Essays on the history of writing II: The isolates of the Fertile Crescent

One can often find in the literature the idea that continents like South America or Africa differ fundamentally from Eurasia in terms of historical language expansions, and that one does not find in those continents a similar phenomenon to Indo-European, which covers nearly all of Europe and a considerable part of Asia. As I was thinking about that, and as I was writing the post about Mediterranean isolate languages, I realised something. If Indo-European has been expanding for the last 5,500 years or so, then it was only relatively recently in its history that it came to dominate all of Europe… or almost all, since Basque is still spoken today.

Etruscan was alive and well in (early) Roman times, as were the Rhaetic languages of the Alps and the unclassified languages of the Iberian Peninsula, including Aquitanian (probably related to Basque). Minoan probably continued to be spoken in Crete after the adoption of Greek, developing into the Eteocretan language. Not to mention the Lemnian and Eteocypriot languages (by the way, I completely forgot about the later in that post). And these are just the ones that left some written record. Who knows how many languages were spoken in Northern Europe at the same time? The point is: if we could map the distribution of language families in Europe 2,500 years ago, I am sure the map would be a lot more colourful.

Historical and modern isolated languages and small families in Mesopotamia, Anatolia and the Caucasus.

Of course, as I wrote before, there might have been one or two large families that expanded with the Neolithic in Europe. That is the nature of spread zones. But what about the area where agriculture originated? In line with what we saw in the Americas, areas of ancient cultural developments in the Old World should also be mosaics with high linguistic diversity. In the map above, I am showing some of the unclassified languages, isolated languages or small families of the Fertile Crescent (plus the languages of the Caucasus). Since Crete and Cyprus are still in the range of the map, maybe I should have included Minoan and Eteocypriot, but let’s focus on the Mesopotamia/Anatolia/Caucasus corridor for now.

There is no doubt that the cereals, pulses and animals that were the foundation of the Neolithic in Western Eurasia and Northern Africa were domesticated in this region, most likely in the highland Levant/Anatolia border. Was the spread of farming and pastoralism in the Fertile Crescent accompanied by the diffusion of a single language family, perhaps coinciding with the PPNA/PPNB cultures (or interaction spheres) that dominated much of the region between 11,500 and 8,000 years ago? I do not think so. In later times, most of the area shown in the map above would be covered by only two families: Indo-European (with Hittie and Luwian in Anatolia, Armenian in the Caucasus, and the Indo-Iranian languages to the East, including Persian) and Afro-Asiatic (namely Akkadian in Mesopotamia and all the other Semitic languages spoken from the Levant to Arabia). As in the European case, languages like Hattic, Urartian and Sumerian, that survived for some time while Indo-European and Afro-Asiatic were “conquering” the Near East, could be remnants of previous expansions, possibly since Neolithic times (although there are good Chalcholithic/Bronze Age candidates, like the Maykop culture in the Caucasus and the Ubaid culture in Mesopotamia…). But the fact is that they are too diverse to fit into a hypothetical single family.

Sumerian… and Euphratean?

The Royal Game of Ur. Like writing, board games were another Sumerian invention.

I have written enough about Sumerian in a few previous posts, so I will not elaborate on it anymore except for two crucial questions: the autochtonous origins of this mysterious language and the supposed Euphratean substratum. I have come across an extremely interesting paper that advanced the idea of Sumeria as a creole language born in the multicultural environment of Southern Mesopotamia. Among other things, the argument is that archaic cuneiform signs were constructed through a logic similar to modern creoles. For example, consider the signs sagka and gu7, meaning “head”, “mouth” and “ration” respectively. The last two are formed by slight modifications of the first. Specifically the last one is a combination of “head” and “bread”, and it can be argued that basic Sumerian nouns are very few, most of the others being derived in such way. However, the actual readings of those signs are sagka and gu respectively, so the creole-like derivation of nouns is mostly a phenomenon of the writing system, not of the language itself.

More interesting is the idea that an even earlier language preceded Sumerian(s) in Southern Mesopotamia. Many toponyms do not have a Sumerian etymology, and the phonetic reading of many cuneiform signs is of obscure origins. Most of the signs gained their phonetic values from the rebus principle, but some do not work that way. For example, the sign for “bird” muszen is read ḫu when used phonetically (instead of mušen), and the sign for “fish” ku is read ḫa (instead of ku). Some words, consisting of CV1C(C)V2C structure, do not fit the typical Sumerian structure. This is particularly the case of words for professions, like adgub adgub “reed weaver”, sipad sipad “shepherd” or engar engar “ploughman”. Many refer to farming or herding activities. Thus it is possible that Sumerians borrowed some specialised vocabulary from a society that was already well established in Southern Mesopotamia, but the theory remains speculative, and we know nothing else about this hypothetical language except for the few words in the alleged Sumerian substratum.


Linear Elamite

Spoken to the east of Mesopotamia, Elamite was the language of Susa, capital of a mighty state contemporary with the Sumerians, and later incorporated into the Persian empire. The Elamites got the idea of writing from the Sumerians, but developed their own “archaic” cuneiform signs that we cannot read phonetically (although we know something of the structure and possible content of the texts). Presumably, they recorded the same language as would be later written with the Assyrian syllabary.

“And thus says Darius, the king: Within these lands, whosoever was a friend have I protected…”

We have a large corpus for Elamite, and the language is understood relatively well. That is thanks to a number of bilingual inscriptions, the longest of which is the famous Behistun Inscription, where Darius I, king of Persia, announces his lineage and deeds in three languages: Persian, Elamite and Babylonian. Despite some efforts to connect Elamite with other language families, especially with the Dravidian languages of India, it remains an isolate. There are also interesting parallels with Afro-Asiatic, e.g. elti “eye” and kassu “horn” (although the later seems more like a Wanderwort). George Starostin has a paper with a review of such hypotheses, as well as a comparative 100-word list between Elamite and major language families.

Gutian and Kassite

Much less is known about the languages of two other neighbours of Mesopotamia, the Gutians and the Kassites, both of which inhabited the vicinity of the Zagros Mountains and invaded Mesopotamia to install their own dynasties of rulers. The Gutians reigned for a few generations ca. 2100-2000 BC after the collapse of the Akkadian empire. The names of theirs rulers are virtually all that is known of their language. Despite some comparisons with Tocharian (a very divergent Indo-European language once spoken near the border with China), the names of the Gutian kings mentioned in the Sumerian king list do not reflect any known family in the region. As for the Kassites, they conquered Mesopotamia after 1500 BC, and their language is marginally better attested, since a few Kassite-Akkadian glossaries were compiled by ancient scribes. Needless to say, the known Kassite words do not resemble Sumerian, Elamite, Akkadian, Hurrian or any other language of the region.

Above, some of the Gutian names from the Sumerian king list in the Weld-Blundell prism. Below, part of the Kassite-Akkadian glossary of the Hormuzd Rassam tablet.


“Argishti, son of Menua, built this temple and this fortress, called Irbuni…”

In the 15th century BC, Hurrian was the language of the powerful kingdom of Mitanni to the north of Mesopotamia. Together with the language of their neighbours, the kingdom of Urartu, near modern Armenia, they form the Hurro-Urartian family. Inscriptions in Hurrian and Urartian are written in the Assyrian cuneiform syllabary. They are found, for example, in the Fortress of Erebuni, modern Yerevan, announcing its foundation by the king Argishti I. The inscription names the fortress irbuuni Ir-bu-u-ni, which is in the origin of modern Armenian Երեւան Yerevan, a name that resisted 2800 years!

The most convincing proposed genetic affiliation of Hurro-Urartian is that it would be related to the Caucasian languages. The geographic location of Hurro-Urartian, its phonology, and a few cognates speak in favour of that connection. For example, Hurrian words consist predominantly of a (C)V(C) structure (pa- “to build”, a- “name”, un- “to come”, ar- “to give” etc.) adhering to the typical pattern of the Northern Caucasus. The shortness of the words is compensated by a seemingly complex consonant inventory. Caucasian languages are famous for having very few vowels, but 50 or 60 different consonants (remember this theory linking phonology and climate?). In line with that, it appears that the Assyrian cuneiform syllabary was ill-suited for rendering the Hurrian language. From what we can reconstruct, scribes had to resort to signs like pi pi and ip ip to represent –w-, -v-, -f and combinations thereof. Moreover, the personal pronouns bear a remarkable resemblance with the Northern Caucasian sets. Compare, for instance, Hurrian 1sg. iša-/šo– and 2sg. fe– with Kabardian 1sg. sa, 2sg. wa and 2pl. fa.

Hattic and Kaskian

The supporters of the Anatolian hypothesis of Indo-European origins forget that the Hittites were newcomers to the region. Their predecessors in Anatolia spoke an unrelated language, conventionally called Hattic. Very little is known of the language, as bits of it were only recorded by the Hittites. Evidently, the hypothesis of a Caucasian connection has been proposed by the Russian school. Comparisons with the hypothetical Sino-Caucasian family (including North Caucasian, Sino-Tibetan and Yenisseian) show some interesting cognates, but the validity of Sino-Caucasian is what is disputed in the first place. The Kaskian language later spoken in the northern coast of Anatolia was presumably related to Hattic, and could be the language of the descendants of the first settlers dislodged by the Hittites.

Deep connections? Not quite

As I was first planning this post, I believed there should be some deep relationship between all the “colours” in the map at the beginning of the post. That was the logical conclusion: these were islands of (a) previous expansion(s) later blurred by Indo-European and Afro-Asiatic. Reality is not so simple: let’s have a look at the basic vocabulary of some of those languages:

* The reconstructions in bold are Starostin’s Proto-North-Caucasian. The ones not in bold are reconstructed for the Proto-Northeast-Caucasian level only.

