How Old Are the River Names of Europe ? A Glottochronological Approach

The names of large rivers are among the diachronically most stable of all words. The present glottochronological study aims to quantify the lexical preservation rate of this set of words, more specifically of the names of the largest rivers of Europe. 210 river names are taken into consideration. Their preservation rate during the past 2000 years turns out to have been about 87%. Under the assumption that this observed rate can be extrapolated into the prehistoric periods, a statistical distribution of the name creation dates can be estimated. One conclusion is that 50% of the large modern rivers of Europe are likely to have received their present names prior to 8000 BC. This implies that the greater part of these names cannot be of IndoEuropean origin as has usually been assumed.

guages) or retarded (e. g. when an earlier version of the same language serves as the literary norm).Based on such counterexamples, there has been much critique to Swadesh's approach.I am aware that glottochronology is a highly controversial method and has been criticized at times very harshly. 4The most frequent objection has been to deny that the lexicon changes at a constant rate.This objection is, however, unjustified because it does not account for the statistical character of the claim.While the wording by Swadesh himself is not always very precise mathematically,5 I understand that, contrary to what has often been purported, Swadesh did not have an absolutely constant preservation rate in mind, but rather a reasonable average rate that admits some amount of statistical variation. 6This already follows from his very term "lexico-statistics", as also from the nature of radiocarbon dating to which he compared his method.If so understood, lexicostatistic procedures will lose much of their dubious reputation.Even if we reject the idea of a constant rate of change, it will nonetheless be possible to compute a statistical average value out of rates that have been observed, and it should be obvious that such an average value can be used if treated with the same care as is appropriate to any statistical assertion.
In essence, glottochronology is thus about creating a statistical model from things known and applying it to the unknown.This is a legitimate procedure, in fact the best available one in the absence of data about the unknown.As the model is more fuzzy, the results will become less reliable.Nevertheless, they will offer the best possible approximation to truth and should therefore be regarded as the default hypothesis until new evidence appears.The lexicostatistical method can therefore be applied even without having to insist on a strictly constant rate of lexical change.

2
The logarithmic Swadesh formula and its amendment Departing from a preservation rate of 0.86 per millennium, Swadesh (1955) and most of his successors assumed that after a time interval of m millennia, the rate of preserved words in Swadesh's 100-item list should be 0.86 m .The discipline of glottochronology has used this formula to estimate the time distance between related languages.For this purpose, word lists of two languages are assembled, and the rate r of shared cognates is measured. 7An observed rate r of shared cognates is then assumed to imply that the total time distance between both languages is log r / log 0.86 millennia.The may mean that one of the languages is a child of the other separated by this distance in time, or that both languages are independent derivatives from a common ancestor, the distance on each branch being half of the total distance.
There is a mathematical flaw in this formula that has usually been overlooked.In any given word list, the preservation probability is not equal for all items, but the list is composed of items with different degrees of stability.The preservation rates of the list items form a statisti-cal distribution around some average value.The empirically measured preservation rate of the list as a whole is just an estimation of this average value.The point is that, while two probabilities can be multiplied to find their combined probability (which leads to a logarithmic formula as the one cited above), the average of the combination of two probability distributions is not simply the product of both their averages.
I intend to clarify the mathematical aspects of this issue elsewhere in more detail and will illustrate the problem here only by means of a very simple example.Let there be a Swadesh list of two items.Let the real preservation probabilities per millennium of these items be 0.7 and 0.9 respectively.In real life, we cannot know the probabilities beforehand but have to determine them from empirical language data, that is by counting cognates within related languages whose time distance is known.If the data base is large enough, we will obviously find the average preservation probability per millennium of our word list as a whole as being close to (0.7 + 0.9) / 2 = 0.8.Now consider the preservation rate of the same word list for languages whose time distance is two millennia.The preservation probability after two millennia will be 0.7 2 for the first item and 0.9 2 for the second.The preservation probability of the whole list after two millennia therefore turns out to be (0.7 2 + 0.9 2 ) / 2 = 0.65, and not 0.8 2 = 0.64 as many would have expected.
