The website that accompanies “Mapping the Origins and Expansion of the Indo-European Language Family” (August 24 Science), maintained by co-author Quentin D. Atkinson, proudly features several maps that allow the easy visualization of the patterns generated by the model. One is a conventional map that purports to show “language expansion in time and space,” depicting and dating the spread of Indo-European languages through a red-to-blue color scheme. The other cartographic product is a sequence of numerous map-frames that ostensibly shows Indo-European (I-E) expansion from the seventh millennium BCE to 1974 CE. This Google-Earth-based animated map, or “movie,” as Atkinson calls it, is explained in terms that are at once simplistic and cryptic:
Watch the Indo-European expansion unfold. This movie shows how our model reconstructs the expansion of the Indo-European languages through time. Contours on the map represent the 95% highest posterior density distribution for the range of Indo-European.
The analysis that I provide below takes these maps on their own terms, as advertised: as if, in other words, they indicate what Atkinson and his colleagues believe to be the “unfolding” of the Indo-European language family in “time and space” as substantiated by their mathematical model. But if one reads the fine print found elsewhere, one discovers that the maps are not actually what they purport to be. The authors admit up front that these figures deliver incorrect information, owing to the fact that crucial pieces of data were excluded from the model:
This figure needs to be interpreted with the caveat that we can only represent the geographic extent corresponding to language divergence events, and only between those languages that are in our sample. The rapid expansion of a single language and nodes associated with branches not represented in our sample will not be reflected in this figure. For example, the lack of Continental Celtic variants in our sample means we miss the Celtic incursion into Iberia and instead infer a later arrival into the Iberian Peninsula associated with the break-up of the Romance languages (and not the initial rapid expansion of Latin). The timing represented here therefore offers a minimum age for expansion into a given area.
This admission is extraordinary, as it amounts to saying that “even though our data set is too incomplete to produce accurate results, our model should nonetheless be regarded as powerful enough to settle the most highly debated topic in historical linguistics,” and that “even though we make no claims as to the earliest dates in which Indo-European languages were established in any given area, our approach still shows that the language family originated in Anatolia.” I do not think that I have ever encountered a more flagrant example of “having one’s cake and eating it too” in an academic work. In fact, as is demonstrated in a previous discussion thread that is reproduced below, the “caveat” itself errs at virtually every turn.*
In a comment on the previous post, co-author Alexei Drummond framed the study’s limitations in more direct language:
Our geographical reconstructions are only for the language lineages that are direct ancestors of the particular sample of IE languages we analyzed. Our inferred geographic distributions don’t say anything about the full extent of IE languages at any time past or present.
If the geographic patterns depicted on the maps say nothing about the “full extent” of I-E languages “at any time,” why are viewers of the animation invited to “watch the Indo-European expansion unfold”? The claim is evidently inherently misleading. But as we shall see below, the problems run much deeper, as in numerous instances the maps fail to accurately show the partial extent of I-E languages. But before delving into such specificities, a few words about the mapping project in general are in order.
Many problems plague the authors’ cartographic depictions. The two maps, static and animated, fail to correspond in their details, often in a glaring manner. The animated map, moreover, lacks anything approaching a key, and hence is difficult to interpret. The temporal framing of the two maps is oddly displaced, as the “movie” purports to take the story up to 1974 CE, whereas the static map terminates at roughly 1800 CE. Potentially confusing is the fact that the static map gives dates in “BP,” or “before present” (which by conventions means prior to 1950 CE), whereas the animated map uses the historically Christian calendar. Both maps, it is essential to note, show only the expansion and not the contraction of Indo-European, although this essential feature also goes unmentioned. Areas that ceased to be Indo-European speaking centuries ago, such as the supposed Anatolian heartland, continue to be shaded as I-E throughout the animation.
Although the contours mentioned in the “explanation” of the animated map are visible in the greenish shading, the overall coloration scheme remains vague. As the animation unfolds, the hypothesized I-E homeland circa 6500 BC—Anatolia, the Caucasus, the northern Middle East, and the greater Aegean—is washed in yellow, whereas later geographical addition to the realm appear in shades of green. Yet at approximately 2225 BCE, most of the heartland abruptly turns green as well, with the exception of a swath extending from Cyprus through what is now Lebanon to central Iraq and two areas on either side of the Black Sea. Another such abrupt color switch occurs later in the animation.