A few isolated resemblances can be found here and there (e.g. “eye” in Elamite and Proto-North-Caucasian, “tongue” in Sumerian and Proto-North-Caucasian), but these might as well be due to chance. The words for “horn” in Elamite, Hattic and Caucasian might be related. To that we must add Proto-Indo-European *k’era(w)- and Proto-Afro-Asiatic *ḳar-. This means the resemblance in the table above is not unique of those languages, and we might in fact be dealing with a Wanderwort.

Perhaps we should not give so much weight to vocabulary. As I explained previously, there are some intriguing morphological similarities between various Eurasian isolates located thousands of kilometres apart. Curiously enough, the same does not apply to this group of relatively close languages in terms of geographical distance. Let’s review the crucial features that distinguish Eurasian isolates in opposition to the large families that surround them: 1. a predominance of prefixes; 2. ergative alignment when marking the pronouns in the verb; 3. possessive pronouns identical to one of the sets used with the verbs; and 4. complex “chains” preceding the verb root. In the table below, I show a quick comparison between the Mesopotamian-Anatolian isolates and Kabardian, a good representative of the Northwest Caucasian family.


Personal pronoun prefixes are marked in red, and suffixes in blue, as usual. Kabardian is the only language that actually conforms perfectly to the aforementioned pattern. Sumerian comes close, especially in relation to the verbal chain, but even it employs a good number of suffixes. Elamite is not even an ergative language, and Hurrian is as prolific in the use of suffixes as a Turkic or Uralic language. Finally, there is almost no resemblance in the actual pronoun particles, except perhaps between Hurrian and the Caucasian languages (more evident with the independent pronouns, as I said above).

What conclusions can we draw from all this? I would like to end with a very simple idea: before the Bronze Age, when large scale warfare – propelled by better weapons but also by the horse – became a major factor in population expansions, the Fertile Crescent was a linguistic mosaic with higher population densities than its surroundings and a long history of ancient cultural innovations, including agriculture. Languages expanded from it, not into it. It is pointless to look for the origins of Indo-European in the first farmers of Anatolia, who spoke Hattic before the Hittites arrived. Neither could the Neolithic peoples of the Levant have spoken Afro-Asiatic, the only Eurasian branch of which (Semitic) having reached the area in relatively recent times. We will never know the language of PPNA, and there might have been many. Perhaps the dwellers of Çatal Höyuk and worshippers at Göbekli Tepe spoke an ancestor of Hattic, or perhaps it was yet another language that contributed to the huge diversity of this ancient cultural mosaic.


Essays on the history of writing: Afro-Asiatic roots in Egyptian

As I have mentioned previously, Afro-Asiatic is an extremely important language family. Not only because it includes two of the earliest languages ever recorded in writing – Egyptian and Akkadian – but also for a reason somewhat related to that: it is old, very old, possibly twice as old as Indo-European (unless you favour the Anatolian hypothesis). Yet, Afro-Asiatic is recognisable as a family and it can be reconstructed, giving hope to the supporters of long-range comparisons in a number of other cases worldwide.

(Modern) extent of the Afro-Asiatic languages and key languages that have ancient written records (expressed in k = thousands of years before present). Berber and South Arabian are rather groups of languages than single languages, but are so named here for convenience.

It could be argued that such recognition is only due to the fact that we have records of some of its branches since nearly five thousand years. Well, Indo-European has also a long (though not as long) tradition of written languages, and the fact that we can compare Latin, Greek and Sanskrit (not to mention Hittite) greatly facilitates the work of reconstruction. It was the similarity between those ancient languages that led to the recognition of Indo-European as a family, but no one would seriously doubt that one could arrive at the same conclusion or that Proto-Indo-European could be reconstructed based solely on the modern languages. As for Afro-Asiatic, the same holds true: even it might have started to split around 10 thousand years ago or even earlier, we would still recognise it as a family solely based on languages spoken today.


The antiquity of Afro-Asiatic cannot be questioned: the Egyptian language is well attested since a bit less than five thousand years ago (or more if you consider Predynastic seals, although they cannot be read with certainty in the way later inscriptions can). Akkadian, the second oldest Afro-Asiatic language ever recorded, starts to be written around 4500 years before present. Yet, the differences between those languages are so large that they must have begun to diverge millennia before that. In the table to the left you can see around 20 words of basic vocabulary in Egyptian and Akkadian. Only two of those (highlighted in red) are cognates, descending from a common root in Proto-Afro-Asiatic (be aware that the words for ‘two’, sn and šina, and for ‘to give’, rdj and nadānu, are not related, despite the superficial resemblance).

The spread of Afro-Asiatic has been thought to be intimately connected with the dispersal of farming, and its age is certainly consistent with that. Interestingly, the modern distribution of the family roughly coincides with the distribution of Y-chromosome haplogroup E1b1b (a.k.a. E-M35), which is also assumed to have arrived in Europe during the Neolithic (though it is not really as frequent in archaeological samples as G2a, and a lot of the presence of E-M35 in the Iberian peninsula could be explained by recent gene flow from North Africa). Recent genetic research has shown that individuals from the earliest culture of settled hunter-gatherers of the Levant, called Natufian, belonged to this haplogroup. Natufians lived around 12500 years ago in the key region of western Eurasia where domestication of a number of cereals and animals has taken place, and they were followed, ca. 11500 years ago, by the first Neolithic cultures of the Near East – samples of which were also found to belong to haplogroup E. On the other hand, the spread of E-M35 could have happened long before the Neolithic, and to assume that its carriers expanded from the Levant to Africa, largely replacing the local populations, simply does not agree with the archaeological record. Furthermore, it would imply a Western Eurasian Urheimat for the Afro-Asiatic family, whereas the highest diversity within the family is undoubtedly found in Africa.

Thus, it is reasonable to situate Proto-Afro-Asiatic somewhere in Eastern Africa around 10 millennia ago or more, but whether its expansion has anything to do with the spread of farming and E-M35 is uncertain. The two latter might be connected in the Levant and in Europe, but I would argue that European Neolithic farmers almost certainly did not speak an Afro-Asiatic language. Rather, it seems that the big expansion of Semitic – the only Afro-Asiatic branch in Eurasia – was, like Indo-European, a Bronze Age phenomenon (more about that in the future…).

P.S.: As I was finishing writing this post, this new paper was published showing a genetic affinity between ancient Egyptians and modern Near Eastern/Anatolian populations, in contrast with modern Egyptians, who have a higher contribution from Sub-Saharan Africa. The individuals that were sequenced for the Y-chromosome belonged, unsurprisingly, to haplogroups J and E1b1b.

In any case, this will be a light, almost recreational post, showing Afro-Asiatic roots that made it into Egyptian and, later, Coptic basic vocabulary. The comparison with Akkadian shown above is somewhat unfair, since many basic Egyptian words actually have cognates in Akkadian and many other Afro-Asiatic languages, even though the semantic correspondence is not exact. We will see how a language family can still be recognised even after ten thousand years (let’s say nine and a half, since Coptic is no longer spoken except as a liturgical language).

NOTE: I offer below a few examples of languages from each branch of Afro-Asiatic. The complete etymologies are available in Sergei Starostin’s site. The Proto-Afro-Asiatic forms (PAA) given at the end of each etymology are taken from the reconstructions of Militarev and Stolbova in that website. When the form originally reconstructed by Orel and Stolbova differs, I note it in parenthesis (OS). Just for the fun of it, some words are accompanied by examples from hieratic papyri, with the respective hieroglyphic equivalents and transliteration.

Man and his Occupationsmanandhis-01

sn sn *san “brother”. In Coptic, ⲥⲟⲛ. A very common Egyptian word, with cognates in quite a few Afro-Asiatic branches. In the Chadic languages, for example, we find Cagu šǝn, Dangla sino, and others. The Cushitic languages have Beja saan, Bilin šan and Gawwada aššinko – the later meaning “nephew”. PAA *san-/sin-.

“Once there were two brothers…” (From the Tale of the Two Brothers, Papyrus D’Orbiney)

Parts of the Human Bodyhuman_body-01-01

an cn “eye”. This is an Old Egyptian word, later replaced by the usual jrt jrt (which also has an Afro-Asiatic etymology). Derived from this obsolete root is the verb an2 cn “to glance”. A number of words with the phonetic combination cn have an_det as a complement. The obvious cognate is the word for “eye” in the Semitic languages: Akkadian akk_iinu īnu, Arabic عين, Hebrew עין, Amharic አይን etc. In the Chadic languages, the cognate for this word appears in the verb “to see”, as in Bole (closely related to Hausa) ‘inn-. Among the Omotic languages, a very divergent branch of Afro-Asiatic, we have Bench an “eye”. PAA *ʕayVn-.

jrt jrt *jārat “eye”. The usual word in Coptic is ⲃⲁⲗ, but ⲉⲓⲁ and ⲉⲓⲉⲣ persisted in prefixes. This has cognates in the word for “eye” among many Chadic languages: Zaar yīr, Musgu arai, Mubi irin etc. Among the Cushitic languages, we find Beja iray- “to see” and Iraqw ara “eye”. PAA *ʔir-.

fnd fnd “nose”. This root is not so well attested in other Afro-Asiatic branches, but it is there. In Chadic, it appears in words for “hole” or “mouth”, as in Sura fuŋ and Pa’a vingi, respectively. In the Cushitic languages, we have a possible cognate in Beja gunuf “nose”. The final d /dʒ/ in Egyptian must have been palatalised from /g/. PAA *fung– (OS *funVg-).