More generally, if the preservation probability of a word list per millennium is p, the preservation probability after m millennia (with m > 1) will not be = p m , as has commonly been assumed, but > p m .This implies that the estimations of time depths in practically all existing applications of glottochronology have been to low. 8 It is difficult to quantify this effect in an exact manner, since this would require, alongside with the empirically measured average preservation rate of the word list, at least one additional constant representing the internal heterogeneity of that list, which would likewise have to be determined empirically. 9Let it suffice here to say that the preservation probability after m millennia (for m > 1) will be higher, to an unknown extent, than p m ; or that for an observed preservation rate r the distance in time will be higher, to an unknown extent, than log r / log p.

River names and glottochronology
I will now estimate the preservation rate of one specific word list, namely a list of the names of the major rivers of Europe.This will allow us to estimate a distribution of the dates at which these names came into existence.Due to the mathematical effect described in the preceding section, the distribution is not going to suggest absolute dates, but can only suggest a threshold stating that a given number of names are likely to exceed (to an unknown extent) a specific age.
River names are widely recognized as being the diachronically most stable of all words. 10ost rivers still have the same names today as attested in the earliest available records.This is true in particular for large 11 rivers, whose names are replaced even more rarely than those of small rivers.Names of large rivers are normally retained even when the area is invaded by speakers of another language.Factors contributing to their extraordinary stability include:  Their semantic isolation.While it is easy for a word to undergo a meaning shift such as 'to think' > 'to believe', 'to want' > 'to love', 'green' > 'yellow', 'woman' > 'wife ', etc., it would be much harder for a river name to change its meaning from 'Rhine' to 'Danube', or to any other meaning. 12Names of large rivers are known to a large number of people, possibly speaking different languages.Even if some of them invent a new name, this new name will not readily be adopted by all peoples concerned.
 While names of small rivers may be forgotten and need to be reinvented after a temporary depopulation of an area, names of large rivers do not elapse completely from human memory.

Selection of rivers
The preservation rate of river names must be estimated empirically on the basis of rivers attested in ancient records.While this could be attempted for several regions of the world, I will limit myself to the rivers of Europe, many of which were mentioned in classical Greek or Latin sources which are about 2000 years old.2000 years is therefore the time span for which the preservation rate will be calibrated.
Swadesh's hypothesis predicts that the results would not differ a lot if rivers of other parts of the globe were examined for which records of a similar age are available.I have to leave this for others to test. 13 10 Four citations may suffice: "C'est la rivière qui résiste le mieux aux substitutions, car elle échappe en général aux changements d'ordre officiel; d'autre part, elle a une individualité plus nette que la montagne ou le lieu dit, dont les limites sont plus imprécises, ou même que la ville, qui peut se déplacer plus ou moins.C'est ainsi qu'en France la très grande majorité des noms de fleuves et de rivières sont celtiques ou préceltiques" (Dauzat 1926:  41); "[Es] soll hervorgehoben werden, daß als die dauerhaftesten und altertümlichsten unter allen Eigennamen, dauerhafter als Siedlungsnamen und altertümlicher selbst als Gebirgsbezeichnungen, sich die Gewässernamen erwiesen haben" (Krahe 1954: 46); "Най-устойчиви от всички местни имена са названията на големите реки: те се пазят най-дълго време, независимо от смяната на населението" (Georgiev 1960: 21); "Gewässernamen (= Hydronyme) sind innerhalb der Onomastik häufig Gegenstand gesonderter Untersuchungen.Dies hat seinen Grund [...] darin, dass Fluss-und Seenamen häufig ein besonders hohes Alter besitzen.Sie können häufig aus der Sprache, die an den Ufern der Gewässer gesprochen wird oder in deren Vorstufen gesprochen wurde, nicht mehr erklärt werden und setzen sich dadurch von den Siedlungs-und Flurnamen deutlich ab" (Udolph 2004: 329). 11I define large as 'long' here, simply because the lengths of rivers are easy to retrieve. 12The exact geographical extension of what to consider a single river can change more easily.It happens, for instance, that two differently named, adjacent parts A and B of one course of water aquire a single name, either that of A or that of B (as in ⇒ Donau; this is what Schramm 1981: 67f.calls "Namensausgleich"), that what used to be considered as one single river comes to be seen as two rivers A and B with distinct names (as in ⇒ Weser), or that when two rivers A and B of roughly equal size confluence into a river C, there is hesitation as to whether A or B should bear the same name as C (cf. ⇒ Severskij Donec). 13East Asia or India would be areas with sufficient historical records, but I am not personally familiar with these.The three rivers with the absolutely oldest attestations on earth are Euphrates, Tigris, and Nile.The Euphrates (Arabic al-Furāt) and the Tigris (Arabic Diǧla) bore their present names already in Sumerian sources of the 3rd Within Europe, I consider all rivers exceeding 250 km in length 14 that flow into the Atlantic Ocean, the Mediterranean Sea, the Black Sea or the Baltic Sea.There are 210 of them.By contrast, rivers flowing -directly or indirectly -into the Caspian Sea (such as Vólga, Urál and their tributaries) and the Arctic Sea (such as Pečóra, Výčegda) are disregarded, because most of them are located too far away from the Classical world to have been mentioned in the ancient sources.In this respect, "Europe" is taken here in a narrower sense than usual.

