The causes of language swapping and language shift

by Edward Pegler on 5 March, 2018

A review post outlining three main methods of language swapping: 1) Biggest language wins, 2) Richest or most violent language wins (this is helped if languages are similar), and 3) Everyone’s second language wins.

This post is aimed at anyone who’s interested, although it comes from a particular discussion between me and others about the spread of the Indo-European languages.

The question is simple; are there any generally accepted laws of language swapping where a geographic region changes its main language? Probably. Have I found any in the literature? No.

Part of the problem seems to be that language shift (the swapping of one main language for another by a particular bunch of people) is extremely politically loaded, coming with the baggage of colonialism, neo-colonialism, oppression of minorities and loss of diversity.

To be harsh, I’m not looking at any of this here. The remainder of this post is aimed at putting forward the main arguments I can find from various academics for why geographic areas swap their main languages. Here is what I’ve found.

Population dominance – Go with the herd

In this model a population coming into a new area brings more speakers of its (main) language into that area than the speakers any other language group already present in the area. This results in these minority linguistic groups showing language shift to the language of the incomers.

The methods of actually achieving linguistic dominance have historically included:

  • Immigration – the arrival of large numbers of new people into an area (e.g. of 17th-18th century English speakers to the colonies of America and Australia, or of 18th century Han Chinese farmers into frontier areas);
  • Displacement – forcing an existing population to leave the area (e.g. of 20th century Poland and Ukraine by the Soviet Union and Germany);
  • Population fall – usually caused by disease introduction, famine, genocide or birth rate decline (e.g. 15th to19th century introduction of European diseases into the Americas, Australia and South-western Africa, the 19th century potato blights effect on Gaelic-speaking western Ireland.)

It would have been good to separate these methods. However, they often go hand in hand. For example, the collapse of the American population was caused by immigration of Europeans bringing disease. Also, the aggressive displacement or extermination of one population by another may be to create new territories for immigrants to settle. It also needs a large population to carry out the displacement effectively.

Importantly, in this model population replacement does not necessarily mean language replacement. Short time scales are essential and language shift happens faster the higher the rate of swamping of minority languages. A slow, steady stream of immigrants over generations may eventually swamp the genetic signal of the original inhabitants of an area, but each small batch of immigrants will come as a minority, switching its language to the language of the majority, the old language of the original inhabitants (this has just been demonstrated nicely in the case of Vanuatu).

Elite dominance – Swap or lose!

This is the opposite of the model above – language shift to a minority language by the majority population of a region. It results when the minority language as used by the elite of society. It has been a significant cause of language shift in the last one to two hundred years. There are two main ways that this has happened:


In this, people are forced by a physically strong elite to learn their language under threat of violence (this includes the education system). It has occurred in a number of countries since the 19th century, including the imposition of English in Wales and Ireland, of Spanish in Latin America and of Turkish in its SE Turkey.

However, it should be pointed out that forcible methods are usually effective only in small geographical areas (up to around 30,000 km2) controlled by neighbouring, large, organised and highly literate states or empires (this is the case with the change from Arabic to Turkish in SE Turkey, or of Scots Gaelic and Welsh to English in the UK, or of Breton to French in France, or of Slovene to German in Austria, etc).

On the other hand, such forcible methods are not known to have been successful where a resident minority has tried to impose its language on the majority of people in a larger area (notable failures are of the Anglo-Normans in 12th century England, the Swedes in 19th century Finland). Many minority elites in a region haven’t even tried, rapidly adopting the language of the subject people.

(There may be an interesting example of Medieval enforced language change in East Prussia, where the German ‘Teutonic Knight’ monastic elite appear to have been reasonably organised and successful in changing the local language from Prussian to German between the 13th and 17th centuries. However, local population loss from plague, together with German immigration, may also explain this language change.)

Social betterment

In this, people wish to climb socially, and learn the language of the elite as well as having their children educated in the new language. Historically, this can be argued to be the case with the adoption of English (either UK or US) by various cultures since the 19th century, including by many Maltese and by city dwellers of India. The same can be said of the adoption of French in the cities of some of France’s ex colonies and even in some former British ones (e.g. Sierra Leone).

However, historically it is urban populations or populations of small islands that have tended to adopt the new languages, although they often stay bilingual (the adopted languages are, at least initially, linguae francae – see below). The population of the greater countryside does not adopt these new languages. This may now be changing. The advent of communications technology even in rural areas means that prestige languages are causing language shift in these areas too (this is clear with the loss of Gaelic and Irish in Scotland and Ireland).


I suspect that we shouldn’t underestimate the effects of mass literacy and the expectation of literacy on the success both of these methods in the last two hundred years. The increasing requirement to be able to read, whether signs or legal documents, means that a knowledge of the written language has become necessary to many inhabitants of modern nation states. If this is to cause major language shift it’s necessary for a large proportion of the population to be able to read. Such effects would not have been significant in earlier, less literate ages or with less organised states.