“The nose is blocked, it cannot breathe…” (from the Precepts of Ptah-Hotep, Papyrus Prisse)

ns ns *nīs “tongue”. In Coptic, ⲗⲁⲥ. This is a widespread root in Afro-Asiatic. In Semitic, as noted in the table in the beginning of the post, we have Akkadian akk_lisaanu lišānu and, among modern languages, Arabic لسان, Hebrew לשון, Amharic መላስ etc. Berber languages have ilǝs or ils. In the Chadic branch, we have cognates in many languages, as in Angas leus, Musgu εlεsi, Dangla lēse etc. Egyptian did not have an independent glyph for /l/, although this phoneme must have been present, given the evidence from Coptic. Cognates of words with /l/ in other Afro-Asiatic languages were written with n, r or j in Egyptian. PAA *lis– (OS *les-).

tst ts “tooth”. There are not many cognates for this word, evidence being mostly limited to the Cushitic branch, with Beja koos “tooth” and Qwadza koʔosiko “molar”. As in the case for “nose”, the initial t /tʃ/ must have been palatalised from the original /k/. PAA *kV(ʔ)Vs– (OS *kos-).

jb jb *jib “heart”. This is a beautiful etymology, with cognates in many branches of Afro-Asiatic. The Semitic languages are an obvious example, with Akkadian akk_libbu libbu, Arabic لب, Hebrew לב, Amharic ልብ etc. In the Chadic languages, this root is represented by forms like Kilba libibi, Musgoy lib (meaning “belly”) and Mokilko ʔulbo. The Cushitic branch has Afar lubbi, Somali laab and Sidamo lubbo – the later meaning “soul”. Finally, Omotic (this very divergent branch of Afro-Asiatic!) can be included in this etymology, with Anfillo yiboo “heart”. PAA *libb-/lubb– (OS *lib-/lub-).

“The good conduct of his heart and his tongue…” (from the Precepts of Ptah-Hotep, Papyrus Prisse)

qsw qs “bone”. In Coptic, ⲕⲁⲥ. This word is not well attested in Semitic, with few examples like Arabic قص “chest (bone”). In the Berber languages, however, we find a good cognate in the usual word for “bone”, iɣǝs or similar variants. The Chadic languages are also well represented, with Hausa k’ašii, Musgu kεskεε, Dangla kaaso and many others, all meaning “bone”. In Cushitic, a few examples exist, like Warazi mik’eče. Finally, among the Omotic languages, we may cite Nao k’us. In summary, this word has cognates in all Afro-Asiatic branches. PAA *ḳ(ʷ)as– (OS *ḳaċ-).

Sky, Earth, Waterskyearth-01

mw mw *māw “water”. In Coptic, ⲙⲟⲟⲩ. This is, of course, a very basic word, and it has many obvious cognates in other Afro-Asiatic languages. In fact, words for “water” with the general form mV– are widespread in other families, as exemplified by a famous Nostratic etymology. But let’s stick to Afro-Asiatic. Curiously, it is not attested in many branches. Among the Semitic languages, we have Akkadian akk_mu , Arabic ماء, Hebrew מים etc. Cognates in the Chadic languages are not many: we have Guruntum ma, Gude maʔine, among a few others. In the Cushitic languages, cognates of this word appear in Beja muʔ “liquid”, Iraqw maʔay and Dahalo maʔa.  PAA *maʔ-.

“She did not pour water into his hands…” (from the Tale of the Two Brothers, Papyrus D’Orbiney)

nfw nf “breath, wind”. In Coptic, ⲛⲓϥⲉ. Perhaps this should have been included among the parts of the human body, especially given the meaning this root acquired in some Afro-Asiatic branches. The cognates of Egyptian nf mean “nose” in the Semitic languages: akk_appu appu, Arabic أنف, Hebrew אף etc. In Chadic languages, it came to mean “to breathe” or “life”, as in Daba nip and Tera nifi respectively. The Cushitic languages also include similar semantic shifts. Thus, we have Beja nifi “to blow”, Saho naf “to breathe; soul”, Afar neef “face”, and Somali naf “soul, life”. Finally, the Omotic languages can be included in this etymology, e.g. Kafa naf “to blow, swell”. PAA *(ʔa)naf-/(ʔa)nif– (OS *naf-).

jmnt jmnt “west”. In Coptic, ⲉⲙⲛⲧ. The West is not just a cardinal direction in Egyptian: it is also the land of the dead, where Osiris reigns. This word also happens to have an interesting Afro-Asiatic etymology. In the Semitic languages, the cognates mean “right” or “right hand”: Akkadian akk_imnu imnu, Arabic يمنى, Hebrew ימין etc. In Hausa, the most widely spoken Chadic language, the meaning is closer to Egyptian: yammaa “westward”. The trick is to think that, if you have the South, not the North, as your reference, then the West will be at your right hand side! PAA *yamin-.


km km “black”. In Coptic, ⲕⲁⲙⲉ. This is an important word, since it is the root of the name for Egypt itself in the ancient language: kmt kmt *kūmat (ⲕⲏⲙⲉ in Coptic), the “black land”, a reference to the fertile soil in the Nile floodplain in contrast with the surrounding desert. Unfortunately, it does not have many cognates in other Afro-Asiatic languages. Evidence is limited to the Chadic and Cushitic languages. In the Chadic branch, semantics have changed a little bit, as in Buduma kaimē “shadow” and Awiya kəmən “evening”. In the Cushitic languages, correspondence is more obvious, with Gawwada kumma “black”. PAA *kum-.

“And the stream carried it to Egypt…” (from the Tale of the Two Brothers, Papyrus D’Orbiney)

wad w3d *wāʀid “green”. In Coptic, ⲟⲩⲱⲧ. This etymology is somewhat ambiguous, but I will assume that the best cognates are in the words for “green” in the Semitic and Berber languages. In the Semitic branch, we have Akkadian akk_warqu warqu, Hebrew ירוק etc. Among the Berber languages, we have awraɣirwaɣ and related forms, sometimes meaning “yellow”. Presumably, as in a few other roots listed above, there was a palatalisation from q > d in Egyptian. PAA *wVraḳ– (OS *wVriḳ-).

qbb qbb *qabab “cool” (actually a verb, “to be cool”). In Coptic, ⲕⲃⲟ. The only cognates are found in the Cushitic languages, as in Somali qabow “cold”. Nevertheless, I thought I should include it! PAA *ḳab-.

Verbs verbs-01

mt mwt “to die”. In Coptic, ⲙⲟⲩ. Unlike the previous one, this is a very common and widespread root – and “to die” seems like an appropriate way to finish this list! In the Semitic languages, cognates include Akkadian akk_maatu mātu, Arabic مات, Hebrew מת, Amharic ሞተ etc. Berber, on its turn, has əmmət. Among the Chadic languages, we can cite Hausa mutu, Buduma matte, Dangla mate, and many others. Cushitic languages also have cognates for this root, as exemplified by Somali mōd “death”. PAA *mawVt– (OS *mawut-).

“Death is reached…” (from the Precepts of Ptah-Hotep, Papyrus Prisse)

Spread Zones: Europe and the Neolithic Survivors

I have been away for too long, due to other writing activities that kept me extremely busy. In spite of that long gap, this will not be a new cycle of posts, but the end (at least momentarily) of a series of previous topics. I will, however, touch on a variety of subjects and express my opinion about the most contentious question of European linguistics and archaeology – the timing of the spread of Indo-European languages. My opinion on this problem has been decided (for now) by the evidence provided by Basque and the isolated languages of the Mediterranean.

From the now classic Guns, Germs and Steel to the more recent Prisoners of Geography, the idea that the terrain inhabited by a group largely determines the events of their history has become widespread among the general public. Although reality is much more intricate than that, the same principle is the basis of mosaic x spread zone models: large navigable rivers and extensive flat plains will tend to experience wave over wave of language replacement. In contrast, mountainous areas, islands and other inaccessible terrains tend to be refugia where millennia of uninterrupted development result in a myriad of languages disconnected from large families. Obviously there are exceptions – as in the Andean case that I examined previously.

In any case, it is undeniable (as Russia well knows) that from the Asian steppe to Central Europe there is really no geographical barrier – as long as one keeps south of the Urals. The diffusion of Indo-European languages into Europe did not need to overcome all that distance, moving only from the surroundings of the Caspian Sea to the West (assuming a Yamna origin!). Curiously, the Neolithic diffusion followed a different path altogether, from the Levant, through Anatolia, to the Mediterranean, resembling part of the (much later) silk road. The big question is whether these two events (Neolithic and Indo-European) coincide, as proposed originally by Colin Renfrew. I assumed this theory to be pretty much dead, at least among linguists, but I was mistaken (see Gray and Atkinson’s paper and, more recently, Bouckaert and colleagues’). Fortunately, genetics have been playing a major role in redefining our views about ancient migrations, and I became convinced that genetic evidence does not support a Neolithic age for Indo-European. Nevertheless, given the difficult association between genes and languages, I am sure the matter will continue to be hotly debated for years to come.

Map of the 5th principal component of 94 genes in Europe (from History and Geography of Human Genes)

The Pre-Indo-European languages of Europe (and their speakers) offer formidable clues to the problem. I have written about the isolates of Eurasia in a previous post, but let us explore some more facts about the last speakers of a Pre-Indo-European language: the Basques. Even before the modern DNA studies, the Basques were known to be different from their neighbours. For example, using only blood types, it was noticed that Basques were predominantly O and had the highest incidence of Rh- in Europe. With the first DNA analyses of a large number of European populations, it became clear that Basques were indeed genetically distinct. I have previously shown some maps of principal component analyses, and here I reproduce the map of the 5th principal component for Europe, based on the 94 genes analysed by Cavalli-Sforza and colleagues. This principal component peaks at the Basque country, which is at the opposite extreme from most of Northern/Central Europe and the Balkans, although with some similarity to the remainder of the Iberian Peninsula.

Keeping in mind that the first principal component shows a gradient from Greece to the northwest (Neolithic?) and the second principal component radiates from the north of the Black Sea (Bronze Age/Yamna?), I believe there are only two ways Cavalli-Sforza’s data can be interpreted:

  1. The Basques are Paleolithic/Mesolithic “survivors” (ergo Neolithic migrants spread the Indo-European languages);
  2. They are Neolithic “survivors” (ergo Indo-European languages arrived during the Bronze Age).

