. Semitic is of particularinterest due to its association with the earliest civilizationsin Mesopotamia (Lloyd 1984), the Levant (Rendsburg 2003) and the Horn of Africa (Connah 2001), which gave rise to several of the world’s first major religious traditions(Judaism, Christianity and Islam) and literary works (e.g.the Akkadian poem  The epic of Gilgamesh ). The import-ance of Semitic dates back at least 4350 years beforepresent (YBP) to ancient Sumer in Mesopotamia, wherethe Akkadian language replaced Sumerian (Buccellati1997). From this time forward, archaeological evidencefor Semitic among the Hebrews and Phoenicians inthe Levant (Diakonoff 1998; Rendsburg 2003) and the Aksumites in the Horn of Africa (Connah 2001) suggeststhat Semitic-speaking populations and their languagesunderwent a complex history of geographical expansion,migration and diffusion tied to the emergence of theearliest urban civilizations in these regions (Lloyd 1984;Connah 2001; Richard 2003 b ; Nardo 2007). Uncertain-ties about key details of this history persist despiteextensive archaeological, genetic and linguistic studies of Semitic populations. A more comprehensive understand-ing of the precise origin and relationship of Semiticpopulations to each other is necessary to fully appreciatetheir complex history.Although multiple genetic studies of extant Semitic-speaking populations have been conducted (Nebel  et al  .2002;Capelli et al  .2006),muchisstillunknownaboutthegenealogical relationships of these populations. Mostprevious genetic studies focus on time frames that areeither too recent (the srcin of Jewish communities in theMiddle East and Africa; Hammer  et al  . 2000; Nebel  et al  .2001; Rosenberg  et al  . 2001) or too ancient (the out-of-Africa migration of modern humans; Passarino  et al  .1998; Quintana-Murci  et al  . 1999) to provide insightabout the srcin and dispersal of Semitic languages andSemitic-speaking populations.Previous historical linguistic studies of Semiticlanguages have used the comparative method to inferthe genealogical relationships of Semitic (for review,see Faber 1997). The comparative method is a techniquethat uses the pattern of shared, derived changes inlanguage (vocabulary, syntax or grammar), termedinnovations, to assess therelativerelatedness oflanguages,although this method cannot date the divergencesbetween languages (Campbell 2000). Cognates, whichare words that generally share a common form andmeaning through descent from a common ancestor (e.g.the English word ‘night’ is a cognate with the Germanword ‘Nacht’), serve as the data used most often incomparative analyses. doi:10.1098/rspb.2009.0408 Electronicsupplementarymaterialisavailableathttp://dx.doi.org/10.1098/rspb.2009.0408 or via http://rspb.royalsocietypublishing.org. * Author and present address for correspondence: Department of Biology, Pennsylvania State University, 208 Mueller Laboratory,University Park, PA 16802-5301, USA (aak11@psu.edu). Received   9 March 2009  Accepted   8 April 2009 This journal is q 2009 The Royal Society 2703 Publishedonline 29April2009 Proc. R. Soc. B  (2009)  276 , 2703–2710  The field of Semitic linguistics has generally coalescedaround a model that places the ancient Mesopotamianlanguage Akkadian as the most basal lineage of Semitic(Hetzron1976;Faber1997).Thisstandardmodeldivides Semitic into East Semitic, composed of the extinctAkkadian and Eblaite languages, and West Semitic,consisting of all remaining Semitic languages that aredistributed from the Levant to the Horn of Africa. WestSemitic is in turn divided into South (consisting of Ethiosemitic, Epigraphic South Arabian and ModernSouth Arabian (MSA)) and Central linguistic groups,but the genealogical relationships of the languages withinthese two groups are poorly defined (Huehnergard 1990,1992; Rodgers 1992; Faber 1997). Additionally, no consensus exists for placing Arabic in either the Centralor South Semitic group (Hetzron 1976; Blau 1978; Diem 1980; Huehnergard 1990, 1992; Faber 1997), which makes Arabic’s genealogical location simultaneouslyuncertain and interesting, as Central and South Semiticare geographically and genealogically distinct entities.Dating language divergences has been controversial,especiallywhenlinguisticclocksareinvolved(fordiscussion,seeRenfrew etal  .2000).Theexistence of a linguistic clockis controversial as it assumes that languages evolve at afixed rate (Ehret 2000), whereas there is evidence forvariation in rates of change between words and languagesand no reason why languages should evolve at fixed rates(Blust 2000). However, recent studies have shown thatmuch variation in the rates of linguistic change may followgeneralized rules that apply across language families(Pagel  et al  . 