Making it easier – related language adoption

A more speculative idea is that elite languages are more readily adopted when they are not too dissimilar to the existing language, as the original population find it relatively easy to change their language.

In this case, the new language introduction could be by an incoming minority, albeit a dominant or prestigious minority, as above. However, the resistance to language shift is decreased.

Nick Ostler argues that this would explain the spread of the Arabic language to the Middle East and North Africa from the 7th century AD onward, as both already spoke Semitic languages (e.g. Aramaic and Coptic). The corollary to this is the failure of the Arabic language to take hold in Iran (where people spoke unrelated Iranian languages). It may also explain the adoption of Latin by Celts in western Europe during the Roman period (see further discussion below) and of Turkic languages by Mongols in the medieval period.

More historically, can this method explain the replacement of Byelorussian by Russian (both closely related Slavic languages) during the 20th century. I don’t know. It would be nice to have more certain examples of such a mechanism.

International chat – adoption of the lingua franca

Many people throughout history have, in fact, spoken more than one language: one at home, and at least one to less well known people. This is quite normal. We live in an age where many people speak English as well as their home language.

These languages are, as above, often used to get on in the world as they allow communication with other groups in trade, education and technical discussion.  Apart from modern English, there are many other historical examples, including Sanskrit across Southern Asia, Greek on the western Silk Road, Swahili in West Africa, Aramaic in Iran and the Near East, Medieval Latin in Europe, German in the Baltic and French in Western Europe.

However, it should be pointed out that these are second languages. As such, they often fail to replace the home language, and their long-term consequences can be ephemeral. This is clearly the case with Sanskrit, which is now extinct with no descendants, despite its widespread use several hundred years ago. The same can be said of Medieval Latin. Arguably, if America fades and China rises the incentive for an internationalist to swap their second language to Mandarin will rise too.

So are there any cases where the lingua franca has been adopted. For example, what about Ancient Latin or Arabic?

Ancient Latin is a most interesting case in that it appears to be a lingua franca which was adopted, in corrupt form, as the first language in central and northern Italy, the Iberian Peninsula, in France and among the Vlachs (‘Romanians’ in the broadest sense) of the Balkans. However, this needs to be analysed carefully.

In the case of Italy, Rome’s dominance of the local region, both in terms of elite power and population (due to establishment of coloniae), may have caused language shift in locally to Latin. This could be an example of population dominance or elite dominance of a small local area.

In the case of the Vlachs it is possible that East Romance languages result from the settlement as farmers in the Balkans of predominantly Italian legionaries, who then maintained their identity as a mountain people after the migration period. Alternatively, polyglot groups of settled legionaries from all over the empire (possibly together with some locals) might have used Latin as a lingua franca between them when they were settled in the Balkans. As more of them spoke Latin than anything else this became the dominant language, even though it was very few people’s first language.

On this basis, the adoption of a lingua franca as a first language may well be the result of linguistically fragmented groups in a region adopting a language of communication simply to unite them.

What happened in France and the Iberian Peninsula to cause language shift from the various original languages of these regions is more difficult to establish (as it is with Arabic). Firstly, it’s not known how uniform the languages of these areas were before their incorporation into the Roman Empire. Caesar suggests that large parts of France spoke Gaulish, although other languages spoken in France included Belgic, Greek, Basque/Aquitanian/Vascon, Iberian and Ligurian. In Spain Iberian, Celtiberian, Lusitanian and Gallaecian/NW Hispano-Celtic are all recorded, whilst Basque must also have been present.

Second, it’s not known how long it took for the language to shift to ‘Latin’. It’s also unknown what proportion of Latin-speaking (either as first or second language) settlers migrated to these regions. Subsequently, there was significant population loss and a large minority of Germanic speakers settled both regions. Therefore the adoption of Latin as a lingua franca in a linguistically fragmented landscape may be correct. However, it’s almost impossible to be certain.

Interestingly, linguae francae, due to their importance in written communications with other language groups, are precisely the ones that we know about historically. The languages that the majority of people daily spoke in Bronze Age Sumer, Iran, Greece, Crete, Anatolia, etc, may have been Sumerian, Elamite, Mycenaean Greek, ‘Linear A’, Hittite or Hurrian, but they may not. This makes determining ancient language shift so much harder.

Could the necessity to use such written languages have ever caused language shift? The answer may be no for peasant societies of the past. However, the effects of communications aided by states/empires, technology or both, mean that second languages can become increasingly important in daily life, not just in long-distance trade or exchange of ideas. Such an effect can be seen clearly in modern India, where many people use the neutral English in preference to Hindi in their communications, both written and spoken, with people outside their own linguistic groups. Whether such linguae francae will become first languages for the majority of such vast regions as India is unknown.

