Relative chronologies

When it comes to individual lexical items, historical linguists look for various clues to establish the etymology and the relative chronology of the development of words. For example, Edward Sapir (Reference Sapir and Mandelbaum1916/1949: 434–435) pointed out that “the objects […] denoted in English by the [unanalyzable] words bow, arrow, spear, wheel, plough […] belong to a far more remote past than those indicated by such [analyzable] words as railroad, insulator, battleship, submarine, percolator, capitalist, and attorney-general”. Another technique, applied with remarkable success by Sapir in his study of Navajo (Sapir Reference Sapir1936), involves deriving historical information from cultural items whose linguistic designations have obviously changed their meaning; Sapir's example of such a phenomenon in English is the word spinster, which changed its meaning from ‘one who spins’ to ‘unmarried female of somewhat advanced age’. Mallory (Reference Mallory1976: 45) takes non-productive (i.e. irregular) morphology to likewise indicate the relative antiquity of the words (and by extension, of the associated concepts); for example, the irregularity of such plurals as ox/oxen, sheep/sheep, and calf/calves is taken as evidence of the great age of stockbreeding among speakers of English and their linguistic ancestors. This argument is problematic, however, as the retention of irregular forms such as foot/feet and tooth/teeth but not of the irregular plural of book (Old English bēc, which after the Great Vowel Shift would have resulted in the form beek) is a reflection not of the fact that feet and teeth predate books but of the higher frequency of the irregular plural forms of ‘foot’ and ‘tooth’ compared to that of ‘book’ in the Middle English period.

Yet another strategy used to reconstruct the relative chronology of words and associated concepts rests on the assumption that “items which are represented by terms which have cognates widely spread across the languages in the family are likely to be older than terms which lack such a wider distribution among the related languages” (Campbell Reference Campbell, Renfrew, McMahon and Trask2000: 11). For example, just as the irregular plural form of sheep is indicative of the word's antiquity, so is the fact that the cognate set for ‘sheep’ across Indo-European languages is broad, including Luwian hawi, Sanskrit avis, Greek o(w)is, Latin ovis, Lithuanian avis, Proto-Slavic *оvьсa (cf. Russian ovca), Old Irish oi, and English ewe.

More important than the reconstruction of relative chronologies of words is the reconstruction of relative chronologies of sound changes that affect the entire lexicon. Our first example comes from the development of modern French from Latin, illustrated in Table 8.1 (the pronunciation of the initial consonant in the first three words is shown in transcription as well).

These words illustrate two sound changes that occurred as (a dialect of) Latin transformed into French. One change is the transformation of /k/ into /ʃ/, mentioned in Chapter 4 above. This change applied only in front of a low back vowel /a/, as can be seen from the fact that coeur ‘heart’ retained the original /k/. Another change involves the diphthong /au/ turning into /o/. Crucially, the two changes must have occurred in a particular order: palatalization of the velar first, and the vowel change second. Had the vowel change taken place before the palatalization, the original au in causam would have become o, and the palatalization rule would not have applied to this word at all, leaving it as /kozam/, eventually to become modern French /koz/ rather than /ʃoz/. In this case, the vowel change “bleeds” the palatalization: the application of the vowel change creates a context in which the palatalization rule can no longer apply. Thus, although we cannot tell without historical records when exactly the two changes occurred on the way from the local variety of Latin to French, we can deduce their relative chronology.

Another example concerns the development of the English word foot and its plural form feet. This word derives from the Proto-Indo-European (PIE) form *pód, which is cognate with Latin pēs/pedis, Sanskrit pad-, and Lithuanian pãdas ‘sole (foot)’. Consider the consonants first. The transformation from pVd (where V symbolizes some vowel) to fVt involves two changes: p > f and d > t. Each of the two changes was part of a larger transformation that involved a whole class of sounds: p was changed alongside t and k into f, θ, and h, respectively, whereas d changed alongside b and g into t, p, and k, respectively. The relative chronology of these two changes is easy to establish: the voiceless stops p, t, and k must have become fricatives first, and the devoicing of voiced stops b, d, and g into p, t, and k must have occurred later. Had the order of the two changes been reversed and the devoicing applied first, the original *d inherited from PIE would have become *t, and then this new t and the original *p would both have changed to fricatives. Thus, the original *pVd would have become *pVt and then fVθ, leaving English with footh instead of foot. Clearly, this is not what actually happened, so we can deduce that the change from voiceless stops into fricatives happened before the devoicing, without attaching absolute dates to these changes (known as stages of Grimm's Law, or the First Germanic Consonant Shift, which happened in Proto-Germanic).

And how did the vowel alternation in foot/feet come about? In Proto-Germanic, as in PIE, the singular and plural forms had the same vowel *ō, as in *fōts/*fōtiz (the macron over the vowel means that it is pronounced long). Both the final z and the i were later lost, for reasons that need not concern us here, and the vowel in the plural form changed from ō to ī. Moreover, this latter transformation happened in several stages: first, ō was fronted into a long ø (this process is known as Germanic umlaut), then it was unrounded into ē, then raised to ī by the process known as the Great Vowel Shift. We know from written records that the latter change must have happened circa 1500 CE; likewise, we also know from textual sources that the form fēt was attested in the Old English period (i.e. from the sixth to late eleventh century CE). From this we can deduce that the initial fronting of ō must have happened earlier, in the Pre-Old English stage. Thus, while we cannot attach a precise date to this change, we can put constraints on it, on the basis of a relative chronology with respect to other changes that can be dated in absolute terms based on extra-linguistic information.

Establishing a relative chronology of sound changes can also help us understand layers of borrowing to and from a given language. This point can be illustrated with Latin words originally containing a velar stop /k/ that were borrowed into Basque at different times (cf. Trask Reference Trask, Renfrew, McMahon and Trask2000: 45–46). The Latin word cellam ‘chamber’ preserves a velar stop in the Basque form gela ‘room’; it must have been borrowed in the early period, before Latin underwent sound changes that affected its velar stops. In contrast, Latin caelum ‘sky’, which appears in Basque as zeru, with the initial consonant pronounced as /s/, must have been borrowed later, after Latin velar stops have been palatalized and assibilated, turning the Latin word into */tselo/. Finally, Latin cepullam ‘onion’ must have been borrowed into Basque in the intermediate period, when the velar stops of Latin were palatalized but not yet assibilated: its Basque form tipula is a reflex of the Latin form containing a palatalized velar *k^j, which Basque turned into an alveolar stop.