2007; Atkinson  et al  . 2008). This suggeststhat variation in the rates of change between words andlanguages can be modelled by applying techniques used inevolutionary biology (e.g. probabilistic modellingof relative rates of word change with relaxed clock orcovarion models of language evolution). Computationalphylogeneticmethodssuchastheseareconsistentwiththephilosophical underpinnings of the linguistic comparativemethod (i.e. inferring relationships by the comparison of similar features between languages) and provide anobjective statistical framework to accurately estimatelanguage divergences. Furthermore, Bayesian phyloge-netic methods offer distinct advantages by allowing forthe inclusion of multiple lines of evidence as priorprobabilities, incorporating the uncertainty of modelparameters in posterior probability estimates, and provid-ing straightforward statistical comparisons of models viaBayes factors (BFs).In this study, we analyse lexical data from 25 Semiticlanguages distributed throughout the Middle East andHorn of Africa (figure 1) using a Bayesian phylogeneticmethod to simultaneously infer genealogical relationshipsand estimate divergence dates of the Semitic languagesinvestigated here. In order to calibrate a relaxed linguisticclock and increase the accuracy of our divergence dateestimates, we use epigraphic data (text inscribed in stoneor tablets) from extinct Semitic languages (Akkadian,Aramaic, Ge’ez, ancient Hebrew and Ugaritic) combinedwith archaeological evidence for the sampling dates of theepigraphic data (the time at which the materials wereinscribed). We employ a log BF model-testing techniqueto statistically assess alternative Semitic histories andinvestigate different waysof modelling language evolution.Finally, we combine our divergence date estimates withepigraphic and archaeological evidence from allknown Semitic languages to create an integrated modelof Semitic history. 2. MATERIAL AND METHODS ( a )  Wordlists and cognate coding Wordlistsweremodified fromSwadesh’s 100-word listofmostconservedwords(Swadesh1955),withthefinallistscontaining96wordsfor25extantandextinctSemiticlanguages(fig.S1inthe electronic supplementary material). Wordlists for theEthiosemitic languages (Amharic, Argobba, Chaha, Gafat,Ge’ez, Geto, Harari, Innemor, Mesmes, Mesqan, Soddo,Tigre, Tigrinya, Walani and Zway) and Ogaden Arabic weredrawn from Bender (1971). Wordlists for Moroccan Arabic, SoqotriJibbaliMehriHarsusisrcin of AfroasiaticMesopotamiaAkkadianUgariticHebrewLevant Soddo / West Gurage Harari / East GurageOgaden ArabicHorn of AfricaTigreGe’ezTigrinyaMoroccanArabicAramaic6000 ybp <30004500NABEFGsrcinancestralSemitic?SemiticdispersalsCAmharicArgobbaGafat Figure 1. Map of Semitic languages and inferred dispersals.The locations of all languages sampled in this study, bothextinct and extant, are depicted on the map. The currentdistribution of Ethiosemitic languages follows Bender (1971)and distribution of the remaining languages follows Hetzron(1997). The ancient distribution of extinct languages is alsoindicated (i.e. Akkadian, Biblical Aramaic, Ge’ez, ancientHebrew and Ugaritic; Bender 1971; Hetzron 1997). The West Gurage (Chaha, Geto, Innemor, Mesmes and Mesqan)and East Gurage (Walani and Zway) Ethiosemitic languagegroupsincentralEthiopiaaredepictedastwocombinedgroups.The map also presents the dispersal of Semitic languagesinferred from our study. An srcin of Afroasiatic along theAfrican coast of the Red Sea, supported by comparativeanalyses (Ehret 1995; Ehret  et al  . 