Combinations of effects – e.g. Hebrew

There is no reason why more than one of these effects cannot combine to cause language shift. Perhaps the classic case is the resurrection of the use of Hebrew, especially in Israel, from the 19th century onward. This was effective for at least three reasons.

  • In the 19th and early 20th centuries, Hebrew could be used as a written and spoken lingua franca between disparate Jewish groups in Europe and the Middle East as chances for communication improved between them due to technological advances.
  • When Jews migrated to Palestine in the 20th century these incomers were all minorities, speaking many different first languages, and needed a language of communication between them. In this they shared Hebrew.
  • The Israeli state, once formed in the mid 20th century, put its weight behind the use of Hebrew in education at the expense of other languages.

There’s no reason to believe that other combinations of effects couldn’t result in language shift in the past, now, or in the future.

Speculations in Prehistory

So far, so historical. Unfortunately, many of us on blogearth want to understand why language change happened in Prehistory. Frankly, most of this revolves around our fascination either with our chosen national identifiers or with Indo-European origins or, often, both. What troubles me is that none of us seems to take the trouble to find out what can cause language shifts either in the present or during well recorded history.

So one particularly prevalent view, as expressed by many enthusiasts, is that prehistoric warrior elites have caused language changes. This is, essentially, a machismo view of prehistory, either despaired of (e.g. Marija Gimbutas’ view of the coming of the Indo-Europeans to peaceful ‘Old Europe’), celebrated (e.g. Madison Grant, Gustav Kossinna or even the Eurogenes Blog (David Wesolowski has taken exception to this slur and I’m happy to retract it), or a cause of mental confusion (e.g. David Anthony).

And yet the historical evidence for language change resulting from nomadic warrior elites is still scarce. Eastern Europe suffered repeated waves of horse-warrior incursions after the fall of Rome. This includes the Alans, Mongols, Pechenegs, Bulgars, Cumans, Magyars, Avars, etc. Some of them went on to rule the areas they invaded (e.g. the Bulgars and Cumans). However, only one of these groups, the Magyars, left a linguistic legacy.

As stated by Jean Sedlar, the success of the Magyars was probably due to their arrival in a sparsely populated area (the Hungarian Plain), meaning that they may have been in the majority there (the previous immigrants to the area, the Avars, were also of eastern origin according to the limited genetic data, and possibly spoke a related language, which perhaps helped).

On the other hand, the slowly migrating mass of Slav peasant farmers, not on horseback, who settled a depopulated Eastern Europe in great numbers from the 6th century onward, have left a huge linguistic legacy, despite often being ruled over by horse-riding elites (e.g. the Bulgars and perhaps the Croats) speaking other languages.

So this appears to be a failure of elite dominance and a success for population dominance. The reason for this is probably simple; nomadic warrior elites were not that organised and their populations not that educated. Furthermore, they were not part of a neighbouring majority.

So maybe there were (possibly horse-riding) elites in the Bronze Age. They may even have sired many children and left their disproportionate legacy in Y-chromosomes (as seems quite likely). And yes, they probably did sweep into India and Europe and Iran and Anatolia during this time. But, from the point of view of language change it’s not them that matter. They could speak any languages they wanted, be they related to Iranian, Finnish, Turkish or something now lost. It is the slow, vast, uncelebrated swarms of peasant migrants, wherever they came from, who probably caused the language shifts.


Amorim, C.E.G. (posted on bioarxv 20-2-2018) Understanding 6th-Century Barbarian Social Organization and
Migration through Paleogenomics, pp27.

A sketchy bit of data so far, but which suggests considerable Italian migration to Central Europe during the late Roman period. Though definitely a minority, perhaps there were local areas where Italians were in the majority. Additionally, a couple of samples may (!!!! too little data) suggest a notable influx of foreign genetics into the Hungarian Plain during the Avar period. Two Avars do not make a population shift, however.

Heggarty, P. 2015 Prehistory through language and archaeology, In: The Routledge Handbook of Historical Linguistics (Bowern, C. & Evans, B.), 598-626.

Very pro Colin Renfrew’s Anatolian PIE theory, but a good overview of the mainstream view on language shift.

Ostler, N. 2005 Empires of the Word: A Language History of the World, Harper pp614.

A really good source for many of the ideas in this post, concentrating on history.

Posth, C. et al. 2018 Language continuity despite population replacement in Remote Oceania, Nature Ecology & Evolution (online).

Appearing after I started writing this, it didn’t phase me as much as it might have. I’ve only seen the summary here.

Sedlar, J.W. 1994 East Central Europe in the Middle Ages, 1000-1500, Uni. Washington, pp556.