More generally, an analogy can be drawn between the historical linguistic and genealogical enterprises. Linguistic evidence alone can reveal relative chronologies but not absolute dates; similarly, in genealogy “it is one thing to know who is whose mother, or grandfather, or sibling, but quite another to date earlier lives in family trees, in the absence of documentary evidence” (McMahon and McMahon Reference McMahon, McMahon, Forster and Renfrew2006: 159). More precisely, generational information can be reconstructed (e.g. on the basis of information from names, with surnames typically reflecting male-line descent and patronymics allowing the placement of people in their respective generations), whereas specific birth and death dates for the various family members need to be known directly. Such dates can be estimated based on biological limitations that restrict the human generation interval to between 15 and 45 years in the case of maternal descent (it can be somewhat longer in the case of paternal descent). Yet such estimates are never precise or very reliable; while the general expectation is that of “stratified generations” (i.e. one's aunts and uncles being approximately the same age as one's parents, one's great-aunts and great-uncles being approximately the same age as one's grandparents, and so on), deviations from that pattern, such as an aunt who is younger than her niece, are not hard to come by. As for languages, the situation is if anything more uncertain as there is no apparent analogue for the human generation interval: there are no external (e.g. biological) limitations on how quickly or slowly a language can change to the extent that it is no longer recognized as an extension of its ancestor. Consequently, the development of a “grandmother” language into its “granddaughters” may take longer or shorter periods of time. For instance, the development of Proto-Germanic into its “granddaughter” languages, such as English and Norwegian, took considerably longer than the development of Proto-Slavic to its “granddaughters”, such as Polish and Russian: Proto-Slavic was spoken approximately from the fifth to the ninth century CE, at which time Old English and Old Norse were already distinct languages.

It is clear from well-documented instances of relatively recent linguistic history that languages change not at a constant rate (see the summary of criticisms of glottochronology in Chapter 5), but in fits and starts. Ringe et al.'s (Reference Ringe, Warnow and Taylor2002: 60) Uniformitarian Principle (see Chapter 3) forces us to assume that the same variability in the rate of linguistic change applies to early periods in human (pre-)history. However, it also forces us to assume that the amount of change – not only lexical but also phonological and grammatical – that happened during comparable time spans in prehistory and in historical times should be comparable as well. Based on this assumption, Garrett (Reference Garrett, Forster and Renfrew2006) argued against the Gray–Atkinson model, indicating that the date that it assigns to PIE is several millennia too early.Footnote ¹ Garrett's argument is illustrated with five numerals (‘three’, ‘five’, ‘seven’, ‘eight’, and ‘nine’) in PIE and three Indo-European branches: Greek, Indo-Iranian, and Italic. For each branch, he considers a modern representative (Modern Greek, Waigali, and Spanish, respectively), as well as intermediate proto-languages of approximately the same age (Proto-Greek, Proto-Indo-Iranian, and Proto-Italic). He compares the temporal distance between PNIE (i.e. Proto-Nuclear-Indo-European, the common ancestor of all Indo-European languages except those in the Anatolian branch) and the intermediate proto-languages, on the one hand, and between the intermediate proto-languages and the modern languages, on the other hand. Garrett notes that, if one assumes the Gray–Atkinson date for PNIE (i.e. 7,000–8,000 years ago) to be correct, the two temporal intervals are approximately of the same length, on the order of 3,500–4,000 years (though Proto-Italic may be somewhat younger). Yet, Garrett (Reference Garrett, Forster and Renfrew2006: 144) notes that during the latter temporal interval (i.e. intermediate proto-languages to modern descendants) “significantly more phonological change has taken place than occurred” in the interval from PNIE to intermediate proto-languages. Moreover, a major restructuring of nominal and verbal inflection occurred en route from the same intermediate proto-languages to their modern descendants, while the morphosyntactic changes from PNIE to the intermediate proto-languages were much less substantial. If PNIE predates the intermediate proto-languages by approximately the same time interval as the one that separates those intermediate proto-languages from their modern descendants, what accounts for the two periods being radically different in the amount of phonological and inflectional change that characterizes them? A relatively conservative period of little change would have been necessary, followed by a period of much more rapid transformations, in contradiction to the Uniformitarian Principle. As a result, the Gray–Atkinson model requires that we make “the unscientific assumption that linguistic change in the period for which we have no direct evidence was radically different from change we can study directly” (Garrett Reference Garrett, Forster and Renfrew2006: 144). Here again, the Steppe hypothesis, which posits that PIE was spoken approximately two millennia after the date produced by the Gray–Atkinson model, is more compatible with the facts described by Garrett (Reference Garrett, Forster and Renfrew2006).

From relative to absolute chronology: the case of Romani

Although relative chronologies of various historical linguistic changes are fascinating in their own right, absolute chronologies, whereby linguistic events are assigned calendar dates, are obviously more valuable. But establishing such absolute chronologies based solely on linguistic facts is virtually impossible, as words and grammatical structures rarely bear direct signs of when they were born. The only way to establish absolute linguistic chronologies is to correlate linguistic facts with dates fixed by other means. In the simplest case, written materials illustrating particular linguistic phenomena can be dated by radiocarbon methods – and in some cases by the text itself. For example, the Kirkdale sundial, found at St. Gregory's Minster in North Yorkshire, England, which exhibits a loss of the genitive case in Old English, can be dated by its reference to “the days of Earl Tostig” – that is, Tostig Godwinson. Brother of King Harold Godwinson, he held the title of Earl of Northumbria from 1055 to 1065 and died in the battle of Stamford Bridge in September 1066. From this and similar texts, we can conclude that the loss of distinct genitive forms happened as early as the mid-eleventh century, during the late Old English period.

Because such direct indications of when a given linguistic development occurred are not always available, historical linguists must engage in more subtle sleuthing, looking for both grammatical changes and loanword patterns that can be dated independently. Work on Romani provides a good example. The origin and migration history of the Roma must be reconstructed based on linguistic evidence, as little information can be gleaned from written records, either their own or those of other peoples (as discussed in Chapter 3). Since the late eighteenth century, it has been known that the Roma trace back to the Indian subcontinent, from where they migrated westward (Rüdiger Reference Rüdiger1782; Marsden Reference Marsden1783). The earliest mentions of the Roma in Europe date from the mid-1300s, by which time they must have reached the Balkans. The reasons for the Roma exodus are controversial. Most scholars now believe that the Roma originated as one of the service-providing castes in India that left the subcontinent as camp followers, making their living by providing crafts and services to military forces (Matras Reference Matras2010: 39). This middle-of-the-road explanation treats the early Roma neither as aimless wanderers, as they are often depicted in the non-specialized literature, nor as a prestigious caste of warriors and priests, as portrayed by some Romani activists (cf. Kochanowski Reference Kochanowski1994).

Whatever the reasons behind the Roma migration that ultimately brought them to Europe, linguistic evidence helps us understand the timing of the differentiation of the Romani language – and consequently, the timing of their migration. An examination of the sound system and grammar of Romani reveals that an earlier form of the language was spoken in India as late as 1,000 years ago. As discussed in Chapter 5, Romani underwent a number of widespread changes that affected the northern Indo-Aryan-speaking area during the second half of the first millennium, including the restructuring of the gender system, the general reduction of the older system of nominal case inflection to an opposition between nominative and non-nominative (oblique), the simplification of some consonant clusters, and the collapse of the past participle. These changes characterize the transition from Middle Indo-Aryan (MIA) to New Indo-Aryan (NIA) languages.