By the way, I am assuming here that the massive genetic legacy of the Neolithic and Bronze Age expansions (see below) must have had a linguistic correlate. In theory, one can imagine a situation in which large numbers of migrants arriving at a region, becoming culturally dominant and having children with locals do not imply language replacement, but I would like to see real world examples of that. In fact, it seems that one only needs a small number of “conquering” migrants with minimal genetic impact to change the language of whole regions (that is the case in some Latin American countries and, further back in time, was the case of Hungary).

Fortunately, we have advanced much since Cavalli-Sforza and colleagues’ original work. For example, it is clear now that the Neolithic expansion in Europe did involve population movement, and that migrations from Anatolia were indeed the source. In my opinion, the most significant genetic piece of evidence is the discovery that modern Basques are the closest living population to Neolithic skeletons from the same region. Sardinians were also found to be very close and, in fact, Sardinians appear in every study cited here as the closest modern match to Neolithic DNA samples. Their status as a “relic” population in Europe due to isolation was noticed long ago, when Cavalli-Sforza left Sardinians out of the PCA due to their singularity within Europe. The genetic similarity between Sardinians/Basques and Neolithic samples deserves special attention.


One important aspect that has been taken into consideration recently is the distribution of haplogroups in Europe. Unlike the autosomal data referred to above, which tells us about the admixture in the ancestry of an individual, Y-chromosome and mtDNA haplogroups are specific mutations that are passed down over generations and preserve the histories of migrations of particular male and female lineages. The Y-chromosome haplogroups most closely associated with the Neolithic are E1b1b and G2a. Both of them are rare in modern Europe as a whole, but G2a was the dominant haplogroup among Neolithic farmers of Central Europe, France and Spain. Both E1b and G originate outside of Europe, in the Levant and the Caucasus respectively, and must have been brought to Europe by Neolithic migrants. I will not reproduce the beautiful maps from Eupedia, but in the map above I highlight the areas where E1b and G are nowadays more common. The Mediterranean (including Sardinia) and the Balkans are modern refugia of those two haplogroups, but what happened to Central Europe? In fact, paternal lineages from most of Central and Western Europe were later replaced by haplogroups R1a and R1b, carried by Bronze Age migrants (confirmed by the fact that Yamna samples were recently shown to be R1a). Surprisingly, most modern Basques actually belong to haplogroup R1b (maybe “bottleneck effects” could easily lead to the replacement of Y-chromosome lineages over a few generations in such an isolated population?).

In summary, the Neolithic expansion in Europe involved considerable population movements, the genetic signature of which can still be seen and is most noticeable in the Mediterranean and the Balkans. These areas were somewhat less impacted by the later Bronze Age migrations, which also changed considerably the genetic make-up of Central and Western Europe. The fact is that, since there is no other major language expansion after Indo-European, this event must have occurred during the Bronze Age. Some supporters of the Anatolian hypothesis do not deny that fact, arguing that several waves of Indo-European expansion could have occurred. Using the principle of Occam’s razor, I would immediately discard this explanation as being too complicated. The question then is: would the Neolithic expansion also have involved the diffusion of a single, widespread language family?

Pre-Indo-European languages in relation to the major cultural traditions of the Neolithic.

In the map above, I am showing some of the main Neolithic cultures of Europe in relation to the isolated languages that we know about. The Linearbandkeramik (LBK) has taken advantage of the plains that extend from Ukraine to France – a huge spread zone – and it is difficult not to imagine that it was accompanied by the diffusion of a single language [family] around 7000 years ago. At the same time, the Cardial/Impressed culture spread along the Mediterranean shores. Both ultimately stem from the Greek Neolithic (with clear origins in Anatolia), but seem to be local developments. Whether both involved the expansion of the same language [family], I will not dare to speculate, as there is no clue to what the LBK people spoke (perhaps it is in the substrate of the Germanic languages). But could there be a connection between the other Pre-Indo-European languages recorded in Southern Europe?

That Basque was part of a larger language family in the past is well accepted, but what was its extension? Some have suggested the widespread occurrence of Basque-like elements in the toponymia of Europe: e.g. the connection between Val D’Aran in Spain, Arundel in England and Ahrntal in the Alps would be the Basque word aran “valley”. There have also been some attempts to connect Basque to the extinct Pre-Indo-European language of Sardinia – which we can call “Paleo-Sardinian” but is also known as “Nuraghian”. Beyond genetic isolation, the linguistic isolation of Sardinia is clear even in (relatively) recent times: for example, whereas all Romance languages have turned Classical Latin C into /tʃ/, /ʃ/ or /s/, in Sardinian it is still pronounce as a hard /k/. As for the Paleo-Sardinian language, we have no record of it except for words that entered modern Sardinian or the toponyms in the island. It is based on those that Blasco Ferrer proposes parallels with Basque – e.g. the triad Lur-beltz, Lur-gorri and Lur-zuri meaning “black”, “red” and “white earth” respectively, which appears in the island as Duru-nele, Lúr-kuri and Lu-tzurró.

Among the well-known Pre-Indo-European isolates is the Etruscan language. There is a possibility that it is actually connected to Rhaetic, a language or group of languages preserved on a few inscriptions around the Alps. Given the small corpus, it is unlikely that it will ever be “deciphered”, but formally there are some resemblances with Etruscan – e.g. a common ending -ce or -ke that, in the later, marks the past tense. Another candidate to form a family together with Etruscan is Lemnian, attested in a few inscriptions in the Greek island of Lemnos. If Etruscan, Rhaetic and Lemnian are indeed part of a single family, later fragmented by the spread of Indo-European, then we are potentially dealing with yet another group of Neolithic “relics”. Then, the location of those languages in relation to the Cardial/Impressed culture and to the modern distribution of Y-chromosome haplogroups E1b1b and G2a, as can be seen in the maps above, starts to make sense.

Finally, let us consider the Pre-Indo-European languages (directly or indirectly attested) from the region closest to the origins of the Neolithic. One of the most interesting things about the Greek language is that it absorbed a large amount of non-Indo-European vocabulary. These are wοrds that have no cognates in other Indo-European languages and are also easy to spot based on their distinctive phonology/morphology. Interestingly, they tend to be cultural items – words like σῦκον “fig”, ἔλαιον “olive”, θάλασσα “sea” and βασιλεύς “king” (in Linear B qa-si-re-u). It is not impossible that, like the Linear B syllabary, the substratum in Greek might be related to the language once spoken in Crete and partly preserved in the Linear A script. This language, which we may call Minoan, is probably never going to be fully deciphered given the small size of the corpus and the fact that most of it consists of accounting tablets full of personal names and toponyms. Names of products are written with logograms, so it is impossible to know how they were pronounced. A few transaction words (see below) and inflected forms (like the famous ja-sa-sa-ra-me that I mentioned previously) offer a window into Minoan, but the known vocabulary is so small that the language is destined to remain unclassified.

In summary, it is likely that all these languages are remnants of a once widespread Neolithic family (or families), diffused together with the Cardial/Impressed culture along the Mediterranean. We simply do not have enough material to show how they are related, although a convincing case can be made for Etruscan, Lemnian and Rhaetic. Basque could be related to this phenomenon or be part of a different family spread further west, perhaps in association with the Megalithic traditions of the Atlantic Neolithic. It is almost certain that the LBK expansion brought a single language or language family to their vast territory, but whether it was also related to the Mediterranean variants is impossible to tell.

Example of a Linear A tablet, now at the museum of Heraklion. Logograms are transcribed with latin words (vir for man, fic for fig). Most of the words spelled with the syllabary are personal and place names. In this tablet, you can see some of the exceptions: ki-ki-na seems to be an adjective describing the figs, or maybe it is the word for fig itself repeated after the logogram. Transaction words usually appear as headers or at the end of the tablet. In this example, a-du and ki-ro could mean something like “balance”. Ku-ro is followed by a number that is the sum of all previous items and must mean “total”.

Spread Zones: South America

As the reader might have guessed by now, one of the theories that I find most attractive in linguistic geography is the opposition between spread and mosaic zones. In the form developed by Nichols, this theory posits that some areas – in general, broad plains with few natural barriers – are prone to the dispersal of single languages or families, whereas other regions (islands, mountain chains) tend to accumulate isolates and small families. Not only specific languages stretch over the spread zones, but also linguistic features. Just look at Eurasia and you will see layer over layer of prehistoric (and more recent) spreads. There is a long East-West axis that served as a corridor for the Indo-European, Uralic, Turkic and Mongolic languages, pretty much along the same steppe that brought Attila and Genghis Khan to the doors of Europe. All the language families in this area also share similar features – as opposed, for example, to the languages of the Caucasus or to the Sino-Tibetan family. The big question is: what are the archaeological correlates of a spread zone? The answer is in the horizon!

Archaeological horizons are geographically extensive phenomena – a ceramic style, a particular form of architecture, or iconography – whose spread happened very fast but whose persistence was often relatively short-lived. If we are trying to identify the periods when some large language families reached their territorial zenith, then we must look for archaeological horizons rather than for very localised cultures. Moreover, it is likely that cultural horizons they corresponded to linguistic spread zones, whereas mosaic zones should be characterised by a myriad of local traditions. Curiously, in South America, the spread zones are very different from the Eurasian case – partly due to the absence of extensive grassy plains with no obstacle to movement (the Argentinian pampas being the obvious exception, although they are not as important a communication route as the Eurasian steppe that connects the continent’s long East-West axis). Instead, the South American spread zones are large river systems (the Amazon and the La Plata Basins) and the Andean cordillera. Yes, mountains are spread zones in South America!

To the left, linguistic mosaic zones (green) and spread zones (pink) in South America. To the right, the major archaeological horizons and some localised cultures.