Source of information on language and changes in central Europe during the migration period.



So, do the different linguistic family trees of Indo-European tell you different things? I think that grammar-based trees are better at telling you about the original splits between language sub-groups, whereas word-based trees can tell you about how isolated those sub-groups became.

The family trees of Indo-European as envisaged by Don Ringe and colleagues on the one hand, and Russell Gray & Quentin Atkinson on the other.

Why are these two family trees for Indo-European so different in shape?

At the beginning of the millennium two papers were produced which both purported to show, using statistical ‘cladistic’ analysis, the structure of the Indo-European language family. The resulting family trees of the two studies were completely different.

Gray & Atkinson’s 2003 study was based largely on modern IE languages, and used the existence of related words in the lexicon of each of the languages to construct a family tree. Ringe, Warnow and Taylor’s 2002 study instead used the earliest languages that they could and, as well as the lexicon, also used grammatical (morphology) and sound change (phonology) features to help classify the results.

And the tree shapes would have, frankly, looked much the same if the linguist of the latter group, Don Ringe, hadn’t known some of the answers that he wanted beforehand. His frustration was that there were loads of shared words between languages but not that many shared morphological or phonological features. Yet his hunch was that the latter features were more important than the words.

So what Ringe’s group did was to weight morphology and phonology hugely compared to the words. This gave them answers that made sense to Don Ringe but left the group subject to some criticism for effectively biasing the data. What he was saying was, in effect, that the words were of lesser importance than the grammatical rules.

Now, what I find frustrating is not that Ringe’s group weighted the data but that they didn’t just go the whole hog and dump the lexicon altogether, just concentrating on morphology and phonology. Maybe they didn’t think that there was enough data. Either way it would have been nice to see the results of such a study.

Building trees from words, sounds and grammar

Don Ringe’s bias was to say that words are important, but not that important in constructing the history of IE languages. He attributes this in part to word borrowing between languages. Whilst Ringe’s group did their best to spot word borrowing and filter out the borrowed words, they admit that they may have failed to detect borrowing between languages.

In this they appear to disagree with Andrew Garrett, who thinks exactly the opposite. Andrew Garrett’s position is that the current phonology and grammatical rules of IE language sub-groups are late features, formed as a result of geographically adjacent IE dialects converging in their grammars and sounds. This, he argues, is the result of crises. Therefore Garrett prefers to see the words as being key to older connections, as these are less likely to be changed in his opinion.

The aim of this post (which has, once again, turned into an essay) is to try to see if it’s possible to look at family trees for either grammar, sounds or words and see how they differ and how they might be affected either through original shared connections or through geographical convergence.

Let’s study Indo-European morphology!

Now personally, I can’t do cladistic analysis, but what I can do is play with spreadsheets.  What I’ve done here is to take phonological and morphological differences recognised between IE language sub-groups and analyse them in order to find if they reveal anything on their own.

To increase the size of the list I’ve added the morphology list of Gamkrelizde & Ivanov (1995), page 345 to the phonology and morphology lists of Ringe et al. (2002) (as supplemented by the updated list of Nakhleh 2007). There may be other good, equivalent lists out there, but I don’t know about them. It should be pointed out that Ringe and G&I’s lists overlap in terms of their morphological criteria, and there are a couple of cases where a morphological criterion from one list can be mapped exactly onto that from the other.

However, the morphological interpretations of Ringe and G&I are clearly not the same. Surprisingly, in many cases where the morphological criteria appear to be describing the same grammatical feature, the list of which language sub-groups share that criterion is often not even the same. In such cases I’ve allowed each criterion to be left separate as this should prevent any bias whilst also preventing any unnecessary loss of data. However, where their morphological criteria are identical I have merged them.

Furthermore, to save complications I have only used criteria which allow the comparison of whole, well established language sub-groups. I have not used criteria which only occur in one language sub-group, as these are not informative when trying to compare sub-groups for similarity. Additionally, all criteria indicate the presence, rather than the absence of a morphological or phonological feature.

This leaves me with 40 morphological and 4 phonological criteria, a total of 44 criteria. As the 4 phonological criteria are not sufficient to be analysed on their own I’ve incorporated them into the morphological analysis.

A list of Indo-European grammatical features and which languages they occur in.

The grammatical and phonological criteria used in this analysis and which Indo-European languages they occur in (Ce = Celtic, It = Italic, To = Tocharian, An = Anatolian, Ar = Armenian, Gr = Greek, II = Indo-Iranian, Sl = Slavic, Ba = Baltic, Ge = Germanic). The fact that this looks like a question mark is pure chance.

The list to the right shows the criteria discussed. Those marked M and P are morphological and phonological features taken from Ringe et al. (2002) p113-120 and Nakhleh 2007. Those marked G* are morphological features taken from G&I p345.