However, an examination of the loan component of the Romani lexicon may suggest an earlier date for the Roma exodus from India, sometime between the sixth and ninth centuries CE (see Campbell Reference Campbell, Renfrew, McMahon and Trask2000: 5–6). As mentioned in Chapter 4, the Romani lexicon has a significant Greek component, suggesting a prolonged sojourn in a Greek-speaking area characterized by intense and widespread Romani–Greek bilingualism (see Matras Reference Matras2010: 35–37). The Romani vocabulary contains numerous lexical roots of Greek origin, including words from the Swadesh-200 list, such as luludí ‘flower’ (< Greek lulúdi), kókalo ‘bone’ (< Greek kókkalo), and drom ‘road’ (< Greek drómos). Grammatical vocabulary has also been borrowed, including pale ‘again’ (< Greek pale), komi ‘still’ (< Greek akómi), and panda ‘more’ (< Greek pánta ‘always’), as well as the numerals eftá ‘seven’, oxtó ‘eight’, and enjá ‘nine’. Derivational markers have also been borrowed, such as the abstract nominalizer -imos, the participal marker -imen, the adjectival markers -itiko/-itko, and the ordinal marker -to. Greek has influenced Romani syntax as well: the default word order in Romani is Verb–Object, as opposed to all other Indo-Iranian languages that obligatorily show Object–Verb order. Other grammatical features that can be attributed to Greek influence are the prenominal definite article, the postnominal relative clause introduced by a general relativizer kaj (as in o manuš kaj giljavel ‘the man who sings’), and the contrast between a factual complementizer kaj and a non-factual one te. This massive linguistic borrowing prompts a speculation that the Roma stayed in a Greek-speaking area, most likely somewhere in the Byzantine Empire, for many generations, possibly as long as two or three centuries (Matras Reference Matras2010: 39).

The second largest set of loanwords in Romani comes from Armenian; these loans include words for everyday items such as bov ‘oven’, grast ‘horse’, and kotor ‘piece’. Different scholars have provided varying estimates as to the number of Romani lexical roots of Armenian origin, ranging from twenty (Matras Reference Matras2002: 23) to fifty-one (Boretzky Reference Boretzky1995). The different counts are partially due to the fact that it is sometimes difficult to distinguish loanwords of Armenian and Iranian origin, because of the massive Iranian influence on Armenian; distinguishing Iranian loans from shared Indo-Iranian cognates is more difficult still. Similarities among Iranian languages make it difficult to determine whether a given loan comes from Persian, Kurdish, or even Ossetian. Nonetheless, several Persian etymologies for Romani words have been accepted, including ambrol ‘pear’ (< Persian amrūd), avgin ‘honey’ (< Persian angubīn), and pošom ‘wool’ (< Persian pašm); see also Hancock (Reference Hancock and Matras1995).

The presence of significant Armenian and Iranian components in the Romani lexicon has often been interpreted as suggesting successive phases of migration, first from India to Persia (where Persian words could be borrowed), then to the northern Fertile Crescent or southern Caucasus (where Armenian words were absorbed), and finally to the Byzantine Empire.Footnote ² Allowing at least a few generations’ sojourn in each of these locations – long enough for a significant amount of borrowing to occur – would indicate an earlier date for the initial exodus from India than the one established on the basis of phonological and grammatical properties of Romani shared with other NIA languages. Paradoxically, Roma could not have left India before 1000 CE to allow for the changes that characterize Romani as a NIA language, and yet the Roma must have left before 1000 CE to allow enough time for consecutive sojourns in Persia, the northern Fertile Crescent, and Greece. The conspicuous absence of Arabic loanwords – except for isolated items such as dzet ‘oil’ from the Arabic zēt ‘olive oil’, which can be explained as indirect borrowings via Persian or Armenian – also suggests that the Roma left Iran before, or at least not long after, the Muslim conquest of 644 CE, which in turn means that they must have left India well before 1000 CE.Footnote ³

So how can the thesis that the Roma stayed in Iran for several generations before leaving in the seventh or eighth century be reconciled with the phonological and grammatical evidence for the Roma leaving India as late as 1000 CE? One solution, originally proposed by Matras (Reference Matras1996; Reference Matras2002: 25; Reference Matras2010: 35), is to look not at the present-day location of the Persian and Armenian languages, but at their historical distribution. This theory takes the Roma in one migration phase in the eleventh century CE from India directly to eastern Anatolia, which had just come under the rule of Seljuks of Rum (1077–1307), who supplanted the Byzantine Greeks in the region after the fateful Battle of Manzikert in 1071 CE. Although the Seljuk elite were Turkic-speaking, the official language of the Sultanate of Rum was Persian. Because Byzantine Greek, Armenian, and Kurdish were all widely spoken in eastern Anatolia at the time, the various loan components of the Romani lexicon could in principle have been picked up in the same period and in the same place. Keeping in mind that the Roma had not converted to Islam and would have been in contact mostly with the non-Muslims (chiefly Christians, but also Yezidis and Jews) of eastern Anatolia, the lack of Arabic loans is not particularly surprising. Assuming that the Roma were camp followers making their living by providing crafts and services to military forces, their arrival in eastern Anatolia at roughly the same time that the Seljuks established their rule in the area makes sense as well. We do find it odd, however, that Romani has few roots of Turkic origin, as would be expected under such a scenario.

As a result of such considerations, we can reasonably settle on the date circa 1000 CE for the Roma exodus from India, and consequently of the Romani split from the rest of the Indo-Aryan family tree – a whopping 2,500 years later than the date calculated by the Gray–Atkinson model (or 1,500 years, according to their newly substituted results). This later date, supported by the historical linguistic evidence outlined above, receives further support from genetic studies, which place the “founding event” (i.e. the Roma exodus from India) “approximately 32–40 generations ago” (Morar et al. Reference Morar, Gresham, Angelicheva, Tournev, Gooding, Guergueltcheva, Schmidt, Abicht, Lochmuller, Tordai, Kalmar, Nagy, Karcagi, Jeanpierre, Herczegfalvi, Beeson, Venkataraman, Carter, Reeve, de Pablo, Kucinskas and Kalaydjieva2004). Assuming 25–30 years per generation, this figure nicely matches the 1000 CE date derived from linguistic studies.Footnote ⁴

The upshot of this discussion of the Roma exodus from India and the consequent differentiation of Romani from other Indo-Aryan languages is that traditional historical linguistic methods, coupled with evidence from written sources, archeology, genetics, and other disciplines, can provide absolute dates, contrary to the claims of the Gray–Atkinson approach that “linguists don't do dates”. In the following pages, we turn our attention to applying such traditional historical linguistic methods to the dating of PIE.

Linguistic paleontology and the “wheel problem”

Although in principle lexical and grammatical borrowings, as well as shared phonological and grammatical changes, can shed light on the migrational history of a language, as discussed above, the farther back one looks into the past, the less reliable such evidence becomes, due in part to the changing locations of the different language groups involved. For example, intense linguistic contact between early Indo-European languages, especially Proto-Indo-Iranian, and Proto-Uralic, has been well established (see Häkkinen Reference Häkkinen2012), yet without knowing where exactly Proto-Uralic was then spoken, we can only use this information to make conjectures about the location of the PIE Urheimat. (Language contact in the context of the PIE problem is discussed further in Chapter 9.)