5
The data table The following table enumerates the longest rivers of Europe, as defined above, sorted by length, together with their ancient names if attested."Ancient" is taken as approximately 2000 years old.These names are attested in the works of Herodotus (5th century BC), Caesar (1st century BC), Strabon (around 0), Plinius (1st century AD) and other authors of that time.I still accept the numerous records in Ptolemaios' Geographikē (2nd century AD) as being roughly 2000 years old, but not so records from Late Antiquity or Byzantine times.Attestations of the ancient names are, for the most part, easy to find.My standard sources are Wissowa et al. (1894-2000) and Talbert (2000).Explicit references will only be provided where they do not suffice.The entries in the rightmost column are to be read as follows: 'yes': An ancient name is attested, the river has one or more modern names, and the ancient name has survived in the modern name (or in one of the modern names).'(yes)': More than one ancient names are attested for the same river, and only one of them has survived to the present day.'no': An ancient name is attested, but the modern name is a different one.'-': No ancient name is attested for this river, so that the age of the modern name cannot be judged.
millennium BC (buranun and idigna, respectively; note that id2 is the Sumerian noun for 'river', so that the name core of the river Tigris is probably just -igna).The river Nile, however, changed its name more than once (cf.Peust 2010: 68f.).The reason may be that its name, the Nile being the only river of Egypt, was perceived as an appellative for 'river' rather than a proper name, and thus did not share the conservativity so characteristic of hydronyms.
14 River lengths are not absolutely straightforward to measure, so that some amount of uncertainty remains.This does not disturb my argument with the exception of possibly a small number of rivers whose length comes close to the 250 km limit.I took my data on river lengths for the most part from Wikipedia.(Greek) Tyras 1360 yes 15 Medieval German spelling Tůnowa (Borchers 2006: 14-17).This form appears to be the regular outcome of an earlier *Dōn-with the High German sound shifts d > t and ō > uo > ū.We would thus expect the modern German name to be *Tūn-, but we find Dōnwhich looks like a return to the Pre-German form.This development has been explained, not entirely convincingly, by local dialect influence (Reichardt 1985).An alternative explanation might be that the modern German name was transmitted via a non-German (Slavonic?)population.
Here, as in several of the names treated below (Bosna, Južnyj Bug, Don, Hron, Olt, Osəm, Morava, Mureş,  Someş, Somme), which are located in South-Eastern Europe for the most part, we find a vowel -a-of the ancient sources reflected as -o-or even -u-in the modern languages.Sound shifts a > o are attested in some branches of Indo-European (e. g. ā > ō in Germanic and Albanian, ă > ŏ in Slavonic), but it remains to be worked out to which languages these sound shifts are to be ascribed in the individual river names.As for the name of the Danube, Georgiev (1966: 189) assumes an influence of Dacian, whereas Schramm (1973: 37-43 and 1981: 231-233)  believes that this name was transmitted to the Slavs by speakers of a Germanic (Gothic) language.As evidenced by the German form, the name of the Danube requires positing a long ā/ō as its first vowel.This is supported by the Latin poets, who treat this vowel as metrically long (e.g.Cēdere Dānuvius sē tibi Nīle negat, a pentameter from Ovid, Tristia ex Ponto).The further sound shift towards u in this name is likely due to the influence of an early Slavonic language.The Slavonic languages represent Proto-Indo-European *ō as ā and *au as ū.We can assume (with Holzer 2007: 59 and Schramm 1981: 144f.) that there was a time at which the shift ō > ā had already been completed, whereas the shift *au > ū had only reached an intermediate stage *ō.At that time, the first vowel of the name of the Danube could have been identified by Slavonic settlers with their *ō, which subsequently shifted to ū and, after the loss of the original length contrasts in Slavonic, was borrowed as u into languages such as Hungarian and Romanian.A shift of borrowed *ō to ū also took place in Russian búkva 'letter of the alphabet' < Gothic bōka (idem).The origin of -r-in the Romanian name has no definite explanation (see Georgiev 1966: 193 and Schramm 1973: 43-47 and 1981: 230 for proposals). 