Where criteria in Ringe and G&I are similar but not identically allocated to language sub-groups, the related criteria are shown in the next column. If a criterion is exclusive (i.e. it clashes with another), then the criteria which cannot occur simultaneously are listed in the ‘opposes’ column. All criteria marked by an asterisk (*) are considered polymorphic (i.e. could develop more than once) by Ringe’s team and I’ve extended these to G&I. Criterion M11(2), although omitted by Ringe’s team from their later analysis, has been left in here, but with additions to show that it may not be unique to Italic and Celtic.

The presence of a particular feature in an IE language sub-group is indicated by a 1. Where 0.5 is given this is where G&I have indicated a potential occurrence in a language sub-group by brackets or, as in the case of M11(2), Ringe’s team suspect that related features may occur in this sub-group. These 0.5s have been ignored in the analysis, although they actually make little to no difference to the analysis.

You’ll notice that the language sub-groups and criteria have not been put in alphabetical order. They are, in fact, in the order that appears to give the highest correlations between each successive family and criterion. While it looks clearer, it also gives a hint as to how the morphology and phonology of the language sub-groups might be connected.

Please note that I have not included Albanian, Phrygian, Messapic and Venetic, which are listed either by Ringe’s team or by G&I. This is because they appear to share very few features with the other sub-groups, presumably due to either heavy loss of features (in the case of Albanian) or due to lack of evidence.

The results

The first table indicates simply the number of criterion correspondences between different language sub-groups. It’s difficult to interpret as different language sub-groups have different numbers of potential corresponding features. The number of potential correspondences in a sub-group is indicated where both column and row have the same label: e.g. row Gr, column Gr shows the value 19, so 19 is the maximum number of correspondences that the Greek language sub-group could have with other sub-groups.

The second table shows the percentage of criterion correspondence between sub-groups. The diagonal line of 100s is simply the result of comparing a language family with itself (which gives 100%). To compare how many features Tocharian shares, say, with Italic, find the column To and compare with row It.

In this case it shows that Tocharian shares 89% of its features with Italic, but note that Italic shares only 47% of its features with Tocharian. This is firstly because Tocharian has only 9 potentially sharable grammatical features, whereas Italic has 17, meaning that Italic could never share more than 100*9/17=53% of its features with Tocharian.

Significantly, the phonological features used have little overall effect on the pattern seen and could have been removed, giving much the same result.

What comes out clearly from the data is the strong morphological relationship between Italic and Celtic as well as, to a lesser extent, of Tocharian with Italic and Celtic.

The morphological relationship of Anatolian with all other sub-groups of IE is fairly weak across the board, perhaps as a result of changes occurring either in Anatolian or in the rest of the IE sub-groups when they were still clustered as ‘late PIE’. However, although distant, Anatolian’s best apparent correlation is with Tocharian. Note however that Tocharian shares just 33% of its features with Anatolian. These are not close cousins.

Amongst the remainder, Baltic shares many features with Slavic (as is well known). Greek, Armenian and Indo-Iranian also appear to be related in some way, Greek and Armenian perhaps being more closely connected out of the three. Interestingly, Germanic shares many features with both Indo-Iranian and Baltic.

All of this evidence tends to suggest a grammatical and phonological divide between Tocharian and ‘Italo-Celtic’ on the one hand; and Armenian, Greek, Indo-Iranian, ‘Balto-Slavic’ and Germanic on the other.

However, for the remaining languages: Greek, Armenian, Indo-Iranian, Germanic, Slavic and Baltic, there appear to be chains of connection between the different sub-groups which don’t lend themselves obviously to making divisions between groups of languages.

Exclusive States

It could be argued that what we’ve got here could be purely based upon chance. Theoretically, all of these morphological and phonological features could have existed in the earliest forms of PIE and been lost randomly in individual sub-groups.

However, this is not possible in a number of cases. Some criteria, such as G13/M5, form pairs; in this case G13b/M5(1), and G13a & M5(2), indicate the type of endings found in the middle verbs of IE languages. These are exclusive states. It’s not possible to have both.

Using these criteria, it’s theoretically possible to cluster the sub-groups into groups:

Group A – Italo-Celtic, with Tocharian and Anatolian – this is based upon the presence of criteria G10a and G13b/M5(1).

Group B  – Those sub-groups not in group A – this is based upon the presence of criteria G10b and related criteria M5(2) and G13a. This group (called ‘Rump IE’ by Jay Jasanoff) can perhaps be further subdivided into:

Group C  – Greek, Armenian and Indo-Iranian (called the ‘Southern Dialect Group’ by Jasanoff) – this is based upon the presence of potentially polymorphic criteria G2 and G4a/M10(4). As Ringe’s team have pointed out, the possibility that these features arose more than once in the development of IE sub-groups makes this grouping weaker than it should be, but it’s the best I’ve got.