Other varieties of linguistic information, however, can help us fill in the gaps in historical knowledge encountered in regard to non-literate, undocumented peoples, such as the original Indo-Europeans. In the absence of written records, scholars can turn to so-called linguistic paleontology, a field of study that examines the reconstructed vocabulary of a proto-language to find clues about the culture, social organization, and geography of its speakers. Introduced by Adolphe Pictet in Reference Pictet1877, linguistic paleontology remains one of the best ways to date linguistic divergence events at the requisite time depth, despite the challenges that it faces from semantic shifts and new archeological discoveries, as we shall see below. For example, the techniques of linguistic paleontology have allowed scholars to date the development of the Uralic language family, and especially of its Finno-Ugric branch, in reasonable detail. This work is based on correlating such historical linguistic information with evidence derived from other fields, such as palynology (for the prehistoric distribution of trees), and archeology (in regard to the local origins of agriculture, metallurgy, and the like) (see Campbell Reference Campbell, Hickey and Puppel1997a, Hajdú Reference Hajdú1969, Hajdú and Domokos Reference Hajdú and Domokos1987, Joki Reference Joki1973, and Anttila Reference Anttila1989, among others).

Linguistic paleontology is based on the assumption that the things that the reconstructed words denote actually existed at the time when these words were used to denote them.Footnote ⁵ For example, if a PIE word *sneig^wh- for ‘snow’ can be reconstructed, it stands to reason that the Indo-Europeans would have known snow. If they had a word *h₂erh₃- for ‘to plow’, they must have known and used some form of plowing. And if their vocabulary included words like *ph₂tér- ‘father’, *meh₂ter- ‘mother’, *b^hráh₂ter- ‘brother’, *su̯ésor- ‘sister’, *suH- ‘son’, and *d^hugh₂tér- ‘daughter’, then they must have cared about such familial relationships.Footnote ⁶ Reconstructing not only individual lexical items, but also well-elaborated lexical fields (semantic domains) with several interrelated terms is particularly valuable for the purpose of reconstructing the culture associated with a proto-language. For example, the reconstruction of terms pertaining to maize and maize agriculture in Proto-Mayan, including words for ‘maize’, ‘corncob’, ‘ear of corn’, ‘atole [a corn drink]’, ‘to sow’, ‘to harvest/pick corn’, ‘to grind’, ‘metate [grindstone for corn]’, ‘to roast [grains]’, ‘flour’, and so on, confirms that Proto-Mayan speakers were already maize agriculturalists before the diversification of the Mayan language family (Campbell Reference Campbell, Hill, Mistry and Campbell1997b). As for the Indo-Europeans, aspects of both their material and non-material culture can be reconstructed in this fashion, though in the following pages we shall focus on material culture, whose elements in the archeological record can be reliably dated by radiocarbon methods. (For a more detailed discussion of the culture of the Indo-Europeans, see Mallory Reference Mallory1989, Beekes Reference Beekes1995, Mallory and Adams Reference Mallory and Adams1997, Mallory and Adams Reference Mallory and Adams2006, and the references therein.)

Recall that the Anatolian hypothesis, originally proposed by Renfrew (Reference Renfrew1987) and advocated by the Gray–Atkinson approach, associates PIE with the spread of agriculture in the seventh millennium BCE, whereas the Steppe hypothesis associates the Indo-Europeans with the domestication of the horse and more broadly with the so-called secondary products complex – plowing, carting, wool, and dairy – that arose in the late fourth and early third millennium (Sherratt Reference Sherratt, Hodder, Isaac and Hammond1981, Reference Sherratt1983, Reference Sherratt, Goodman, Lovejoy and Sherratt1997; Anthony Reference Anthony2007). In this regard, the parts of the reconstructed Indo-European vocabulary that pertain to secondary products are particularly instructive.

To begin with dairy, PIE vocabulary had a word for cow, *g^wóu- (cf. Latin bōs, Greek boũs, Sanskrit gaús, Latvian gùovs, Old Church Slavonic govęždь, Tocharian A ko, Tocharian B kau, Old English cú).Footnote ⁷ Also reconstructed are words *h₂ou̯i- ‘sheep’ (cf. Latin ovis) and *h₂eg^wno- ‘lamb’ (cf. Latin agnus), although the reconstruction of the word for ‘goat’ is more problematic (see Beekes Reference Beekes1995: 36; Wodtko et al. Reference Wodtko, Irslinger and Schneider2008: 233–235). The word for ‘milk’ is also difficult to reconstruct conclusively, though ‘milking’ in PIE was apparently *h₂melg̑- (cf. Greek amélgō, Old English melcan, etc.). It is important to note that dairy consumption was made possible by a genetic mutation that led to a development of lactase persistence, and hence lactose tolerance, in adults. Several such mutations occurred independently in Europe, India, and parts of Africa (cf. Ingram et al. Reference Illič-Svityč2009). Crucially, these mutations were not directly linked to the spread of agriculture some 3,000 years earlier: to this day, southern Europe has significantly lower levels of lactase persistence (and hence higher levels of lactose intolerance) than northern Europe, although agriculture developed much earlier in the south than in the north of the subcontinent. Moreover, the Fertile Crescent, where agriculture first appeared and the European cropping complex is thought to have originated, remains relatively lactose intolerant. Even in populations that eventually developed some degree of lactase persistence, the relevant mutations and consequently the ability to consume dairy products did not spread quickly, nor did they affect entire populations. For example, the famous Tyrolean Iceman, who lived several millennia after the arrival of agriculture in southern Europe, was genetically lactose intolerant (cf. Keller et al. Reference Keller, Graefen, Ball, Matzas, Boisguerin, Maixner, Leidinger, Backes, Khairat, Forster, Stade, Franke, Mayer, Spangler, McLaughlin, Shah, Lee, Harkins, Sartori, Moreno-Estrada, Henn, Sikora, Semino, Chiaroni, Rootsi, Myres, Cabrera, Underhill, Bustamante, Vigl, Samadelli, Cipollini, Haas, Katus, O'Connor, Carlson, Meder, Blin, Meese, Pusch and Zink2012).

Besides milk, sheep must have also provided the original Indo-European speakers with *HulHn- ‘wool’, which was the basic material out of which clothing was made (cf. Latin lāna, Gothic wulla; other derivatives attested also in Anatolian, Greek, Indo-Iranian, Baltic, Slavic, Celtic, and Germanic; see Garrett Reference Garrett, Forster and Renfrew2006: 145). Another useful domesticated animal was the ox, *uksen- (cf. Tocharian okso, Welsh ych, Gothic auhsa, etc.).