16Here and in the following, I cite forms from Ancient Greek without accent marks, since these marks are not usually attested in the original manuscripts but were supplied by modern editors (cf.Clarysse 1997).We have no reliable knowledge about how the ancient Greeks accentuated foreign names. 17Only the ancient term for the upper river has survived, having been extended to the river as a whole. 18The modern name is first attested in 6th century Greek as Danapris and is spelled Dъnҍprь in Old Russian (cf.Fasmer 1964-87, I: 518; Nepokupnyj et al. 1979: 173).The early Slavonic pronunciation must be reconstructed approximately as Dunäprí.On this river name ⇒ also below in section 9. 19 On this river name ⇒ below in section 9. On the phonetic development ⇒ Donau. 20The modern name first appears in the 4th century as (Latin) Danastius (cf.Fasmer 1964-87, I: 518f.;  Nepokupnyj et al. 1979: 173f.; -r-possibly omitted by error).The ancient name Tyras was preserved up into Osmanic Turkish as Turla but is now obsolete.A diminutive derivation survives to the present day in the name of the Turunčúk, a tributary to the Dnestr.Since the initial segment dnof the modern name is almost certainly a former term for 'river' (⇒ below in section 9), the remaining -estr is similar enough to the ancient name to make it probable that it in fact continues that name.This is what Schmid (1976: 439f.)envisages who suggests that the second part might be an Iranian reinterpretation (cf.Ossetian aestyr 'big') based on the ancient Tyr-.I wish to forward an alternative proposal that -scould be the remnant of an Indo-European nominative case ending, here preserved in liaison as opposed to the loss of the same ending in final position, so that the modern name would derive from a compound *d˘năs-turăs 'river Turas'.Hrin; Borchers 2006: 56f.), which was a regular notation in Old High German to denote a sound different from r-, either a cluster /hr/ or possibly a voiceless [r̥ ]-.This spelling certainly cannot be ascribed to Greek orthographical practice.Is it possible that the local (probably Celtic) population pronounced the name with an initial voiceless r̥ -, which was adopted as such into Old High German and which was rendered as Rhby the Romans?R˳ (spelled rh-) is still a common sound in Welsh, another Celtic language, particularly in word-initial position (r̥ -is in fact the regular Welsh representation of an Indo-European initial *r-).A similar argument may apply to ⇒ Rhône.
As for the stem vowel, all modern names (including the Swiss German dialect form /rī/) presuppose an earlier -īwhich contrasts with -ē-of the Latin attestation.It can be assumed either that the original name contained a Proto-Indo-European *-ei-which developed into Celtic *-ē-(= base of the Latin transcription) and Germanic *-ī-(= base of all modern name forms) (thus Krahe 1954: 42; Jung 1970: 440 and others), or that the name originally had a Proto-Indo-European *-ē-which survived as such in Latin and developed into Celtic *-ī-(= base of all modern name forms). 22The name Desna, which appears several times as a hydronym across the Slavonic territory, might be explained from a (South) Slavonic term for 'right (hand)' although, in modern terminology, Desna is the left tributary of the Dnepr.Tolstoj (1984) discusses examples for Desna and for rivers that would allow for a Slavonic translation as 'left (hand)' (Leva, Levaja, Šuja etc.), finding that most 'right' rivers are actually left tributaries of a larger river and most 'left' rivers are right tributaries, although there are some exceptions to this principle of inversion that he needs to explain away.Tolstoj explains this inversion from naming habits of the older Slavonic peoples who, contrary to modern hydrographical usage, would have viewed rivers in streamupward direction.