On the Andean side of the debate, the most important question pertains to when the largest language families in the region – Quechua and Aymara – were diffused. Once thought to be closely related in a “Quechumaran” family, the resemblance between the two is now known to be the result of extensive contact. Quechua is much more widespread than Aymara, partly because it was adopted as the language of the Inca empire (although it was already used over a broad territory even before that), but both functioned as linguae francae in different parts of the Central Andes. As correctly pointed out by Paul Heggarty, Quechua and Aymara must have spread during the Early and Middle Horizons of Andean Prehistory (the Late Horizon corresponds to the Inca expansion, but Quechua was already  widespread by then). These are periods when shared architecture and/or iconography diffused throughout the coast and the cordillera – presumably, hand in hand with new ideas and, of course, languages. The Andean Horizons are separated by periods of fragmentation, called the intermediate periods, when local cultures flourished.

How can a mountain chain like the Andes have served as a spread zone for languages in the past? One possible explanation is that the population in the whole “spine” of South America has always been connected through river valleys that often link the highlands to the coast. Some of them are extensive, like the Callejón de Huaylas (near Chavín), a major trade route throughout the Andean prehistory.

The Early Horizon, around 900 B.C., is the time of Chavín de Huantar – a labyrinthic pilgrimage ceremonial centre in the highlands once thought to be the “mother culture” of the Andes, but now known to be just one manifestation of a broadly shared tradition. For Heggarty, this is when Aymara started to spread. The reasoning behind that is that there are “islands” of Aymara, as well as place names, in what is now a “sea” of Quechua (including in the region around Chavín), i.e. Aymara must have spread first. I have some doubts about this hypothesis, especially because the family seems a bit shallow (the two other languages, Jaqi and Aru, spoken in central Peru, are not that different from the southern Aymara varieties). However, I admit that it is the most parsimonious explanation. This leaves us with the Middle Horizon as the time when Quechua first expanded. Starting around A.D. 600, the Middle Horizon is the time when two major centres made their influence felt over a wide sphere in the Andes: to the North, Wari, and to the South, Tiwanaku. Although the latter is usually associated with Aymara in most people’s imagination, we know that this language only arrived recently in the region. Could the original one have been Quechua? The linguistic diversity in this area points to other possibilities, including Uru, Chipaya and Puquina. Finally, with the onset of the Late Horizon by A.D. 1450, the Incas continued to propel the expansion of Quechua, as did the Spanish after the conquest. What is important to keep in mind is that there are clear correlates of a linguistic spread zone in the archaeological horizons of the Andes.

The famous Shipibo art style looks like a direct descendent of the Amazon Polychrome Tradition. But does that tell us anything about languages in the past?

Let’s move now to the East, to the lowlands of South America. Here, there are two related phenomena whose continental scale grants them the status of archaeological horizons: in the Amazon river, the Polychrome Tradition; in the other large rivers and Atlantic coast, the (distantly) related TupiGuarani Tradition. Starting around 500 B.C., the TupiGuarani Tradition spread from southwestern Amazonia, bringing with it virtually identically decorated ceramics, among other traits. The distribution of the sites and their chronology coincide almost perfectly with the historical extent and depth of the TupiGuarani language family. Thus, unlike in the Andean case, here we know exactly with which languages to associate an archaeological culture. The issue with the Amazon Polychrome Tradition is not so simple. Starting around A.D. 1000 or maybe a few centuries earlier, this tradition rapidly spread over the basin. Some suggest it might also be associated with TupiGuarani languages, based on the fact that the historical Kokama and Omagua, who spoke languages of that family, were still producing similar designs during colonial times. I would really like that to be true, as it would confirm the Spread Zone nature of the main South American waterway. However, there are other groups producing similar pottery until the present, like the Shipibo-Conibo, who are Pano-speakers. Of course, the adoption of the pan-Amazonian polychrome designs does not necessarily imply the diffusion of a single language. However, because the Kokama and Omagua languages apparently result from an ancient TupiGuarani-based lingua franca, I would take that as a serious hypothesis.


Following on the tradition of showing a little bit of the languages in question, the examples above offer a brief comparison of some widely spaced Quechua and TupiGuarani languages. I have selected sentences with similar structures, though I could hardly find identical ones. The Quechua of Pastaza and Napo, spoken in the Ecuadorian Amazon, and the Quechua of Santiago del Estero, spoken in northwestern Argentina, are separated by over 3,000 km. Kayabi, spoken near the Xingu river, southern Amazon, and Mbyá, spoken in southern Brazil and neighbouring Argentina and Paraguay, are separated by nearly 2,000 km. Yet, the similarities are striking. Moreover, although the two families are completely distinct in terms of morphology (as expected form the west-east divide in South America), they tend to have simpler phoneme inventories and “easier”, more regular structures than many languages that remained in the mosaic zones (see a few examples of isolates and other South American families here). Quechua, in particular, has a regularity reminiscent of an artificial language. I would tentatively suggest that this is a characteristic of languages that succeed in disseminating over the spread zones.

What about the archaeological correlates of linguistic mosaic zones? As I said in a previous post, these tend to be areas rich in localised archaeological cultures. I have highlighted a few of them in the map at the beginning of this post. North- and southwestern Amazon are typical mosaic zones. Recently, the southern fringe of the Amazon has been shown to hide a myriad of archaeological sites with earthworks, ditches, enclosures and roads. Among these are the Geoglyphs of the state of Acre and the “Garden cities” of the Upper Xingu. The Llanos de Mojos, about which I wrote previously, is another example. However, these are very different archaeological cultures, and even within one of these regions there is enormous variety. Therefore, they are perfect correlates of linguistic mosaic zones: there is not a widespread ceramic style or other material trait over these areas, which is matched by their linguistic diversity.

Satellite imagery with added plans of a Geoglyph site (left) and one of the “garden cities” of the Upper Xingu (right).

A genetic detour

This will be a short “detour” about a topic that might have crossed some readers’ minds after the latest American series of posts: can external evidence help decide which linguistic classification is more likely? Genetics is a discipline that often uses linguistic data (in the case of the Americas, it is usually Greenberg’s classification). Of course, a straightforward relationship between genes and languages would imply that peoples’ movement is the only mechanism responsible for language change, which is definitely not the case. However, it can offer insights – after all, if one can prove that some people did move, well, that must have had some linguistic consequence. I have here and there touched on some of these issues, but now I want to explore the genetic data more seriously.

About twenty years ago, Luca Cavalli-Sforza and other Italian geneticists published the venerable History and Geography of Human Genes, summarising decades of research all over the world. Beyond the presentation of phylogenetic trees for all human populations and a summary of the history of human colonisation of the globe, the book offered an innovative perspective with the use of principal component analysis (PCA) to shed light on finer scale movements within continents. While we could analyse map after map of the distribution of individual genes, what PCA does is to summarise the main trends in all that variability into just a few maps (for example, if there are several genes that show more or less the same distribution, they can be transformed into a single component).

Some of the principal component (PC) maps of Europe correlate with known or hypothetical migrations in the past, and a lot of them are repeated in the more popular Genes, peoples and languages. For example, Cavalli-Sforza found that the first PC – which explains 28% of the total variance – peaked in the near east and decreased towards north-western Europe. What better proof that the Neolithic expansion involved migrations, not just diffusion of ideas? Such results have now been confirmed with actual archaeological DNA from the Neolithic settlers. The third PC, which explains about 10% of the variance in Europe, peaks north of the Black Sea and decreases towards the west. This can be seen as proof that the spread of Bronze Age cultures such as the Beaker and Corded Ware, thought to originate from the Yamnaya of the Russian steppe, again involved a wave of migrants. DNA from ancient skeletons has once more confirmed those results. The tricky part is that both population waves could be associated with the spread of Indo-European languages, so it becomes difficult to decide between Renfrew and Anthony.

Well, I have an answer to that question (I will explain one day, but here is a clue: the Basques are the key). The important thing to keep in mind is that genes denounce when migrations took place or not, and thus have immense potential to unravel historical linguistic questions. Let us have a look now at the Americas – they turn out to be quite homogeneous (in comparison to Eurasia) when it comes to genetics – and therein lies the problem.

 A forest of genes in South America

Phylogenetic (or neighbour-joining) trees are an intuitive way of showing how populations branched out based on their genetic distances. Because new analyses are conducted all the time and new results are constantly being published, we often don’t have a single tree, but a forest of possible phylogenies! For the Americas, they can vary a lot from the earliest  to the latest research. Because the trees are usually colour-coded according to linguistic phyla (taken from Greenberg) in almost all publications, let’s do the same thing here and see how well my own classification fits the genetic data.

sforzaThe first tree shown here was taken from The history and geography…, first published in 1994. It is based on 60 to 70 markers. Cavalli-Sforza and his colleagues present different trees depending on whether North or South America is included (I have used the second). One of the things we can immediately see is that there is no correlation between proposed language phyla and genetic proximity. In fact, there’s no correlation with well-established, smaller families either. The Parakanã, a Tupi population of eastern Amazonia, is the first to split, instead of grouping with the other Tupi-speakers. The Mapuche are almost in an isolated branch instead of appearing close to other Andean and Patagonian peoples. Even worst, the Quechua are shown as being closer to the Maya than to their Aymara neighbours! I believe the problem in this tree is the intrinsic homogeneity of South American populations, coupled with the small number of markers analysed. However, when the whole continent is considered (and groups from the same language family are averaged), the tree is insightful: 1. the Eskimo are closer to Siberian populations; 2. the Na-Dené from the Northwest coast (but not those further south, like the Navajo and Apache) are a bit more distant, but in the same cluster; 3. all the other populations form a separate branch. Needless to say, this was seen as the confirmation of the theory of three waves of migrations – one of them originating the Na-Dene languages, whose distinctiveness and purported links to Siberia are still given serious consideration.