But even more important than the ox was the horse, *h₁ék̑u̯o- (cf. Lycian esbe, Cuneiform Luvian azzuwas, Hieroglyphic Luvian á-zú-wa-, Tocharian A yuk, Tocharian B yakwe, Old Irish ech, Latin equos, Sanskrit áśvas, Avestan aspō, Old Persian asa, Old English eoh, Lithuanian ašvà, etc.).Footnote ⁸ If the ox made plowing and carting possible, the horse made these processes easier and faster. The reconstructed PIE vocabulary contains the words ‘plowing’ (reconstructed as *h₂erh₃- based on such cognates as Lithuanian ariù, Greek aróō, Latin arō, Old Irish airim, and Gothic arjan) and ‘carrying in a vehicle’ (reconstructed as *u̯eg̑^h- based on cognates such as Sanskrit váhati, Lithuanian vežù, Latin vehō, Greek ókhos ‘wagon’, and Gothic ga-wigan ‘to move’).Footnote ⁹

Much ink has been spilled on the topic of whether *h₁ék̑u̯o- referred to wild or domesticated horses, and if horses were ridden by early Indo-Europeans or served to pull military chariots or peaceful plows (Renfrew Reference Renfrew and Mithen1998; Renfrew Reference 309Renfrew, Renfrew, McMahon and Trask2000: 431–434; Anthony Reference Anthony1986, 2007, Reference Anthony2011; Levine Reference Levine1990; Anthony and Vinogradov Reference Anthony and Vinogradov1995; Owen Reference Owen1991; Gamkrelidze Reference Gamkrelidze, Hänsel, Schlerath and Zimmer1994; Mallory Reference Mallory1989). A related discussion (see Clackson Reference Clackson, Renfrew, McMahon and Trask2000: 443–446; Anthony Reference Anthony2011) concerns the role of the horse in early Indo-European myth and ritual. As interesting as these debates are, they are not particularly relevant to the topic under consideration here: whatever the role of the horse in early Indo-European society, the linguistic reconstruction clearly indicates that speakers of PIE (or at least PNIE, as we shall see below) were quite familiar with horses. Since horses could have been domesticated only in an area where wild horses were common, the fact that PIE had a word for ‘horse’ narrows down the timing and location of the language group, regardless of whether the word referred to wild or domesticated animals.

Whether the early Indo-Europeans also had words for ‘wheel’ and other parts of wheeled vehicles – and were therefore familiar with wheel technology – has also provoked intense debate. The importance of these reconstructed words derives from the fact that there is no evidence for wheeled vehicles before the middle of the fourth millennium BCE (Mallory Reference Mallory1989: Anthony Reference Anthony2007). Yet, the existence of such terms in PIE and/or its immediate daughter languages strongly suggests that their speakers were familiar with wheeled vehicles.

As a result of these facts, the advocates of an early Anatolian origin of Indo-European must explain why there is no archeological evidence of wheels for several millennia after the supposed origination of the language family. They have generally responded by pointing out that “parallel semantic shifts or widespread borrowing can produce similar word forms across different languages without requiring that an ancestral term was present in the proto-language” (Atkinson and Gray Reference Atkinson, Gray, Forster and Renfrew2006: 92). A particularly good example of how this phenomenon works in modern, well-documented languages is outlined in Heggarty (Reference Heggarty, Forster and Renfrew2006: 189): the English word mouse acquired a new sense in computing, then this new sense was “borrowed” (or more precisely, calqued) into other European languages; as a result, German Maus, Dutch muis, and Russian myš (as well as their equivalents in other languages) all acquired this new meaning more than a thousand years after the languages split. However, a closer look at the reconstructed PIE wheeled-vehicle terms, drawing heavily on the work of Don Ringe (cf. Ringe Reference Ringe2006), reveals that while a few of the items might be explicable as independent innovations or borrowed terms, a set of wheeled-vehicle terms must have been inherited from PIE or its near-descendants. To show such necessity, we need to examine the evidence of linguistic contact between Indo-European branches – or the lack thereof – and of puzzling, irregular reflexes of PIE forms in its daughter languages (or, again, the lack thereof). In general terms, if a reconstructed word fits perfectly into the phonological and inflectional systems of its language, with nothing falling out of line, the default hypothesis is that it is an inherited word. Admittedly it remains conceivable that it could still be a loanword, but the odds are vanishingly low.

With such considerations in mind, let us begin our exploration of the Indo-European wheeled-vehicle vocabulary with the root of the family tree. As the Anatolian branch is widely accepted as the first one to have split off from the trunk, any lexical item that has reflexes in at least one Anatolian language and at least one non-Anatolian language can be reconstructed for PIE. As it turns out, at least two wheeled-vehicle terms satisfy that criterion, together constituting good evidence that the early Indo-Europeans used draft animals to pull vehicles (or plows). The first such word means ‘thill’ (i.e. the pole that connects the yoke or harness of the draft animals to a vehicle). Ringe reconstructs this term's basic form as *h₂éyHos ~ *h₂éyHes or *h₃éyHos ~ *h₃éyHes, and its collective form as *h₂iHséh₂ or *h₃iHséh₂ (see also Pokorny Reference Pokorny1959; Melchert Reference Melchert1994). Its derivatives are found in Hittite (Anatolian), as well as in Proto-Slavic and Vedic Sanskrit (non-Anatolian). Ringe argues that because all these derivatives in descendant languages mean ‘thill’, we must reconstruct the meaning of the PIE term unambiguously as ‘thill’. Moreover, this word is not derived from any known verb root, so the possibility of parallel semantic shifts must be excluded. Finally, this word could not have been borrowed into or out of Anatolian languages, as the separation of this branch from the rest of the family was “clean”, as evidenced from the fact that Anatolian shares no distinctive innovations with any other subfamily of Indo-European (cf. Melchert Reference Melchert1994: 60–91; Ringe Reference Ringe2000). Thus, this single word constitutes one piece of prima facie evidence that speakers of PIE used draft animals to pull vehicles.

A second term that pertains to draft animals and is attested in both Anatolian and non-Anatolian languages is ‘yoke’; it is reconstructed as *i̯ugóm, based on Hittite iukan, Vedic Sanskrit yugám, Latin iugum, Welsh iau, Old English ġeoc, Old Church Slavonic igo, and so on. Although it is related to the verb root *i̯eu̯g- ‘join’, this term does not mean just any kind of apparatus for joining. Instead, it consistently means ‘yoke’ in descendant languages; as a result, Ringe reconstructs that meaning for PIE. Moreover, he points out that “the formation of ‘yoke’ is a peculiar archaism, and that its shape – like its meaning – is a kind of linguistic fossil in Indo-European languages” (Ringe Reference Ringe2006: 3; see also Garrett Reference Garrett, Forster and Renfrew2006: 145).

It has been suggested that a third word, meaning ‘wheel’, can be reconstructed based on both Anatolian and non-Anatolian daughters of PIE: Hittite hurkis vs. Tocharian A wärkänt and Tocharian B yerkwantai. However, Ringe (Reference Ringe2006) convincingly shows that the words in the two Tocharian languages reflect different reconstructed Proto-Tocharian forms, and the form that can be reconstructed on the basis of the Hittite form is different from the other two. It is possible that all three attested forms are derivatives of the verb root *h₂u̯erg- ‘turn’, but all three must have derived as independent innovations. Thus, we are left with two words indicating that speakers of PIE were harnessing draft animals by using a till and a yoke, to conveyances or plows of some sort. Whether at this point these conveyances had wheels or were more similar to sleds (in winter) or the North American travois, remains to be seen.