2004), is indicated in red,although other African origins of Afroasiatic have beenproposed (e.g. southwest Ethiopia; Blench 2006). Theassumed location of the divergence of ancestral SemiticfromAfroasiaticbetweentheAfricancoastoftheRedSeaandthe Near East is indicated in italics. Semitic dispersals aredepicted by arrows coloured according to the estimatedtime of divergence (see coloured time scale at top of figure),and important nodes from the phylogeny (figure 2) areplaced on the arrows to indicate where and when thesedivergences occurred. A. Kitchen  et al. Bayesian analysis of Semitic languages Proc. R. Soc. B 2704  (2009)  SouthArabianlanguages(Jibbali,Harsusi,MehriandSoqotri)and extinct non-African Semitic languages (Akkadian, BiblicalAramaic, ancient Hebrew and Ugaritic) were constructedfrom previously published lexicons (Leslau 1938; Gelb  et al  .1956; Sobelman & Harrel 1963; Rabin 1975). Cognate classes were determined for each of the 96 wordsusing a comparative method that emphasizes the similarity of consonant–consonant–consonant roots and known conso-nant shifts when comparing two words. The cognate datawerecoded intwoways:(i)asa25-by-96 multistate charactermatrix of cognate classes (‘A’–‘Q’) for each of the 96meanings (fig. S2 in the electronic supplementary material),and (ii) as a 25-by-673 binary matrix coding the presence(‘1’)orabsence(‘0’)ofeachofthe673cognateclassesineachlanguage (fig. S3 in the electronic supplementary material).Loanwords were identified using lexical information fromdistantly related, but geographically close, language families(such as Cushitic), as well as comparisons with lexicons of languages within the Semitic family. Identified loanwordswere excluded from all subsequent analyses.(  b )  Phylogenetic analysis and divergencedate estimation Phylogenies were constructed under a Bayesian frameworkusing BEAST v. 1.4.8 (Drummond & Rambaut 2007).BEAST uses a Markov chain Monte Carlo (MCMC)simulation technique to estimate the posterior distribution of parameters. All Markov chains were run for 20 000 000generations with samples taken every 1000 generations. Thefirst 4 000 000 generations were discarded as burn-in, andpost-run analysis of parameter plots in T RACER   v. 1.4(Rambaut & Drummond 2007) suggested all chains hadreached convergence by the end of the burn-in period.MCMC sampling and run conditions, and all prior distri-butions,wereidenticalforallanalysesunlessotherwisestated.An unordered model of cognate class evolution with equaland reversible instantaneous rates of changes between allpairs of cognate classes (i.e. the rates of A-to-B, A-to-C andB-to-A changes were identical) was used to analyse themultistate coded data, while a model with a single reversiblerate was used to analyse the binary coded data. Rateheterogeneity across lexical items was modelled by a gammadistribution of item-specific rates. This model accommo-dated variations in the rate of change across lexical items,such that conserved items (a single cognate class for alllanguages) were assigned a slower rate than the mean, whilehighly variable items (few shared cognate classes betweenlanguages) were assigned a faster rate than the mean.Priors for the gamma shape parameter were uniform on theinterval 0–50.Divergence times were estimated using an uncorrelatedlognormal relaxed-clock model that assumes a singleunderlying rate for the entire phylogeny, but allows forvariations in rates between branches (Drummond  et al  .2006). In order to calibrate the clock, we used sampling datesfor the five extinct languages in our dataset (Akkadian Z 2800YBP, Biblical Aramaic Z 1800 YBP, Ge’ez Z 1700 YBP,ancient Hebrew Z 2600 YBP and Ugaritic Z 3400 YBP;Rabin 1975) in a manner similar to how sampling dates areused in the studies of measurably evolving populations, suchas fast-evolving viruses or ancient DNA (Drummond  et al  .2003). These dates come from archaeological and epigraphicevidence associated with the linguistic source material, andthus provide the time at which the wordlists of the extinctlanguages were sampled (although the languages themselvesoften continued to exist for some time). Additionally, a set of five constraints taken from a combination of archaeological,epigraphic and historical evidence was placed on interiornodes. Such constraints allow for the inclusion of priorinformation and uncertainty regarding Semitic divergencetimes, which are strengths of Bayesian methods and havebeen successfully used to date the divergences of Indo-European (Gray & Atkinson 2003; Atkinson  et al  . 2005) andAustronesian (Gray  et al  . 