wangNow let’s examine a more recent tree. This one was produced by Wang and colleagues in 2007 and takes into consideration over 600 markers. I am only showing the South American portion of the tree. Overall, the conclusions of this study do not support three waves of migrants, but a single founding population: all Native American peoples grouped closer to each other than to the Siberians. As for linguistic correlates, it seems that a little more structure is emerging: Andean-Patagonian speakers are the first to branch, Quechua and Aymara groups being particularly close genetically. Strangely, some Central and North American populations appear in the following branch. Finally, we have a cluster for Chibcha speakers and another for my purported Chaco-Amazonian phylum. Arawak speakers are distributed across both, something I will comment on ahead. Two things worthy of notice here: 1. as suggested by Wang’s study, the outlier position of Andean-Patagonian might indicate a Pacific coast colonisation route for South America; 2. Chibcha-speakers are closer to Amazonian populations, which would invalidate my claims that it belongs on the western group (assuming that genetic distance equals linguistic distance).

reichAnd now for the most recent tree, one published in 2012 by Reich and colleagues in Nature. This time, over 300,000 markers were used. This impressive study confirms some of the early conclusions of Cavalli-Sforza about the continent as a whole, but presents a different picture of South America. Eskimos appear closer to Siberian populations than to the rest of the Americas; Na-Dené speakers, however, are not clustered with them, but form a highly divergent branch in the American side of the tree. The authors support the idea of at least three gene flows from Siberia to the Americas, the last two originating the Eskimo and Na-Dené groups. As for South America, we see a similar structure as to the previous tree, but with significant changes: 1. now all the Central American groups are on a branch of their own, instead of mixed with South America; 2. Andean-Patagonian speakers are closer to Chaco-Amazonian ones, with Chibcha splitting earlier than them. In any case, if this is reflected in the language phyla, maybe Chibcha shouldn’t be part either of the western or the eastern South American branches, but maybe as a distinct branch connecting South and Central/North America. Arawak speakers in the latest tree appear in different branches, and that is the only family that never shows a clear genetic correlate. I think this is extremely interesting, as it may confirm the suggestion that Arawak languages spread through trade rather than migration.

Why are the trees so different in this “genetic forest”? Well, maybe it’s difficult to arrive at a consensus because Amerind populations are genetically so close. For example, most people know that nearly all Native Americans, with the exception of some North American groups, belong to blood type O. This is because of a population bottleneck, as not many people crossed Beringia to populate the New World – and that happened relatively recently, so there hasn’t been much time for internal drift. The estimates for the founder population are usually not more than a few hundred persons, and even if the initial separation of this population from the other Siberians might have occurred some 20-30 thousand years ago, the “bottleneck” for the colonisation of the Americas proper is estimated on genetic grounds to be somewhere between 15 and 18 thousand years (with later migrations of the Eskimo and, possibly, the Na-Dené speakers). Moreover, Wang and colleagues find that a coastal route of colonisation of the Americas explains certain variance in the data better, a conclusion that is reinforced by the fact that ice sheets were blocking the inland route in North America before 13,000 years ago. A quick advance of the first settlers along the Pacific is in agreement with the early dates of some archaeological sites like Monte Verde in Chile. Needless to say, it also explains the linguistic divide between the western and eastern parts of the continent.

Maps of the principal components 1, 2 and 3 in Cavalli-Sforza, Menozzi and Piazza’s analysis of 66 Native American genetic markers. These three PCs explain around 53% of the total genetic variability in the continent. Gradient is from red to blue in all cases, but the direction is not particularly important.

Let’s now have a quick look at the maps for the principal components in the Americas. The first PC has a very regular North-South gradient in North America. Cavalli-Sforza and colleagues interpret it as summarising the main differences between the Eskimo and (Canadian) Na-Dené, on the one hand, and all other Native Americans, on the other. In South America, there is little variation in this PC, but it does show a West-East gradient. Maybe that is related to the colonisation along the Pacific coast, coupled with a relative isolation between the highland and the lowland groups. The second PC shows several trends. In North America, it summarises again the difference between the northernmost groups (Eskimo and NW Na-Dené) and the remaining peoples. A little dot in the U.S. Southwest may relate to the local Na-Dené speakers, the Apache and Navajo. The peak in the American northeast is interpreted by Cavalli-Sforza as relating to European admixture (which is documented by other means). In South America, I believe this PC shows essentially the same trend as the previous one, but in more detail: the Central Andes, together with the northern coast of South America, are in one extreme, whereas the region of the lower Amazon is in the other. In my view, this shows perfectly the colonisation of South America along the Pacific coast, another entry along the Atlantic, and the genetic isolation between Andean and Amazonian populations. Finally, the third PC is a bit more difficult to interpret. It forms a mainly West-East gradient. In North America, the Eskimo and Na-Dené are again well differentiated, including the outlying position of the Apache and Navajo (in the little orange dot). In South America, there is a peak in the Mapuche area of Chile, decreasing towards the Guyanas. Cavalli-Sforza sees that as evidence of African admixture in the latter. Possibly this PC resumes the distinction between (southern) Andean and Patagonian peoples on one side, and the remaining groups on the other.

Amerind from a South American perspective: part III

It is time to examine some of the vocabulary that, I believe, provides good evidence for the deep relationship of all South American language families. I have amassed this dataset over the last months and, although I am convinced of its plausibility, I will not passionately defend the validity of these etymologies. Although the items that follow pass the important criteria of being basic vocabulary (less prone to borrowing) compared between proto-languages (not picked and chosen between hundreds of individual languages), the sound correspondences are not completely regular – even though I have paid attention to regularity as much as possible. One must notice that violation of perfectly regular correspondences is not always a reason to discard cognates in some well-established language families. I will not attempt reconstructions, but entries in the AED which are more or less equivalent to the etymologies below will be noticed.


And. = Andean; Ama. = Amazonian; PQ = Proto-Quechua; PAy = Proto-Aymara; Kun. = Kunza; Map. = Mapudungun; PTp = Proto-Tupi; PMJ = Proto-Macro-Jê; PK = Proto-Karib; PP = Proto-Pano; PAr = Proto-Arawak; PTk = Proto-Tukano; PG = Proto-Guaykuru; All. = Allentiac.

Body parts

1. EYE : And. PQ *ɲawi, PAy *najra; Ama. PMJ *nʌm, PK *ənu-ru, All. new. || AED 242 *(i)to(ʔ), 243 *tene ~ *tele. || Comments: PK also has *əne ‘to see’, which can derive from the same root. The reconstruction of PMJ would usually have *nd-. Such initial clusters, which also appear in proto-Tupi reconstructions (e.g. *np-), are an artefact of the phonology of modern Amazonian languages. These languages have no phonetic distinction between voiced consonants and nasals, the second being the realisation before nasalised vowels. For instance, the word for ‘eye’ in many Northern Jê languages is tɔ < *dɔ < *ndɔ < *nɔ. Thus, there is no need to postulate an initial nasal + occlusive cluster to explain why nasal consonants in one language correspond to occlusives in another. I will adopt the procedure of reconstructing nasals, and never initial clusters, for PMJ and PTp, given the Andean cognates where nasals regularly appear. The Aymara form has a suffix -ra that is common to other body parts (ampara ‘hand’, laχra ‘tongue’, maqhura ‘testicles’). The similar suffix in PK is probably unrelated, as it appears in all types of nouns. PK also exhibits an extra initial vowel, a phenomenon that occurs in several other cognates.

2. HAND; TO TAKE : And. PQ *maki; Ama. PTp *po ( ~ *m-), PMJ *mo, PK *amo-rɨ, PP *mɨ-kɨr ~ *mɨ-βɨ, PAr *ama ‘to bring’, All. mameje- ~ mamjek- ‘to take; to bring’. || AED 370 *ma-n ~ *ma-k ~ *ma-r. || Comments: This was probably a CV root to which different suffixes were added, as the AED reconstruction correctly captures. This is evident in the case of PP, where two different roots for ‘hand’ can be reconstructed, one with the suffix *-kɨr (also appearing in Quechua?), the other with *-βɨ. As for the PTp and PMJ words, the same observation regarding initial nasals can be repeated from the previous etymology. PK again displays an extra initial vowel. The main problem with this root is that it is of a CV type starting with an unmarked consonant, so that the chances of chance resemblance are quite high.

3. HAND, ARM; TO TAKE : And. PQ *apa ‘to carry, bring’, PAy *ampara; Ama. PMJ *paC, PAr *po ‘to give’, All. lpuɨ. || Comments: Possibly this root can be conflated with the previous one, but I have chosen not to do so, given that this set consistently exhibits unvoiced occlusives, whereas the one above contained nasals in the same position. I am uncertain about the inclusion of Allentiac here instead of in the preceding etymology. The initial cluster lp- is uncommon but not unparalleled in that language, e.g. lka: ‘one’. The weakness of this root is the same as for the preceding one: CV words with unmarked consonants are the usual suspects for chance resemblance.


4. EGG : And. Kun. qoro, Map. kuram; Ama. PMJ *ŋrɛC. || Comments: A very characteristic trait of Macro-Jê is the presence of initial *Cr- clusters whose origin is not clear. Although this is the reconstruction for PMJ, many derived languages (Maxakali, Rikbaktsa) realise this cluster as CVt- or CVr-. In some modern Jê languages like Xerente, we can observe transitions of the pattern *par > *para > pra ‘foot’. This is a clue that PMJ *Cr- might actually be derived form an earlier *CVr-. The Andean cognates in this set (see also 8. STONE) confirm that possibility, further suggesting (based on Kunza) that the contraction in PMJ might have occurred when the vowels in both syllables were identical.

5. LEAF, TREE, BUSH : And. PQ *satʂa ‘forest, tree, bush’; Ama. PMJ *ʃ-ɔj ‘leaf’. || Comments: In spite of being attested in only two families, and even though the semantic connection is not perfect, this etymology illustrates an important aspect of the comparison between Andean and Amazonian roots, in particular those of the ‘Neo-Amazonian’ core. As I mentioned previously, some PMJ and PTp nouns are distinguished by the presence of a ‘detachable’ initial *ʃ- and *tʲ-, respectively. This initial consonant is lenited or disappears when preceded by a possessive pronoun or when in a genitive construction with another noun. This set, and a few others below, show that this initial consonant regularly corresponds to *s- (and sometimes *ʃ-) in PQ. The situation in PAy is a bit more complicated, possibly involving a correspondence with *h- (also found in PP). Medial *-tʂ- in PQ regularly corresponds to *-j in PMJ (see also 11. WATER below).