Group D – Balto-Slavic and Germanic (called the ‘Northern Dialect Group’ by Jasanoff) – this is based upon the presence of the polymorphic criteria G2c and the interrelated G4/M10a(10), M10b(13) and M9b(10), again with the same potential weaknesses as group C.

The major problem with Groups C and D is the presence in both groups of the ‘satem’ language sub-groups Baltic, Slavic and Indo-Iranian, defined by phonological criteria P2(2) and P3(2). ‘Satemization’ is not a reversible process, so if the division is correct then P2(2) and P3(2) would need to be polymorphic. One way or the other, polymorphism is a problem for subdividing group B.

Comparison with Ringe’s Analysis

The Ringe group’s best fit, family tree of Indo-European sub-groups, based on all criteria but weighted for morphology and phonology, showing the latest positions that individual grammatical or phonological criteria could have arisen.

My preferred tree for Indo-European branches based purely on grammar and phonology. Note that I have chosen to see phonology as potentially able to be passed between branches.

My preferred tree for Indo-European sub-groups based purely on grammar and phonology. Note that I have chosen to see phonology as potentially able to be passed between sub-groups.

What’s shown here is the morphological tree that appears to fit the data best to me, compared to the best tree of Ringe’s team (from Nakhleh et al. 2005), produced using words as well as morphology and phonology, but heavily weighted toward morphology and phonology.

I have added morphological and phonological criteria to both trees in the position of their latest possible appearance on each tree. Criteria in black are exclusive, and are probably the most important. However, those starred are potentially polymorphic, making them less useful. Crossed out, grey criteria indicate the last possible occurrence of these criteria. Phonological criteria are red.

As Ringe’s team have pointed out, the aim of any tree is to have a continuous part of the tree occupied by one feature, except in the case of polymorphism. However, I would like to go further, and add that the aim of any tree like this is to get the latest possible appearance of the criteria as far to the right of the tree as you can.

I have chosen to downgrade the significance of phonology, accepting the argument of some linguists that satemization, as represented by P2(2) and P3(2), was a polymorphic development, happening in more than one sub-group, perhaps by convergence (e.g. the discussion over Nuristani in the Indo-Iranian family – see Hegedűs (2012)). In fact, even Ringe’s team have said ‘Because P2 and P3 are less secure than the other phonological characters and than most morphological characters, one cannot easily judge the performance of any given method by how it treats these two characters’.

Either way, this is a matter of choice. If I had taken P2(2) and P3(2) as being important I would have clustered Balto-Slavic with Indo-Iranian and made Germanic the next most closely related language, with Greek and Armenian separate.

My reason for not doing so is that it appears to push too many mutually exclusive criteria to the left of the diagram, even meaning that criteria such as M10(4) and M10(10) are required to be common at the same level, something which again only makes sense if polymorphism is used to justify their emergence repeatedly, as Ringe’s group do.

Apart from this point, the other important difference is in my grouping is of Italo-Celtic with Tocharian, which to me fits the morphology data better.

Incompatibility with words

In terms of incompatibilities (i.e. criteria or words which appear not to fit the trees), both show some incompatibilities with the filtered word list of Don Ringe’s team. For my tree, 15 words are incompatible, one twice (allx2 beard break1 breast1 drink  fish1 fog2 free give1 one pour long1 straight tearx2 young2) not forgetting the additional 2 phonological incompatibilities. In the case of Ringe’s tree, 14 words are incompatible (all1 arm beard break1 breast1 float2 free head one pour straight tear thousand1 young2). Ringe clearly wins.

Words connecting incompatible sub-groups are, for both trees, largely between Germanic and either Italic or Celtic. In Ringe’s tree this is 7 out 14, whereas for mine it is 11 out of 16. As Ringe pointed out, this suggests some peculiarity of Germanic. However, the positioning of Germanic with Balto-Slavic on my tree eliminates 3 of Ringe’s incompatibles (arm float2 thousand1).

The major problem with my tree however, which lends weight to Ringe’s analysis of the position of Tocharian, is that two of my incompatibles (drink give1) are between Tocharian and Anatolian, and these could be eliminated by pairing Tocharian with Anatolian against the rest. Reordering this brings my incompatibles down to 14, of which 11 are still for Germanic with Italic or Celtic, and 2 of the remaining 3 are for 1 word.

Comparison of word and rule based trees

My modified tree, based on the word evidence of Tocharian's early separation before Italo-Celtic.

My modified Indo-European tree, based on the word evidence of Tocharian’s early separation before Italo-Celtic.

A Ringe/Gray&Atkinson composite Indo-European tree, based purely on the evidence of words.

A Ringe/Gray&Atkinson composite Indo-European tree, based purely on the evidence of words.