2009) languages. These constraintsare:(i)theoriginofancientHebrew3200–4200YBP(Steiner1997), (ii) the srcin of Ugaritic 3400–4400 YBP (Pardee1997), (iii) the srcin of Aramaic 2850–3850 YBP (Kaufman1997) and (iv) the origin of Amharic 700–1700 YBP(Hudson 1997). Each of these constraints spans a 1000-year interval since the earliest epigraphic or historicalevidence for the language. An additional constraint (v) wasplacedonthetime ofthemostrecentcommonancestoroftheincluded Semitic languages to 4350–8000 YBP (the lowerdate is based on the earliest known epigraphic evidence of Akkadian; Buccellati 1997). An analysis was also performedwithout the constraint on the age of the root, which returnedan estimate of 4300–7750 YBP for the root, i.e. almostexactly our constraint range. All divergence time constraintsareintheformofuniformpriorsover theindicatedinterval.Auniform prior of 0.01 to 0.00001 cognate changes per wordper year (0.001–1% replacement rate per year) was placed onthe mean of the lognormal-distributed clock. The mean rateestimated from analysis of the binary data is 6.1 ! 10 K 5 replacements per cognate per year (95% highest probabilitydensity (HPD) Z 4.4 K 7.9 ! 10 K 5 ).The robustness of our results was investigated usinglog BF tests to compare phylogenies that were constrained tomodel alternative Semitic histories. Specifically, we firstcompared two versions of the standard model of Semitichistory: a model that placed Akkadian (i.e. East Semitic) atthe root versus an unconstrained analysis to assess indepen-dent support for a non-African Semitic root. We theninvestigated the position of Arabic in Semitic history bycomparing two variations of the standard model, one withArabicnestedwithinCentralSemiticandanotherwithArabicwithin South Semitic. We also tested the ability of differentmodels to account for variation in rates of linguistic changebetween lexical items and languages. In this case, wecompared the standard model to two alternatives: (i) nogamma distribution to model variation in the rate of changebetween lexical items and (ii) no relaxed clock to modelvariation in the rate of change between languages. All log BFtests of Semitic history incorporated a gamma distributionand relaxed clock since our log BF tests showed support forthese models. Marginal likelihoods for each model wereestimated using the smoothed harmonic mean of thelikelihood distribution (Newton  et al  . 1994; Redelings &Suchard 2005), and all log BF values were calculated bytaking the difference in the log of the marginal likelihoods of each model (Kass & Raftery 1995) with log BF valuesreported in log units. 3. RESULTS ( a )  Genealogy of Semitic languages Our phylogenetic analysis of Semitic languages producedthe phylogeny shown in figure 2. This phylogeny is basedon the binary dataset and incorporates all model features Bayesian analysis of Semitic languages  A. Kitchen  et al. Proc. R. Soc. B 2705  (2009)  that showed significant log BF support (log BF testswere equivocal in the placement of Arabic, so we placedArabic in Central Semitic based on previous comparativestudies; e.g. Hetzron 1976; Faber 1997). A brief summary of the phylogeny highlights include: (i) the greater age of non-African versus African Semitic languages (non-overlapping HPDs of 4150–7400 YBP for East/WestSemitic versus 2000–3800 YBP for Ethiosemitic); (ii)the near-simultaneous divergence of East, West, SouthandCentralSemiticlanguages;(iii) theearlydivergence of Arabic of approximately 4450 YBP (HPD: 3650–5800YBP); and (iv) the well-resolved and recent divergences(less than 3800 YBP) of Ethiosemitic languages in amonophyletic (single srcin) clade (a group of relatedlanguages). It is important to note that each node in thephylogeny represents an ancestral language that ishypothesized to have existed at the time of divergenceestimated for that node, whereas branch tips representactual languages at the time they were sampled. Longbranches are representative of long intervals betweendivergences and the presence of unsampled languages(e.g. the long branch between nodes E and F), whereasshort branches indicate rapid language divergence.Posterior probability estimates are shown for each branchand indicate the probability that a group of languages ismore closely related to each other than to other languages.Branches with posterior probability estimates % 0.70 wereconsidered to be unresolved (the relative pattern of divergence among the languages could not be ascertained)and were collapsed to reflect this uncertainty.