6. LOUSE, NIT : And. PAy *k’utʂi, Kun. qeʧir ~ qiʧe; Ama. PMJ *ŋot, PG *aq’ete. || AED 294 *k’wit’ || Comments: The entry in the AED includes only Quechua, but the form cited is in fact a loan from Aymara. Otherwise, it brings interesting parallels in North America. The Kunza word is very close to Aymara, except for the initial q-, so that borrowing is unlikely. The Andean *k’- / q- has an interesting parallel in PG *-q’-, but otherwise I am not certain of the inclusion of the latter. PMJ *ŋ- corresponds to Kun. q- also in 1. EGG, so we can postulate that an initial nasal velar/uvular was lost in the Andean branch, but preserved in the Amazonian one. The final in PMJ represents a problem, since we would expect *-j.

7. ROOT : And. PQ *sapi; Ama. PTp *tʲ-apo. || AED 600 *tap || Comments: Another case of little attestation, but with good phonological correspondence. PP *tapon cannot be related to this set, as initial *h- (and possibly medial *-β-) would be expected for reasons given below. It is probably an ancient Tupi loan. Otherwise, PQ *s- is in good correspondence with the ‘detachable’ PTp *tʲ- / PMJ *ʃ- (as in 5. LEAF).

8. STONE : And. PAy *qala, Map. kura; Ama. PMJ *kraC, PTk *k’ɯ̃ta. || AED 371 *kwele ?, 719 *k’atla, 722 *got. || Comments: The AED etymology 719 is not too different from this one, except that the Tukano cognates are relegated to a separate entry (722). That might be right in the end, as I am not too confident with the correspondence between PTk *-t- and the remaining forms. One difference from the AED is the inclusion, in the latter, of Kawesqar kiella ~ čella, a word that I could not find in the wordlists available to me (at least not with the meaning of ‘stone’). Kawesqar does, however, have kjesáu and qalqajésqe meaning ‘stone’, and these might be related to the proposed etymology after all. In any case, I am not using that particular family in my comparisons. Another unfortunate difficulty is why would Map. have -r- where Aymara has -l-, since both phonemes are available to the first (unlike the Amazonian languages, which normally only have r). I thought of a solution when I noticed that PAy *-l- corresponds to PQ *-r- in loanwords, but soon realised that in these cases the direction of borrowing was from Aymara to Quechua (apparently PQ didn’t even have *l). Apart from that, this etymology confirms the rule of 4. EGG about the formation of PMJ *Cr-.

9. WATER; TO DRINK : And. PQ *jaku; Ama. PK *woku-ru ‘to drink’, PP *waka, PTk *okko. || AED 852 *aq’wa / *uq’wa. || Comments: As in the case of AED 370 *ma-n ~ *ma-k ~ *ma-r ‘hand, give, measure’, this root is supposedly found in Eurasia and could go back to ‘Proto-World’. Just think of Italian mano and acqua! Although such interesting ‘cognates’ are probably nothing more than coincidences, the potential cognates in this entry, all occurring in South America, should be taken seriously. Greenberg and Ruhlen include in this root some Macro-Jê words for ‘water’ and ‘to drink’, though I think they belong to a different etymology altogether (see below). The PQ *j- is difficult to explain, but a similar alternation in the Andean languages could explain PQ *ɲawi vs PAy *najra.

10. WATER, RAIN, RIVER : And. PQ *maju ‘river’, PAy *uma- ‘water; to drink’, Map. mawən ‘rain’; Ama. PTp *(ã)mãn ‘rain’. || AED 853 *man. || Comments: Not much to comment here, except that the AED gives some extremely interesting extra-South American parallels, and many South American forms that I haven’t considered here (e.g. Yanomami ma: ‘rain’).

11. WATER, RIVER, LAKE : And. PQ *qutʂa ‘lake’, Map. ko; Ama. PTp *k’ɨ, PMJ *koj ‘river’, All. kaha. || AED 857 *kwati. || Comments: Ideally there should be more examples of PQ *q- corresponding to PTp *k’-, but the correspondence of PAy *q- with PTk *k’- in 8. STONE means that this is not implausible. In any case, the replacement of q by k or a variant is expected both in the core Amazonian languages and in Huarpe, which lack the first consonant (in fact, even Mapudungun does). The correspondence of PQ *-ʧ- with PMJ *-j also occurs in 5. LEAF.


12. TO COME, TO ARRIVE : And. Kun. t’e-; Ama. PMJ *tε̃C, PK *(w-)ətepɨ, PG *t’ek. || AED 325 *tem || Comments: As I noticed in some of the first etymologies, similar roots of the type CV are not usually good evidence of genetic relationship, given the chances of chance resemblance. In this case, the final consonant of PMJ was most likely *-m, as suggested by Proto-Jê *tẽm and Rikbaktsa tama. This is in good agreement with the PK form (which also includes the usual extra vowel at the beginning), but not with PG or Kunza. The latter misses a final consonant altogether. Apart from that, the use of ejectives in both Kunza and PG is an interesting parallel.

13. TO GO, TO COME, TO WALK : And. PQ *-mu- ‘translocative verbal suffix’, PAy *maja-, Map. amu- ~ miaw-; Ama. PMJ *mɔ̃ŋ, PK *(w-)əməkɨ, All. majek-. || AED 324 *min ~ *man. || Comments: this should really be considered a CVC root. The final consonant was most likely **-ŋ, lost or lenited in Quechua, Aymara, Mapudungun and Allentiac. In PK, it is a common phenomenon that final nasals in PMJ correspond to unvoiced plosives, and that roots end in vowels, hence *-kɨ (see previous root for the same changes). The PQ root is not a verb per se, but a very productive suffix to change the direction of movement, e.g. apay ‘to bring’, apamuy ‘to take’. It is present in *ʃamu- ‘to come’. For the correspondence, with PAy *-j-, see also 1. EYE.

Other words

14. NAME : And. PQ *ʃuti ( ~ *s-); Ama. PTp *tʲ-et, PMJ *ʃ-(ij)it, PK *ətetɨ, PP *harɨ, All. hene. || Comments: This is my favourite etymology. It is a cultural term, it has a wide distribution, and it illustrates a series of regular sound correspondences. The ‘detachable’ initial consonants of PTp and PMJ are by now well known. They correspond to PQ *s-, which is one possible reconstruction in this case, the other being *ʃ-. This is one of those tricky words with nearly identical reconstructions in PQ and PAy for which we do not know the direction of borrowing. It would be crucial to know, because Aymara only has *ʃuti. Anyway, in all such cases, I only show the PQ reconstruction. In PK, *(V)t- is the usual correspondence (PK likes to add vowels to the beginning and end of the words). In PP, *h- is the regular equivalent in this set, but I do not have enough examples of a medial *-t- appearing as *-r-, so I will give it the benefit of doubt.

15. NO, NEGATION : And. PQ *ama; Ama. PTp *ãm, PP *(ja)ma. || Comments: Surprisingly absent from the AED.

16. PATH, ROAD; TO WALK : And. PQ *puri- ‘to walk’; Ama. PT *(a)pe, PP *βaʔi, PTk *ma, PA *apu. || AED 595 *p’en. || Comments: Another unfortunate CV root. The *-ri- in the PQ verb is probably a crystallised inchoative suffix. This root has an interesting Maya parallel, as noticed in the AED.

17. TWO : And. PAy *paja, Kun. p’oja, Map. epu; Ama. PA *api. || AED 821 *(ne-)pale, 825 *pit. || Comments: There are 12 reconstructions for ‘two’ in the AED, which suggests that the speakers of the proto-Amerind language were obsessed with arithmetics. Possibly PP *raβɨt fits here, but we would have to explain the initial *r-.



Appendix: establishing sound correspondences

 In establishing regular sound correspondences between South American proto-languages, the Neo-Amazonian core (Tupi, Macro-Jê, Karib) is extremely useful, due to certain initial consonants that have been shown to reveal perfect cognates. PTp words starting with *tʲ- regulary correspond to PMj *ʃ- and PK *(V)t-. In the first two, these are the famous detachable consonants: they disappear or change to *j- (*ɲ- before nasals) in genitive constructions. A good comparison of basic vocabulary between the three proto-languages, where this rule of correspondence is confirmed, can be found here. If one can find that in other (proto-)languages there are regular correspondences to this set, we would be one step further to establish more cognates and prove the deep genetic relationships of most South American languages. It so happens that there are regular correspondences in PQ and PP. Consider the following examples:

PQ = Proto-Quechua; PTp = Proto-Tupi; PMJ = Proto-Macro-Jê; PK = Proto-Karib; PP = Proto-Pano.

It is clear that in Proto-Quechua the Neo-Amazonian trio *tʲ- : *ʃ- : *(V)t- corresponds to *s-. There certainly are more examples, with less strict semantic correspondence, than the ones I have selected. In Proto-Pano, there are even more cognates and a perfect correspondence with *h- in this proto-language. This allows us to identify previously undetected loans, such as the word for ‘root’ *tapon in PP. I had once thought that this was a nice looking cognate for PT *tʲ-apo and PQ *sapi, but now I think it is probably borrowed from the first, given that everywhere else PT *tʲ- corresponds to PP *h-. As for Aymara, there are certain difficulties, as the number of cognates that I could establish are not so large as to prove (or disprove) regularity. I currently sustain the hypothesis that the most likely correspondence is PAy *h-, showing a similar development as in PP. The main reason for this are the following potential cognates:


Now, PMJ has an interesting feature: many body parts include a prefix for ‘flesh’ *ʃ-ĩt. This appears, for instance, in *ʃ-ĩ-krãj ‘knee’ and *ʃ-ĩ-pV ‘ear’.  This may be the case in the example for ‘nose’, shown above. The word for ‘flesh’ proper has a perfect cognate in PTp *tʲ-ẽt. Although ‘flesh’ itself has no likely cognates in PQ, the word for ‘nose’ might preserve a prefix similar to that in PMJ. I had once considered that the potential PQ correspondence to the Neo-Amazonian detachable consonants was *r-, as in the word *rinri ‘ear’, precisely because of the ‘flesh’ prefix, and because I saw that particular word as cognate to PAy *hinʧu. However, all the evidence shown above points to PQ *s- as a better candidate. Therefore, I currently see PQ *sinqa ‘nose’ as somehow related to the Amazonian ‘meat’ and ‘nose’ words, and I assume the first part of the word to be cognate to the first half of PAy *hinʧu ‘ear’. An even better potential cognate to the Amazonian words is PAy *hanʧi ‘meat, flesh, skin’, although, admittedly, both are not mutually exclusive.