Here are the diagrams of my modified morphology-based tree and Ringe’s purely word-based tree from Nakhleh et al. (which is much the same as Gray & Atkinson’s tree). In terms of word problems, this tree has just 9 words which don’t fit, compared to the 13 from my tree. However, as shown by the exclamation marks, at least three morphological criteria don’t fit, as well as three phonological criteria.

Apart from Anatolian separating first, the trees shown here are essentially different. Groups A, B, C and D do not occur in the lexical tree. Three morphologically exclusive sets, M5/G13, G10 and M8, are not obeyed by the lexical tree.

If you take the view that morphology trumps words then the tree on the left or my previous tree is more likely to be correct. If you agree with Garrett that words trump rules then (if you had to draw a tree) it would be more like that on the right. If you take the view of Ringe et al. (2002) then some kind of compromise tree is best.

So which, if any, is right?


A cladistic tree for Indo-European allowing late interconnection of dialects after their split.

Is this the best tree for Indo-European languages? In making some branches interact while still mutually intelligible, this forces Anatolian, Tocharian, Armenian and Greek to the edge, whilst allowing the remaining groups to interconnect.

I think that it’s essential to bring in geography at this point and to consider which language sub-groups are likely to have been closest to which others during history. In this case, the extremes of geographical location are of Tocharian in the East (within the Tarim Basin of modern Western China) and of Celtic in the West.

Notably, these two language sub-groups appear to be quite close morphologically. This means that morphological convergence is immensely unlikely for these two language sub-groups. Conversely, Germanic, Celtic and Italic were adjacent to each other at the beginning of the Iron Age and yet Germanic is about as far, morphologically, from Celtic and Italic as it’s possible to get.

Therefore it seems to me that Andrew Garrett must be wrong, at least in his conclusions about large-scale convergent morphology. On the other hand, Garrett arguably has a better case with phonology, as it’s difficult to reconcile the morphological evidence with the phonological evidence. Furthermore, Indo-Iranian was possibly in geographical proximity to Slavic and/or Baltic during the late Bronze or Iron Age, so satemisation may have occurred during contact.

The branching that appears to be happening on the word-based tree is, on this reading, a result of long-term geographical separation. Therefore the apparent grouping of Baltic, Slavic, Germanic, Italic and Celtic is a result of these language sub-groups coming into geographical proximity within Europe whilst still mutually intelligible but after dialect variations had occurred.

On the other hand, Anatolian and Tocharian and, to a lesser extent, Greek or Armenian, appear to have become separated from the rest of IE for long enough to make them unintelligible when IE languages came into contact with them again later in history.

Was this exercise useful? Probably not. Do I feel like I’m just re-inventing Ringe’s tree? A bit. However, I do think that it’s useful to see both the Ringe group tree and the Gray & Atkinson-type tree as both potentially useful in giving clues as to the prehistoric development of IE languages.


Arkadiev, P. 2016 How Languages Borrow Morphology (conference powerpoint presentation).

Gamkrelizde, T.V. & Ivanov, V.V. 1984 Indo-European and the Indo-Europeans (translated by Joanna Nicholls 1995), New York, Mouton de Gruyter, pp1128.

Source of some of the grammar rules.

Garrett, A. 2006 Convergence in the Formation of Indo‑European Subgroups: Phylogeny and Chronology, In: Phylogenetic methods and the prehistory of languages, (Forster, P. & Renfrew, C. eds.) Cambridge, 139-151.

 Q.D. 2003. Language‑tree
 divergence times 

Hegedűs, I. 2012 The RUKI rule in Nuristani, In: The Sound of Indo-European: Phonetics, Phonemics, and Morphophonemics., Museum Tusculanum, 145-167.

Just in case you’re wondering, this is criterion P3(2), which some was not fully applied in one sub-group of Indo-Iranian.

Jasanoff, J.H. 2009  Notes on the Internal History of the PIE Optative, In: East and West: papers in Indo-European studies (K. Yoshida and B. Vine eds). Bremen: Hempen, 47—67.

As well as providing alternative names for Groups A, B and C, he argues here for the sharing of optative ending ‘*-oi’ (M6(2) in all of the ‘Rump IE’ dialects (Gk, II, Ar, Ba, Sl, Ge, including Albanian).

Nakleh, L., Warnow, T, Ringe, D. & Evans, S.N. 2005 A Comparison of Phylogenetic Reconstruction Methods on an IE Dataset, Trans. 
Philological Soc. 103,

This is an extension of the work of Ringe et al. 2002, showing the different results from weighting the trees for grammar and not applying the weighting.

Nakleh, L., Warnow, T, Ringe, D. & Evans, S.N. 2004 A Comparison of Phylogenetic Reconstruction Methods on an IE Dataset (original technical report), pp23.