But in the Gray–Atkinson model the highest-order split of the Anatolian branch is not the only diversification event that long predated the appearance of the wheel in the archeological record in the middle of the fourth millennium BCE (i.e. around 5,500 years ago). The second split, whereby PNIE differentiated into the Tocharian and non-Tocharian branches, is dated at around 7,900 years before present in Gray and Atkinson (Reference Gray and Atkinson2003) and at about 6,900 years in Bouckaert et al. (Reference Bouckaert, Lemey, Dunn, Greenhill, Alekseyenko, Drummond, Gray, Suchard and Atkinson2012).Footnote ¹⁰ These authors therefore face an additional challenge posed by words pertaining to wheeled vehicles that can be reconstructed for PNIE based on cognates in both Tocharian and non-Tocharian languages. At the level of PNIE, the word for ‘wheel’ is reconstructed as *k^wek^wlo- (collective form *k^wek^wléh₂). The cognates in descendant languages include Vedic Sanskrit cakrám, Avestan čaxrō, Homeric Greek kúklos, Old English hwēol, as well as Tocharian A kukäl and Tocharian B kokale meaning ‘chariot’.

Gray, Atkinson, and their colleagues dismiss this reconstruction of ‘wheel’ in PNIE by arguing that the cognates in descendant languages derived independently from the verb root *k^welh_1/2- ‘turn’. Although this scenario is in general possible, linguistic analysis reveals further complications. As Ringe convincingly shows, the formation of *k^wek^wlos from *k^wel- is unique, as it involves reduplication, a zero-grade root, and a thematic vowel. According to Ringe (Reference Ringe2006: 4), “the probability that it could have been formed independently more than once is virtually nil”.Footnote ¹¹ Nor is it likely that descendant languages would have inherited a word with the abstract meaning of ‘circle’ or ‘cycle’ and then shifted it to the concrete meaning of ‘wheel’; although semantic shifts from concrete to abstract are well attested, the reverse meaning shift, from abstract to concrete, is unusual at best (Sweetser Reference Sweetser1990; Traugott and Dasher Reference 313Traugott and Dasher2001; Garrett Reference Garrett, Forster and Renfrew2006: 145). Finally, it is not possible that the Tocharian languages borrowed this term from non-Tocharian languages or vice versa. Based on the lack of shared innovations and of other borrowings, Ringe concludes that the linguistic ancestors of Tocharians severed all contact with other Indo-European groups early and completely. Although the Tocharians subsequently came into contact with speakers of Iranian languages (Winter Reference Winter and Schmitt-Brandt1971; Ringe Reference Ringe1991: 105–115; Ringe Reference Ringe1996: 92; Kim Reference Kim1999), by that time numerous regular sound changes had occurred in all the relevant languages, allowing loanwords from this later period to be easily identified by their foreign “sound signatures”. However, the Tocharian word for ‘chariot’ shows no signs of foreign phonology and is therefore not a loanword. Consequently, the word for ‘wheel’ must have existed in PNIE, and therefore its speakers must have been familiar with the object.

Similar considerations allow us to reconstruct another PNIE root that is indirectly related to the wheeled-vehicle vocabulary, the word for ‘horse’. Based on cognates from both Tocharian and non-Tocharian languages (e.g. Tocharian A yuk, Tocharian B yakwe, Old Irish ech, Latin equos, Sanskrit áśvas, Avestan aspō, etc.), it can be reconstructed to PNIE as *h₁ék̑u̯o- (cf. also Wodtko et al. Reference Wodtko, Irslinger and Schneider2008: 230–233). Some scholars have attempted to reconstruct it even further, to PIE itself, based on forms found in Anatolian languages (Lycian esbe, Cuneiform Luvian azzuwas, Hieroglyphic Luvian á-zú-wa-). Although the word for ‘horse’ is found on numerous clay tablets in Hittite, it is unfortunately impossible to tell how it was pronounced in that language because the scribes never spelled the word out but used a logogram (word-sign) instead, one of many adopted as part of the cuneiform writing system. Could the word for ‘horse’ have been borrowed into or out of the Anatolian branch? The answer, however, is inconclusive as different sequences of sound changes could have produced the attested Lycian and Luwian forms, depending on whether the word was inherited from PIE or borrowed from a non-Anatolian language. Therefore, PIE reconstructions of this word remain controversial, while the PNIE reconstruction stands on firm evidential ground.

The same logic applies to later splits that also predate 3500 BCE (when wheeled vehicles are attested in the archeological record) in the Gray–Atkinson model: if a word pertaining to wheeled vehicles can be reconstructed for a common ancestral language based on cognates from the two branches found after the split, the speakers of that ancestral language must have had wheeled vehicles, meaning that the early dates advanced by the Gray–Atkinson approach are probably wrong – unless of course evidence is unearthed for wheeled vehicles from an earlier era. According to Gray and Atkinson's early study (Reference Gray and Atkinson2003), the third split in the Indo-European family tree is that of the Greco-Armenian branch, at around 5300 BCE. They also split off from the Indo-Iranian languages (plus Albanian) at around 4900 BCE and the Balto-Slavic languages at approximately 4500 BCE. Even the date assigned to the Celtic split from the Italo-Germanic branch (4100 BCE) predates the appearance of the wheel in archeological record by a good margin. According to the refined model presented in Bouckaert et al. (Reference Bouckaert, Lemey, Dunn, Greenhill, Alekseyenko, Drummond, Gray, Suchard and Atkinson2012), the differentiation of the Greco-Albanian plus the Indo-Iranian branch (as well as the split between the Greco-Albanian and Indo-Iranian branches), and the separation of the Balto-Slavic languages from the rest of the tree, occurred before 3500 BCE (although the split between Italo-Celtic and Germanic languages is dated after that time).Footnote ¹² All of these early dates are challenged by the reconstructions of words pertaining to wheeled vehicles based on cognates from Greek, Indo-Iranian, or Balto-Slavic languages and cognates from Italic, Celtic, or Germanic languages.

One prominent example is the word for ‘axle’, *h₂ek̑s-, based on cognates such as Greek áksōn, Old Church Slavonic osь, Lithuanian ašìs, as well as Latin axis, Welsh echel, Old English eax, and Old High German ahsa. Thus, the most recent common ancestor of all these languages must have split from the Indo-European stem not circa 5300 BCE (as in Gray and Atkinson Reference Gray and Atkinson2003) or circa 4600 BCE (as in Bouckaert et al. Reference Bouckaert, Lemey, Dunn, Greenhill, Alekseyenko, Drummond, Gray, Suchard and Atkinson2012), but after 3500 BCE.Footnote ¹³ Another reconstructed form that must have existed in the common ancestor of all non-Anatolian/non-Tocharian Indo-European (i.e. PSIE; see Figures 1, 2, and 4 in the Appendix) languages is the verb root meaning ‘transport/convey in a vehicle’. Its derivatives include Vedic Sanskrit váhati, Avestan vazaiti, Lithuanian vẽža, Old Church Slavonic vezetъ (and probably Greek ókhos ‘wagon’), as well as Latin veh- (cf. also Old Irish fecht ‘trip’, Gothic ga-wigan ‘to move’, and Old English wegan ‘to move’). The probative value of this item is questionable, however, as most of the derivative words shifted meaning in one of two directions: most of the nominal derivatives refer to wheeled vehicles, while verbal derivatives in many languages, including Latin and Old Church Slavonic, widened in meaning to denote conveying in other types of transport, such as boats. However, this term can still be used as evidence for wheeled vehicles as far back as the common non-Anatolian/non-Tocharian Indo-European language (i.e. PSIE) “in the context of the whole set of reconstructable forms” (Ringe Reference Ringe2006: 6).