Amerind from a South American perspective: Part II

Continuing on the subject of South American languages, possible long-range relationships between them, and whether such relationships can enlighten us about the viability of the Amerind hypothesis, I would like to propose a tentative ‘macro’-classification. This will be quite different both from that of the splitters (who don’t propose macro-families at all) and from that of the lumpers (by which I mean Greenberg’s classification, so often repeated in the scientific literature).

sam_amerindThe readers possibly saw the map on the right in a previous post. Greenberg’s classification of the South American languages divided them into broad phyla such as “Andean-Chibchan-Paezan” and “Ge-Pano-Carib”. Ideally, such divisions should be based on shared innovations and vocabulary, but that is not always the case (in fact, it seems that Greenberg’s classifications were ready in his mind before he started looking at the data). Now, I believe a much more accurate, “data-driven” scheme can be devised if the following criteria are taken into account: 1. similarity in morphology; 2. personal pronouns; 3. shared retentions or innovations in basic vocabulary. Let us examine each on its turn.

Prefixing vs Suffixing

First of all, the similarities in morphology. I have written a whole post about the differences between “Andean” and “Amazonian” languages, stressing, among other things, how the first tend to be suffixing whereas the second typically employ more prefixes. I will not repeat the evidence here, but surely one could argue that it only bespeaks of areal features. However, the argument could also be the other way round, i.e. that grammar is hardly borrowed. Otherwise, how to explain anomalies such as Ket and its complex verb chain in the middle of a sea of typically Eurasian languages (Indo-European, Uralic, Tungusic) with which it has probably been in contact for a long time? A language’s morphology is not immutable (remember the example of Egyptian?), but it might tell a lot about it’s historical relationships. After all, is it not on the very origin of the recognition of the Indo-European or Afro-Asiatic families?

N : M vs I : A

Shared personal pronouns are usually considered strong evidence that languages are related. At the same time, they can deceive, since all languages tend to use unmarked phonemes for pronouns (e.g. n, m, k, t…) and, therefore, similarities may arise by coincidence. Nevertheless, as I have been repeating for a while, the question then is: shouldn’t the distribution be random? That is, why would some languages choose a set of unmarked consonants (m : t in Eurasia) while others choose a distinct set (n : m in the Americas)? That is when genetic explanations are more likely. The mention of the controversial n : m series was not fortuitous, as it is quite productive in the classification of South American languages: most families exhibit this ‘pan-American’ pair, but not all! Consider the following example:


A quick look at the evidence above, even without prior knowledge of the geographical distribution of the languages, would lead to their classification in two, or maybe three, groups. Most of them use the pan-American n : m series for the 1st and 2nd persons. However, the Andean languages – Quechua, Aymara, Mapudungun – add a suffix to the 1st person (*-qa, *-ja, -ʧe) and a prefix to the 2nd person (*qa-, *xu-, ej-). This, of course, occurs only in the independent form of the pronouns; in Mapudungun, the 1st and 2nd person verbal suffixes are -n and -mi. The other languages highlighted in blue are Amazonian (or intermediates between highlands and lowlands, in the case of Chibcha). They use the n : m  series in a monosyllabic or prefixed form. Pano and Tukano have particularly similar forms. In contrast, other lowland languages appear to have replaced the pan-American pronouns by a vowel pair. This pair, which is i : a in the case of Macro-Jê and Guaykuru, also appears in the Maya languages.

Does that pattern have any implications for classification or is it just a coincidence? As I have mentioned before, the Macro-Jê and Tupi languages have long thought to form a macro-family, together with Karib. It could be suggested that the Guaykuru languages are part of the same group, though a bit removed (this is confirmed through shared vocabulary, as I will show below). Look at a map and you will see that these languages occupy a huge part of eastern South America. They appear to have originated in SW Amazonia. The other Amazonian languages, the ones that preserved the n : m  pair, have a more restricted distribution in NW Amazonia (except for Arawak). Thus, we can start to envisage a distinction between the Andean/Patagonian phylum and two separate Amazonian phyla, one of which is characterised by widespread geographical distribution and the i : a innovation in the personal pronouns.

*LAQ’w vs *NENE

Shared vocabulary is the last evidence, and the most prone to borrowing. That is why I insist it must be basic vocabulary with close semantic correspondence. This is a major problem with some of the etymologies in the AED, as in the case of the famous *t’ina ~ *t’ana ~ *t’una, where any relative (or, in fact, human being!) can be compared to any other. This problem exists even in works like the Altaic etymological dictionary of Starostin, and is one the reasons why I think the Dené-Yeniseian hypothesis is plausible (items of basic vocabulary, with close meanings and regular correspondences connect the two families).

Shared innovations in basic vocabulary can help distinguish subgroups. Let us take as an example the Romance languages. Even though they are all derived from vulgar Latin, we can show that some are closer relatives than others, and that is reflected in vocabulary: for “leg”, Portuguese and Spanish have perna and pierna respectively, whereas French and Italian have jambe and gamba. For “head”, we have cabeça and cabeza against tête and testa. For “morning”, manhã and mañana against matin and mattina, and so on. Needless to say, Portuguese and Spanish are closer to each other than each is to either French or Italian.

After scrutinising wordlists of South American languages (admittedly not the best way to do historical linguistics), I believe I have identified six very stable words that are particularly useful in distinguishing subgroups. These are six words that include mostly body parts, but also a simple noun and a verb: ashes, foot, to sleep, tongue, tooth, two.


The table above shows these six words in thirteen languages (or, rather, proto-languages in most cases). I have highlighted, in red and blue, contrasting pairs of words that could be derived from common roots, given their phonological similarity. The gaps in some of the languages are either because the words are not reconstructible, or because I do not have a complete wordlist at my disposal. Some of the “cognates” are only tentative. The possible relationship of PQ *qaʎu and Kaw. qala- to forms such as Kun. lassi has been explained previously. PQ also has *ʎaqwa- ‘to lick’, so metathesis cannot be ruled out. The PTp word for ‘tongue’, despite the resemblance, cannot be related to that set, as I will show in the next post. What is clear from the comparison above is that some languages appear all in red, others all in blue, and some fall in between. This is exemplified by the title of this section, where I make reference to two words for tongue (or ‘to lick’) found in the AED, the first of which appears in the Andes, the second in the lowlands. On the one extreme, in red, we have the core Andean/Patagonian languages. On the other, in blue, we have the core Amazonian languages – Tupi, Macro-Jê and Karib – confirming the evidence from personal pronouns. Some Amazonian languages fall in between: Arawak and Pano. They are also the ones that preserved the pan-American n : m pronouns, also present in the Andean languages. I don’t know enough about the Tukano languages to risk a classification with such a limited number of words as in the table above, but other vocabulary evidence (plus the pronouns) would place it not far from Pano and Arawak. The usefulness of such approach, against mere geographical conveniences, is that some surprises may emerge: Allentiac, a Huarpean language of NW Argentina, is completely Amazonian in its basic vocabulary.


I end this post with a map that reflects my current view of the macro-relationships between the South American families. The primordial division, about which I’ve written before, is that between “Andean” and “Amazonian” languages. To put it in more neutral terms (since there are “Amazonian”-type languages well outside of that basin), we can call them “Highland” and “Lowland” or, even better, “Western” and “Eastern” South American languages – the option adopted in the map. There is a core group of “Andean-Patagonian” languages that includes Quechua, Aymara, Mapudungun, Kaweskar and Chon. Possibly Chimu (Mochica) falls in that group, but I do not have a large enough material at my disposal to ascertain that. Chibcha is probably part of the Western division, though it displays some Amazonian features. I cannot specify the position of isolates like Kunza or small families like Puelche within the group, but they are definitely Western.

Among the Eastern languages, the situation is complex. There is a core group of three families – Macro-Jê, Tupi and Karib – whose deep relationship is given as certain. I like to call them the “Neo-Amazonian” languages, given their relatively recent spread from an Amazonian homeland over enormous areas of the South American continent. Pronominal and vocabulary evidence show that Guaykuru (and almost certainly Mataco, though I do not have the data with me) is not far removed from that group. In the map, I show all of them under the label of “Chaco-Amazonian”. They contrast with languages that did not spread as much, remaining since early times in their original homelands. On the one hand, we have the Amazonian families Pano and Tukano, whose deep relationship I take as a serious hypothesis. We can call them “Palaeo-Amazonian” families. On the other, we have those that seem to occupy a similar historical position in the Chaco (or its border), such as Nambikwara and Zamuco. I propose to call them “Palaeo-Chacoan”. Many small families and isolates (Huarpe, Witoto etc.) are Eastern languages, but their relationship with the others is less certain. Finally, Arawak occupies a strange position. I used to think it had to be related to the ‘core’ Amazonian languages, but after long staring at the pronominal, vocabulary and morphological evidence, I am sure it is far removed even within the Eastern group. Its likely western Amazonian origins, close to the Palaeo-Amazonian languages, would confirm that it is indeed an ancient split.