(  b )  Semitic language divergence dates In addition to delineating the relationship betweendifferent Semitic languages, our phylogenetic analysisprovides dates for the divergences of the investigatedlanguages. The mean estimates of all language divergence UgariticHebrewAramaic350040504450850575054004650 Moroccan ArabicOgaden Arabic 205010501450 1.001.001.001.001.000.99 Akkadian MehriHarsusiSoqotriJibbali ABCDEF 1400950300750450 SoddoMesmesChahaInnemorMesqanGeto 1.00 1900160021009501200900170028002250 1.001.001.001.001.000.88 0.850.87 0.99 G TigrinyaTigreAmharicArgobbaGafat Ge’ez WalaniZwayHarari 0 ybp2000 CE1000 ybp1000 CE2000 ybp 0 CE3000 ybp1000 BCE4000 ybp2000 BCE5000 ybp3000 BCE6000 ybp4000 BCE7000 ybp5000 BCE Central Semitic South SemiticWest Semitic    E  a  s   t   S  e  m   i   t   i  c   C  e  n   t  r  a   l   S  e  m   i   t   i  c   M  o   d  e  r  n   S  o  u   t   h   A  r  a   b   i  a  n   E   t   h   i  o  s  e  m   i   t   i  c Figure 2. Phylogeny of Semitic languages. Our phylogeny of 25 Semitic languages based on binary encoded data is presentedwith mean divergence times to the right of each node and 95% HPD intervals indicated by light grey bars. The scale baralong the bottom of the phylogeny presents time in YBP. Posterior probabilities of branches are printed in italics above eachbranch with O 0.75 support. Extinct languages are underlined and all other languages are considered to evolve to the present.Subgroups of Semitic are identified by colour bars to the right of the phylogeny (purple bars, East Semitic; green bars, CentralSemitic; red bars, MSA; and blue bars, Ethiosemitic) and by three boxes (West, Central and South Semitic). Important nodesare indicated by letters: A, West Semitic; B, Central Semitic; C, Ugaritic–Hebrew–Aramaic; D, Arabic; E, South Semitic;F, MSA; and G, Ethiosemitic. The dashed line leading to Arabic reflects the fact that log BF tests were equivocal in theplacement of Arabic, so we placed Arabic in Central Semitic based on previous linguistic studies (e.g. Hetzron 1976; Faber 1997). The topology is rooted with Akkadian, which is preferred by our log BF analyses, and follows the constraints of thestandard model. A. Kitchen  et al. Bayesian analysis of Semitic languages Proc. R. Soc. B 2706  (2009)  times, with associated 95 per cent HPDs, are depicted inyears on the phylogeny in figure 2. Our phylogenyindicates the most basal divergence within Semiticoccurred at 5750 YBP (HPD: 4400–7400 YBP),suggesting an srcin of Semitic during the Early BronzeAge (Ehrich 1992). This result implies that a hypotheticalancestral language was extant during this period and gaverise to all of the Semitic languages investigated in thisstudy. The deepest four branches of the phylogenyindicate the divergences of East (root), West (node A),South (node E) and Central (node B) Semitic;these divergences are nearly coincident with largelyoverlapping HPDs (3300–7400 YBP), suggesting thatSemitic underwent a period of rapid diversification uponits srcin.Central Semitic (node B) initially diverges at approxi-mately4450YBP(HPD:3650–5800YBP)intoArabicanda group of ancient languages from the Levant (Aramaic,ancient Hebrew and Ugaritic), which in turn diverge(node C) at approximately 4050 YBP (HPD: 3750–4400YBP). The Arabic languages (node D) have anestimated divergence time of approximately 850 YBP(HPD: 400–1370 YBP).On the other half of the phylogeny, the South Semiticclade (node E) shows an ancient divergence of Ethio-semitic and MSA languages approximately 4650 YBP(HPD: 3300–6250 YBP), which overlaps with thetransition from the Early to Middle Bronze Age. Theearly divergence between Ethiosemitic and MSA isconsistent with previous historical linguistic proposalsthatMSAisadeepbranchofSemitic,linguisticallydistanteven from its closest relatives within the Semitic family(e.g. Murtonen 1967). The hypothetical ancestor of theMSA clade (node F) dates to approximately 2050 YBP(HPD: 1100–3100 YBP), which, coupled with the narrowgeographicaldistributionofMSAalongthesoutherncoastof Arabia, suggests that the diversification of MSAoccurred in this region.The single, well-supported (posterior probability Z 0.9976) branch leading to modern Ethiosemitic indicatesa single srcin for Semitic languages in the Horn of Africawith their diversification into North and South clades(node G) occurring at approximately 2850 YBP (HPD:2000–3800 YBP), during the Iron Age in the NearEastandoverlappingwiththepre-AksumiteandAksumiteperiods in the Horn of Africa (Connah 2001). Thelarge number of small internal branches in the Ethio-semitic group indicates a rapid diversification of theselanguages. The South Ethiosemitic languages separateinto three monophyletic clades that correspond toaccepted groupings of Ethiosemitic (Bender 1971)and show near-coincident divergences at approximately1200–1600 YBP.Our analysis of the multistate-encoded data produceddivergence date estimates and 95 percent HPDs that wereconsistent with those estimated from the binary encodeddata(see fig.S4 intheelectronic supplementary material).Themeandivergencedatesarealsoaltered:thedivergencesof East versus West Semitic, Central versus SouthSemitic, MSA versus Ethiosemitic and Ethiosemitic areolder in the multistate estimates, and the divergences of CentralSemiticandMSAareyoungerrelativetothebinaryestimates. The topologies are essentially the same withseveral small changes within the Ethiosemitic languagesand a closer clustering of Arabic and Aramaic in themultistate analysis. Importantly, all of the mean divergencedate estimates from the binary analysis fall within theHPDs of the multistate analysis. For figure 2, we chose topresent the phylogeny based on the binary datasetfollowing conventions of previous linguistic phylogeneticstudies (Gray & Atkinson 2003; Atkinson  et al  . 2005; Gray et al  . 2009).( c )  Log Bayes factor tests We assess the robustness of our analysis by statisticallytesting alternative Semitic histories. This was done usinglog BF model tests, which compare the probabilities thatvarious models produced for the observed data (i.e. thelexical list data). Log BF values (all values are in log units)in the intervals 0–0.5, 0.5–1, 1–2 and greater than 2 areconsidered ‘not worth mentioning’, ‘substantial’, ‘strong’and ‘decisive’support, respectively, for the primary model(Kass & Raftery 1995). We test alternative Semitichistories using two comparisons. The first comparisontests models that root Semitic with Akkadian (i.e. a NearEastern srcin of Semitic) relative to an unconstrainedmodel that allows for Near Eastern (i.e. Akkadian),African (i.e. Ethiosemitic) or Arabian (i.e. MSA) srcinsfor Semitic. This comparison shows substantial supportfor a model with an Akkadian root (log BF Z 0.641),consistent with the consensus of comparative linguisticanalyses (Faber 1997). The second comparison concernsthe placement of Arabic and compares the standardmodel, in which Arabic is placed within Central Semitic(e.g. Hetzron 1976; Faber 1997), with a single topological modification that places Arabic within South Semitic.This comparison showed little preference for a modelwithArabic within Central Semitic over one with Arabic withinSouth Semitic (log BF Z K 0.438). Interestingly, thelocation of Arabic within Semitic is the only discrepancyin topology and divergence date estimates between ourbinary and multistate analyses (figure 2; fig. S4 in theelectronic supplementary material).We also use log BFs to test the ability of differentmodels to accurately represent variation in the rates of linguistic change between words and languages. Our firstcomparison was between versions of the standard modelthat did and did not include a gamma distribution tomodel variation in the rate of linguistic change betweenlexical items. This log BF test shows substantial(log BF Z 0.574) support for a model that includes agamma distribution to model rate variation betweenwords. To place this in perspective, our estimate for theshape of the gamma distribution ( a Z 24.9) indicates thatthere is less variation in the rate of change between lexicalitems than there is within codon classes in mitochondrialcoding genomes of primates (Yang 1996). Our secondcomparison was between versions of the standard modelthat used relaxed and strict linguistic clocks to model ratevariation between languages. This log BF test showsdecisive support (log BF Z 13.0) for a model that includesa relaxed clock to model between-language variation.These two results demonstrate that our inclusion of ratevariation components in our model of linguistic evolutionsignificantly improves the fit between the data and ourmodel, and that there is substantial variation in thelinguistic rate of change between lineages and betweenlexical items. All log BF tests of Semitic history reported Bayesian analysis of Semitic languages  A. Kitchen  et al. Proc. R. Soc. B 2707  (2009)
Search
Tags
Related Search
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks
SAVE OUR EARTH

We need your sign to support Project to invent "SMART AND CONTROLLABLE REFLECTIVE BALLOONS" to cover the Sun and Save Our Earth.

More details...

Sign Now!

We are very appreciated for your Prompt Action!

x