Also unspecified are the thick green lines, which begin as a several-pixel splotch at roughly 6200 BCE that gyrates in place for about 1,500 years before spreading across the map to form a web. An unwary reader might assume that such lines indicate pathways of migration, but he or she would be mistaken, as movement along specific corridors defies the underlying diffusional model, which postulates gradual expansion along broad fronts with scattered outliers pushing into new territories. The lines actually indicate supposed examples of family-level linguistic divergence. Such relational links often extend into areas that are not shaded as I-E; note, for example, the green lines pushing into unmarked western Russia and northern Sweden on the first map. A naïve reader might wrongly assume that such extensions signal relatively recent movement, with little actual settlement to date.
As mentioned above, the static map and its animated companion do not correspond well. Unlike the animated version, the conventional map shows Corsica, the Balearic Islands, Crete, and Cyprus, for example, as never having been occupied by Indo-European speakers. The animation, to the contrary, puts Cyprus in the initial I-E homeland in the seventh millennium BCE. (Both depictions of the island are incorrect; the first known language of Cyprus, non-I-E Eteocypriot, was supplanted by the Greek (I-E) dialect of Arcadocypriot in the late Bronze Age.) Also notable is the static map’s depiction of Indo-European occupation in areas unmarked on the animated map, including western Norway and western Russia. (Neither map manages to show northern Norway as ever having been occupied by Indo-European-speakers.)
Although the discrepancies between the two maps are never explained, a few of them might be deduced. Consider, for example, the different treatments of western Russia in the maps posted here. In the animated depiction of 1974, only a small portion of this region is shaded as ever having been I-E speaking, yet the static map shows a sizable area as having become largely Indo-European over the past 500 to 1,000 years. This map depicts the distribution of I-E languages in western Russia with discontinuous blotches, seemingly placed at random, which would apparently indicate that the language family spread into this area in a spatially sporadic manner and never managed to fill in the gaps. On the basis of this particular disparity, one might assume that only areas of (supposedly) continuous I-E occupation receive shading on the animated map frames. But if this is indeed the case, the guideline is apparently reversed elsewhere. Note that sizable portions of Central Asia are similarly splotched on the static map, yet are shaded on the animated map. The area that now constitutes Kyrgyzstan is fully shaded on one map, yet remains almost entirely blank on the other. A swath across what are now Syria and Iraq is blobbed red on the static map, apparently indicating partial I-E expansion in the Neolithic, yet is blanketed with yellow on the animated map from the earliest frames. Cartographic consistency is evidently not high on the authors’ agenda.
Far more troubling than disparities between the two maps, however, are inconsistencies between both of them and the historical record. Overall, the fit between the modeled spread of I-E languages and what we know of its actual expansion is poor. In pointing out some of the more flagrant errors, I will begin at the end of the “movie,” which shows the accumulated spread of I-E languages to 1974 CE, contrasting it with the depictions on the static map. I will subsequently work backward in time on the “historically unfolding” movie, pointing out crucial errors for several particular periods. To reiterate, I will consider what the maps literally show, ignoring for the most part their hidden meanings.
As mentioned in the previous post, the most obvious blunder in the 1974 depiction is the omission of Russia and Eastern Ukraine from the Indo-European-speaking realm. On the final map frame, the only parts of Russia that are shaded are the Pskov district, the far southern Crimea, and the largely non-I-E-speaking northern Caucasus. The same map also fails to mark other areas long characterized by I-E speech, such as southern Iberia, Balochistan, southern Sri Lanka, and Orissa in eastern India. The static map, however, does successfully mark most of these places as I-E speaking, yet conversely errs in placing several non- (and never-) I-E-speaking areas in the Indo-European zone, such as northeastern Sri Lanka as well as Manipur and environs in northeastern India. Unlike the animation, this map does show I-E in Western Russia, but only in the past 1,000 to 1,500 years, as discontinuous as late as 1800 CE, and as disappearing entirely in far western Siberia. Such depictions, needless to say, are erroneous; although pockets of Uralic languages persist to the present in eastern European Russia and Western Siberia, the bulk of the region was solidly Russian speaking well before the termination date of 1974. Compounding such errors is the sprinkling of bluish dots in southern Tibet, northern Nepal, and northwestern Burma. Some of the most inhospitable parts of the central Sahara are also vaguely marked with blue to show I-E expansion over the past millennium.