This is similar to the final paper, but includes un-rooted trees.

Ringe, D. & Taylor, A. 2007 (last update) – CPHL Datasets

This includes all the Ringe group datasets used in the post (with the exception of criterion M7, which they dropped). Sadly, they never appear to have made the software available.

 D., Warnow
, T. &
Taylor, A. 
and computational 
Philological Soc. 100,

The main paper discussed in this post.

Additional references

Josephson, F. 2013 How aberrant are divergent Indo-European subgroups? In: Approaches to Measuring Linguistic Differences (L. Borin & A. Saxena eds.) de Gruyter, 83-105.

A good discussion of some of the grammatical points at issue.

Carling, G. et al. 2014 Modelling sound change in relation to time-depth and geography: a case study on the Indo-European and Tupían language families (powerpoint presentation).

An interesting one, talking purely about phonology and attempting to make family trees based on innovations in sound features. They argue that the modern European IE subgroups (Italo-Celtic, Germanic and Balto-Slavic) are at phonological extremes from each other (bold is the most extreme), and from the rest of the Indo-European subgroups, which are all argued to be phonologically relatively conservative.

On this basis, they also group Greek with Italo-Celtic and Tocharian, and Anatolian with the rest (Albanian, Armenian, Indo-Iranian, Balto-Slavic and Germanic).

If I were to make any kind of argument for what’s going on here I would guess at two possibilities for the causes of phonological innovation:

A) it may be the result of hybrid populations changing the sounds of language. So, on this basis a relatively pure migration of a population without admixture might lead to little innovation, but a migration of a dominant minority (although still substantial) might lead to sound changes as pre-existing populations speaking different languages adopt the new language.

B) it may be the result of small, relatively isolated populations showing high mutation rates in the sound of language (a sort of ‘linguistic drift’).

In fact, the two may not be exclusive as both potentially involve relatively small population migrations.

Maybe this would be better as a post.



Proto-Indo-European homelands – ancient genetic clues at last?

November 12, 2017

(Neditors note – You know I now realise that so much of the argument below depends on my belief that elite dominance is not a sufficient mechanism for language change. I prefer to see major intrusion of a new language carrying people as necessary (i.e. that 30% or more of the population will now be […]

Read the full article →

No South American DNA in ancient Easter Island – and…?

November 4, 2017

Regardless of whether Rapanui ever made contact with South America, is the lack of ancient DNA evidence of South American contact in Easter Island that surprising? A very recent paper by Lars Fehren-Schmitz, Pontus Skoglund, and others, based on the ancient DNA (mitochondrial and nuclear) of five Easter Island / Rapanui individuals dated before 1700 […]

Read the full article →

European ancient DNA – the movie

October 1, 2017

We have learned loads from the autosomal DNA analyses of Europe’s ancient populations which have poured out of Harvard and other universities in the last two years. Europe was a restless place, changing people more than some of us would have guessed. Here’s the movie adaptation. I put together the video above for my own […]

Read the full article →

What type of grass are you on, Jared?

July 5, 2017

Does agriculture start with lucky grass, as Jared Diamond says, or does lucky grass start with agriculture. I’ve been sitting enjoying the weather in my garden in Swindon, England. It’s a mess, full of random weeds and overgrown grass. But who cares? The Sun’s out. And there is a particularly fine, architectural grass growing just […]

Read the full article →

Violence in the Near East at the dawn of agriculture: conflict vs co-operation

December 22, 2016

Small population size and little need for competition is an excellent explanation for the lack of conflict during the transition to agriculture. However, there’s also another possible explanation – co-operation. In my previous post I tried to gather together the evidence for violence in the transition from foraging to farming across the Near East. Much […]

Read the full article →

Violence in the Near East at the dawn of agriculture: the evidence

December 10, 2016

Does farming lead to violence, violence lead to farming, or was the period during and after the invention of farming in the Near East as peaceful as it gets? You’ll find many opinions on the role of farming on the origins of violence and of violence on the origins of farming in human history. I’m […]

Read the full article →

Fatal Epidemics of the Bronze Age

October 25, 2016

Fatal epidemics in the Eastern Mediterranean have been going on since at least the start of the Bronze Age. However, the best evidence for big outbreaks is in 18th century and 14th century BC. After about 500 BC historians occasionally took the trouble to chronicle the fatal epidemics they experienced. However, due to a lack of historians […]

Read the full article →

Third Century Rome’s inflation crisis: ‘bad’ leaders or plague?

October 12, 2016

Is the massive inflation seen in the third century Roman Empire a result of reckless spending by megalomaniac emperors, plague or loss of faith? Take your pick. Is late Rome like the modern US? Probably not. There has been a general (though not universal) assumption in studies of the later Roman Empire that increasing militarisation […]

Read the full article →