A third reconstruction concerns another root meaning ‘wheel’, *(H)roth₂ó- (collective form: *(H)roteh₂‘set of wheels’). Its derivatives in descendant languages include Vedic Sanskrit ráthas ‘wagon, chariot’ and Lithuanian rãtas ‘wheel, circle’ (plural rataĩ ‘wagon’), as well as Latin rota (derived from the collective form but reinterpreted as the singular), Old Irish roth, Welsh rhod, and Old High German rad. While this word is derived by productive PIE morphological processes from the verb root *(H)ret- ‘run’, it is unlikely that all the derivatives came about via independent parallel developments. Moreover, the attestation of reflexes of the collective form in Italic and Indo-Aryan languages suggests that a derivative meaning ‘a set of wheels’ is also old, inherited from a common PSIE ancestral tongue. An instructive contrast is provided by the words for ‘hub’: most Indo-European languages derived this term from ‘navel’ by using different morphological means (e.g. different ablaut [regular vowel variation] grades of the root, different suffixes, etc.), which strongly suggests parallel developments.

To recap, comparative reconstruction yields not just isolated words but a much larger “terminological ensemble in a coherent semantic field” (Garrett Reference Garrett, Forster and Renfrew2006: 145). First, the words for ‘thill’ and ‘yoke’, and possibly ‘horse’ (whether domesticated or wild), can be reconstructed to the oldest common ancestor of all Indo-European languages, PIE. Second, the word for ‘wheel’ can be reconstructed for the last common ancestor of the non-Anatolian languages, PNIE. (The word for ‘horse’ may belong to this category as well.) Finally, additional terms for ‘axle’, ‘transport/convey in a vehicle’, and another term for ‘wheel’ (as well as the collective form for ‘set of wheels’) trace back to PSIE, the common ancestor of all non-Anatolian/non-Tocharian Indo-European languages. In the Gray–Atkinson model, all these proto-languages predate 3500 BCE, a date from which archeological evidence of horse domestication and wheeled vehicles is not available.

Archeological evidence of horsemanship and wheeled vehicles

It is now time to give an equally close look at the archeological evidence, especially as it pertains to horses and wagons. Recall that a word for ‘horse’ can be reconstructed at least for PNIE (even if its reconstruction for PIE remains questionable). Therefore, by the logic of linguistic paleontology, speakers of PNIE knew horses, whether wild or domesticated. According to the Gray–Atkinson model, PNIE split into Tocharian and non-Tocharian branches as early as 4900 BCE (Bouckaert et al. Reference Bouckaert, Lemey, Dunn, Greenhill, Alekseyenko, Drummond, Gray, Suchard and Atkinson2012) or even 5900 BCE (Gray and Atkinson Reference Gray and Atkinson2003). Moreover, according to the same model, proto-Tocharian speakers migrated eastwards via the northern Fertile Crescent and Iran. Therefore, the first question we must ask is whether horses were known in Anatolia, the Fertile Crescent, or the larger Middle East some 6,500–8,000 years ago.

Humans were certainly familiar with horses ages ago, depicting them in Paleolithic cave art as early as 25,000 BCE (the famous “Spotted Horses” in Pech Merle cave, Dordogne, France, were painted circa 18,000 BCE). These animals were wild horses that were hunted for meat. But about 10,000–14,000 years ago, wild-horse habitat contracted significantly when the favorable Ice Age steppe was replaced by dense forest over much of the Northern Hemisphere. In Eurasia, horses virtually disappeared from warmer areas such as Iran, lowland Mesopotamia, and the Fertile Crescent. Nor do horses appear in the faunal record of early Neolithic sites in Ireland, Italy, Greece, India, or, most significantly, in western Anatolia. In all these regions, the earliest horses appear to arrive from somewhere else, after 4000 BCE in western Anatolia or even later in most parts of Europe and India (Mallory and Adams Reference Mallory and Adams1997: 275). Large herds of wild horses survived only in the steppe zone in central Eurasia, although small populations persisted in isolated pockets of naturally open pasture in Europe, central Anatolia, and the Caucasus Mountains. However, in those areas where small populations of wild horses survived into the Holocene, they were rarely eaten by people, as they were too scarce. In central and eastern Anatolia, all equids together total less than 3 percent of the animal bones excavated from sites near human habitations. Horse bones, however, constitute only a small fraction of equid bones and total less than 0.3 percent of the total animal remains. In comparison, in archeological sites in the western steppe zone, horse bones constitute 13 percent of the total, while in the Volga–Ural burials they constitute 40 percent of the total animal bone finds (Anthony Reference Anthony2011). In central and eastern Anatolia, where horse bones are meager, more than 90 percent of equid remains come from onagers (Equus hemionus; the “Asiatic wild ass”) and another donkey-like animal that went extinct circa 3000 BCE, Equus hydruntinus (Russell and Martin Reference Russell, Martin and Hodder2005; Anthony Reference Anthony2007: 198). Onagers were by far the most common wild equids of the ancient Near East. Curiously, no word for ‘donkey’ can be reconstructed for PIE (Beekes Reference Beekes1995: 36). While absence of evidence should not be confused with evidence of absence, this vocabulary gap still strongly suggests that speakers of PIE knew horses but not donkeys or their close relatives, indicating yet again that Anatolia is the wrong place for the PIE homeland.Footnote ¹⁴

As for the invention of the wheel, and consequently of wheeled vehicles, the evidence points in the same direction, away from Anatolia and later than 6500 BCE. Instead, 3500 BCE does seem to be in the right time frame. The earliest, though uncertain, piece of evidence is a track that might have been made by wheels, preserved under a barrow grave at Flintbek in northern Germany and dated as far back as 3600 BCE. However, as discussed in detail by Anthony (Reference Anthony2007), the real explosion of evidence across the ancient world begins about 3400 BCE, with four independent lines of evidence appearing between 3400 and 3000 BCE: written and pictorial depictions, clay models, and actual wagon remains.

The oldest written evidence of a wheeled vehicle, a ‘wagon’ sign impressed on clay tablets, was found in Eanna temple precinct in Uruk, in what is now Iraq. Radiocarbon dates for charcoal found in the same layer as the tablets averaged about 3500–3370 BCE. These dates, however, may be too early, as the radiocarbon dates tell us not when the wood was burned, but rather when it was created by the tree that it came from. It is therefore possible that the charcoal found at the site came from the dead heartwood of a large tree, implying a later date. Also significant is the fact that the ‘wagon’ sign appears only three times in the tablets, whereas the sign for ‘sled’ – a transport device dragged on runners rather than rolled on wheels – occurs 38 times, suggesting that sleds were far more common than wagons in Uruk at that time.