The static map is, in a word, preposterous. What possible Indo-European language could ever have been spoken in the Kachin uplands of Burma over the past 1,000 years, much less in essentially uninhabited areas of the Tibetan Plateau and the Sahara Desert? Note as well that northern Tunisia and northeastern Algeria are clearly marked as having been substantially I-E speaking in recent centuries. On first glance, I wondered whether the authors were trying to show the spread of Latin in this region under the Roman Empire; if so, the coloration is wrong, as blue indicates I-E expansion in the past 1,000 years. But as we have seen, Latin does not count in Atkinson’s scheme, as it supposedly spread too quickly as an individual language (it actually spread quite slowly here; non-I-E Punic continued to be spoken in the region as a minority language up to Augustine’s time). But as it so happens, the blue splotches around Tunis do not indicate anything nearly so specific. Rather, like the light red blobs in central Arabia, they merely show that the model occasionally spits out randomly (and incorrectly) placed outliers at some remove from main areas of Indo-European speech.
Other inaccuracies abound on the static map, including incomplete I-E occupation at the termination date (1974) in western France, Andalucía (but not in Spain’s Basque Country!), and northeastern Scotland, as well as a complete absence of the language family from Gotland in the Baltic along with the previously mentioned Mediterranean islands. The map seems to show that Indo-European languages have never quite yet reached the Atlantic, although of course the authors would likely counter that the map does not actually depict what it claims to depict. Or consider the model’s portrayal of non-I-E-speaking areas in Fennoscandia with that of an actual language map of the region, as can be seen to the left. The fit is poor.
The Fennoscandia map detail also presents evidence that contemporary geopolitical boundaries anachronistically mold the hypothesized language-family distribution in the Science model. As can be seen on the actual language map, linguistic and political boundaries do not correspond particularly well in this area; Estonia and Finland may be non-I-E-speaking countries, but not over their entire expanses. On Atkinson’s map, however, I-E coloring abruptly and transhistorically ends exactly at the modern Estonian border, a most suspicious situation. The general lack of I-E shading for Moldova also makes me wary—and is completely bizarre. A clear example of contemporary geopolitical contamination is found in the portrayal of Central Asia. Note the salient of solid I-E coloration extending northward into Tajikistan’s portion of the Fergana Valley, avoiding the core of the valley held by Uzbekistan. Such a portrayal would be understandable if the map depicted merely present-day conditions, as Tajikistan is mostly I-E-speaking whereas Uzbekistan is not. But the sorting of “Sarts” into Uzbeks and Tajiks, along with the forced “Uzbekization” of many previously Persian speakers, in this historically heavily bilingual area is largely the product of Soviet geo-ethnic machinations. If one delves back to the first millennium CE and earlier, the entire region was heavily I-E-speaking (Sogdian and other Iranian languages).
Consider, for example, the depiction of western Europe circa 1000 CE. At this time, western France, Sicily, and the entire Iberian Peninsula are shown as non-I-E-speaking, although a line of I-E linguistic relationship has been etched across southern France roughly to the Spanish border at the crest of the Pyrenees. The false implications conveyed here—which are fully admitted as erroneous by the authors—are that Roman Hispania and Aquitania were never Latinized, and that the preexisting Celtiberian and Gaulish tongues were not I-E. The same 1000 CE map frame also incorrectly excludes from the I-E realm the South Asian areas that now constitute southern Gujarat, southern Balochistan, most of Maharashtra, and southern Sri Lanka. Note as well that most Norse areas are not given an I-E shading, nor is northern Scotland. Yet at the same time, southern Tibet is placed within the I-E zone! Even the essentially uninhabited and uninhabitable region of Aksai Chin is depicted as Indo-European-speaking at this time; I can’t help but imagine proto-Dardic speaking yetis.
Turn back to the portrayal of the year 18 BCE, and the errors compound. The most conspicuous I-E omission here is the Scythian/Sarmatian realm, which by itself is enough to discredit the model; it almost seems as if the authors intentionally manipulated their data to exclude the linguistically hypothesized steppe homeland of the I-E family. The northeastern salient of I-E languages depicted for the time, which denotes the Tocharian languages, oddly excludes a significant portion of the Tocharian homeland in the Tarim Basin to focus instead on the lofty Tien Shan Mountains. Tellingly, the diffusional front hypothesized here has the ancestors of the Tocharians advancing along ridges well in excess of 20,000 feet in elevation.
Several nice examples of demonstrably false information are found on the depiction of the Mediterranean Basin circa 700 BCE. Here we see the greater Aegean along with the Italian Peninsula clearly colored as I-E, but with little else falling in the same category; Sicily, most of Sardinia, and most of the littoral zone of southern France and eastern Iberia are excluded. Yet we have incontrovertible knowledge that Greek-speaking colonies had been firmly planted in western Sicily, Cyrenaica in North Africa, and over a large expanse of the northwestern Mediterranean coastlands. The spread of the Greek language to Crete, moreover, occurred much earlier, as attested by the Bronze Age Linear B script. The model fails here in part because it does not count the “rapid” spread of individual languages; Greek colonization, however, took place over hundreds of years, and some of the dialects of ancient Greek were differentiated enough to be classifiable as separate languages.
While the 700 BCE map frame unduly restricts the spread of I-E over much of the Mediterranean, it also improperly extends it in other parts of the basin. Several relatively well-known non-I-E languages persisted in the map’s “green zone” well beyond 700 BCE. On the island of Lemnos, the non-I-E Lemnian language vanished only with the Athenian conquest in the fifth century BCE, while Etruscan and Raetic survived into the first millennium CE. Together, Lemnian, Etruscan, and Raetic seem to have constituted the extinct Tyrsenian language family, which might have included Minoan (Eteocretan) and Eteocypriot as well. The scattered distribution of this family in antiquity probably signals that Tyrsenian languages had blanketed a much broader area before the incursion of I-E speakers. In the Science model, however, the entire Aegean region is mapped as I-E speaking as early as 6500 BCE. Are we to imagine a post-I-E migration of Tyrsenian speakers into the Aegean from Etruscan- or Raetic-speaking areas further to the west? Yet historians who have viewed the Tyrsenian Etruscans as non-indigenous have instead tended to locate their homeland in Anatolia, the hearth of I-E in the Science model! Today, however, a near consensus has emerged that the Tyrsenian languages represent a pre-I-E substrate that likely extended across much of the northeastern Mediterranean in the fifth millennium BCE, if not significantly later as well.
Finally, consider the depiction of supposedly I-E-speaking “greater Anatolia”—including what is now Syria and northern Iraq as well the Caucasus—in the Bronze Age, circa 1500 BCE. Yet we have unassailable historical evidence of widely spread non-IE languages over much of the region at this time, including Hurrian, Hattic, and, for a somewhat later period, Urartian. Much evidence suggests, moreover, that the three (or perhaps four) extant Caucasian language families covered much broader swaths of land in ancient times than they do today; modern Azerbaijan, for example, was a largely NE-Caucasian-speaking area, as attested by both historical sources and the extant language of Udi. For the Science model to make sense, later migrations of several different non-I-E groups would have had to have pushed through long-inhabited I-E lowlands to settle in inhospitable areas of mountainous terrain. Such a scenario, to say the least, strains credulity.
*. Let us consider here the various elements of the authors’ “caveat”:
1. “we can only represent the geographic extent corresponding to language divergence events.” Do languages really diverge in discrete events? Does not language divergence happen continually? Whenever one segment of a language community adopts a new word, a new sound, or a new grammatical feature, some degree of divergence has occurred. It is always an open question as to when diverging dialects become separate language; in the modern world, the issue is more political than linguistic (cf Serbo-Croatian, Serbian, Croatian, Bosnian, and Montenegrin).
2. “only between those languages that are in our sample.” That is interesting, seeing as Atkinson claims in an interview (to be cited later) that “all” I-E languages were included (an impossibility, as there are no hard and fast divisions between languages and dialects). But more to the point, if one can simply exclude languages at will from the sample, then one can then mold the results. Drop a few more languages, and the maps will differ. In such a manner, one can get the results that one wants.
3 “nodes associated with branches not represented in our sample will not be reflected in this figure.” Yes indeed, which is one reason why the figures are so spectacularly wrong.
4. “the lack of Continental Celtic variants in our sample means we miss the Celtic incursion into Iberia and instead infer a later arrival into the Iberian peninsular…” I am glad that the authors begin to acknowledge their own errors here, but they still do not go far enough; they do make an inference, and that inference is simply incorrect. They also miss not just Celtiberian and Latin, but also Mozarabic, Ladino, and several other I-E languages of the Iberian Peninsula (the map frame for 1000 CE still shows only partial I-E coverage).
5. “associated with the break-up of the Romance languages.” The model assumes that Latin began to “break-up” with the fall of the Western Roman Empire. That is incorrect, as divergence began much earlier. The “vulgar” Latin of the distant provinces was not the language of Cicero.
6. “not the initial rapid expansion of Latin.” Latin did indeed expand rapidly as a language of administration, but not necessarily as a language of everyday use. Basque remained in use throughout, although the maps produced by the study indicate otherwise.
7. “The timing represented here therefore offers a minimum age
for expansion into a given area.” This proviso is particularly rich, as it alone undermines the approach. In other words, I-E languages could have been found in any part of the study area at much earlier times than indicated? If so, how can one pinpoint Anatolia as the place of origin? If one claims to “find” a location of origin, then one is automatically making an argument for “maximum ages” in areas that fall outside that supposed birthplace.