The oldest pictorial evidence of a wagon is a two-dimensional image, about 3 centimeters by 6 centimeters, incised on the surface of a decorated clay mug of the Trichterbecker (TRB) culture found at the settlement of Broncice in southern Poland. The image appears to portray a four-wheeled wagon, a harness pole, and a yoke – recall that the words for ‘thill’ and ‘yoke’ are reconstructed for the common ancestor of all Indo-European languages, PIE. The mug is dated about 3500–3350 BCE, representing the earliest and latest radiocarbon dates of other finds, such as cattle bones, found in the same pit. Two other images of what appears to be draft animals pulling a two-wheeled cart, found in central Germany and in the Russian steppes near the mouth of the Volga River, date from approximately the same period: the image from Germany is placed between 3400 and 2800 BCE, and that from southern Russia dates between 3500 and 3100 BCE.

The earliest three-dimensional ceramic models of wagons, found in two graves of the Late Baden culture in Budakalász in eastern Hungary, date to about 3300–3100 BCE. Other graves in Budakalász and other Late Baden sites in Hungary, as well as graves of the partly contemporary Globular Amphorae culture (3200–2700 BCE) in central and southern Poland, also revealed paired oxen that had been sacrificed. A small circular clay object that might be a model wheel, perhaps from a small ceramic wagon model, found at the site of Arslantepe in eastern Turkey, also dates from the same period, 3400–3100 BCE.Footnote ¹⁵

Remains of the oldest actual wagons and carts, some 250 in total, have been found under earthen burial mounds known as kurgans, in the steppes of southern Russia and Ukraine. These vehicles had a revolving-wheel design; their wheels were solid rather than spoked and were 50–80 centimeters in diameter. The wagons themselves were small, only about 1 meter wide and 2 meters long, presumably designed to reduce drag while keeping the vehicle adequately sturdy. The earliest radiocarbon dates on wood from these wagons average around 3300–2800 BCE. Remains of wagons, wooden wheels, and axles have also been discovered in the mountains of Switzerland and southwestern Germany. These vehicles, with revolving axles, which are less efficient yet easier to make than revolving wheels, date to about 3200 BCE, suggesting that separate regional wheel-design traditions already existed by that time. Sometime after 3000 BCE, revolving axles were largely replaced by the revolving wheels, as revealed by finds in the Netherlands and Denmark.

To recap, the 3500 BCE date for the first wheels and wagons poses an insurmountable challenge for the Anatolian theory of Indo-European origins, proposed originally by Colin Renfrew and advocated more recently by the Gray–Atkinson approach. According to this theory, by 3500 BCE the Indo-European language family would have been “bushy, multi-branched, and three thousand years old” (Anthony Reference Anthony2007: 75). As a result, David Anthony argues that the sheer age and diversity of the family would have necessarily precluded horizontal transmission (i.e. borrowing) of vocabulary elements across multiple Indo-European subfamilies. But lexical items pertaining to technological innovation can spread quickly over vast areas; witness the rapid spread of computer terminology among – and well beyond – the languages of the same Indo-European family in the last seventy years or so. Still, the possibility that the common Indo-European vehicle vocabulary stems from borrowing alone is excluded on purely linguistic grounds, such as the phonological changes that occurred independently in the family's various descendant branches, as discussed above.

The only alternative solution available to proponents of the Anatolian theory is to contend that PIE was spoken in essentially an unchanged form for some 3,000 years, from the first farming era to the days of the first wagons/wheels (cf. Renfrew Reference Renfrew and Drews2001; Anthony Reference Anthony2007: 77–81). Such a scenario would require a homogeneous and extraordinarily slow rate of change over the enormous area supposedly encompassed by the Indo-European family by 3500 BCE, which was characterized at the time by small-scale tribal groups with highly diverse material cultures. This supposition, however, directly contradicts the Uniformitarian Principle (see Chapter 3): everything that we know about language change in historical times tells us that modes of speech do not remain static over such a long period of time and over such vast expanses of territory. Moreover, several studies (see Nettle Reference 307Nettle1996, Reference Nettle1999, Reference Nettle, McMahon, Trask and Renfrew2000a, Reference Nettle2000b) suggest that linguistic change proceeds at a higher speed in small-scale societies than in large, homogeneous ones. In a small-scale social order, an innovation introduced by one individual can rapidly become known to all members of the community; moreover, an innovation is more likely to survive in a small community because the number of individuals who might reject the innovation is relatively low. In regard to lexical replacement, the rate of change can be further accelerated by the effects of the common word taboo phenomena (see Comrie Reference Comrie, Renfrew, McMahon and Trask2000: 37–38). It has also been established that tribal languages unaffected by outside forces tend to be more varied than tribal material cultures; therefore, one would expect that the linguistic diversity of Neolithic/Eneolithic Europe would have been more pronounced than its material-culture diversity (Mallory Reference Mallory1989: 145–146; Anthony Reference Anthony1991; Anthony Reference Anthony2007: 80).Footnote ¹⁶ Finally, the territory that the Indo-European family occupied during the relevant period according to the animated maps produced by the Gray–Atkinson approach is vastly too large to have been covered by a single language; there is no ethnographic or historic warrant for postulating such a large, stable language territory among tribal farmers or pastoralists. Thus, PIE could not have been spoken in Neolithic Anatolia and still have existed in largely the same form three thousand years later when wagons were invented.

We are thus brought to the inevitable conclusion that the evidence from linguistic paleontology, especially as it pertains to horses, wheels, and wheeled-vehicle vocabulary, simply cannot be brought into agreement with the Gray–Atkinson model, which dates the highest-order Indo-European split as early as 6700–6500 BCE. Thus, the evidence from the wheeled-vehicle vocabulary determines the terminus post quem circa 4000 BCE. In other words, PIE could not have started splitting into its daughter languages prior to that time. Can we also determine the terminus ante quem, a time after which PIE could not have split? The answer is positive: the time depth of PIE can be determined by the dates of the earliest attested Indo-European languages. The earliest Anatolian (Hittite) texts inscribed on clay tablets date from 1700 BCE. The oldest Greek texts date from circa 1400 BCE (though it is possible that Greeks were present in Greece as early as 2200 BCE). The oldest evidence of Indo-Iranian (in the form of the “Mitanni” words preserved in cuneiform texts) also dates from circa 1400 BCE. Since these three groupings have each undergone a series of changes from the reconstructed language, PIE must be dated to sometime earlier than 1700 BCE. Most scholars thus set the terminus ante quem circa 2500 BCE. In other words, PIE must have split into daughter branches at some point between 4000 BCE and 2500 BCE, though it is likely that the split happened in the earlier part of this range. In the following chapter, we shall look for clues from linguistic paleontology as to where PIE was spoken.