I am sceptical about this argumentation.All rivers so named are isolate examples, there being no pair of neighboured, opposing tributaries that would bear both names 'right' and 'left'.This raises the suspicion that these names aquired their present shapes by folk-etymological reinterpretation of older names rather than by original coinage. 23For similar names of other rivers cf.Krahe (1964: 52f.).Several scholars have identified this name with Old Norse elfr = Swedish älv 'river' and some (e.g.Krahe, Kitson 1996: 75) further with Latin albus 'white'. 24For details and considerations on the original name form see Udolph (1990: 303-311).
25 Donec is a diminutive form of the ⇒ Don, of which it is a tributary.The diminutive name form has been attested since the 16th century, whereas in earlier sources the Severskij Donec was either called 'small Don' or considered a part of the Don (cf.Nepokupnyj et al. 1979: 503; Pospelov 1998: 377).It is rather common for a tributary B of a river A to be named by a diminutive derivation of A (e. g.Bužok as a diminutive of ⇒ Južnyj Bug; ⇒ Ingulec as a diminutive of ⇒ Ingul; ⇒ Moselle as a diminutive of ⇒ Meuse; Mürz as a diminutive of ⇒ Mur; see also ⇒ Dnestr).As explained by Pospelov, the attribute Severskij does not, as it might seem today, mean 'northern' (which would not make sense geographically), but was derived from the name of a people settling in that area. 26The Portuguese name is pronounced /tεʒu/.This name and other toponyms in the Iberian peninsula evidence striking phonetic developments that differ from the regular sound shifts of the inherited vocabulary.In topo- -940 nyms, ancient a can be reflected as e, and ancient g and s as Portuguese /ʒ/ ~ Spanish /x/ (written g or j), whereas Latin a, g and s are generally retained as such in common vocabulary.Other testimonies for these peculiar developments include ⇒ Esla, ⇒ Genil, ⇒ Júcar, some names of smaller rivers not discussed in the present article such as Jalón < Salo, Sella < Salia, Tâmega < Tamaga, and the name of the Portuguese town Beja < Paca.(But note that Latin s > Spanish j is found as a sporadic sound change in a few items of the common vocabulary as well, such as Span.pájaro 'bird' < Lat.passer.)These peculiarities have traditionally been explained (e. g. by Meyer-Lübke 1930: 416f.; Guérios 1956: 142f.;  Coutinho 1967: 193) by the assumption that these proper names were arabicised during the Muslim rule over the Iberian peninsula.The Romance phonemes /a/, /g/ and /s/ would thus have been realized as [ae], [ǧ] and [š], for which modern continuations as /e/, /ʒ~x/ and /ʒ~x/ would be conceivable.But it is hard to see why Arabic pronunciation habits should have affected Spanish or Portuguese toponyms more easily than Spanish or Portuguese appellatives.I consider it more probable that sounds such as ae, ǧ and š existed in these names already in antiquity, of which the Latin spellings were only approximate notations. 27First attested as Duna by Saxo Grammaticus (12th cent.).On this river name ⇒ below in section 9. 28 French oi (here and in ⇒ Oise) is the regular outcome of a Latin short ĭ. 29 For similar names of other rivers cf.Krahe (1964: 50). 30First attested in the 14th century as Poxor (Pospelov 1998: 450). 31I follow Pospelov's (1998: 417) suggestion that the initial element was misunderstood as a preposition po 'on' by a speakers of a Slavonic language.By contrast, Georgiev (1961: 91f.) does not consider the early name forms with Pto be related.He believes that the name might originally have contained a medial consonant and thus was identical to the name of the ⇒ Temeş. 32Most likely the source of the modern name despite difficult vowel correspondences.A 10th century Greek source (cited by Schramm) attests both varieties side by side as Bourat ~ Broutos.Former explanations of this variation include a secondary assimilation to the name of the ⇒ Răut (cf.Schramm 1973: 49-54) or an apophony on the level of Proto-Indo-European (*purn̥ t > Pyretos/Bourat vs. *puront > Prut/Broutos, cf.Schmid 1976:  438).I wish to forward a third and perhaps easier suggestion that Prut may have developed from an earlier *Purt by the well-known Slavonic liquid metathesis whose effect we also observe in Labe for ⇒ Elbe and in ⇒ Łyna for Alna. 33The -а-of the Russian form is a false orthography for *-о-, both of which merge in unstressed position.This vowel, which is mobile in the name forms of Belorussian (N'óman, genitive N'ómna) and Polish (Niemen, genitive Niemna), must reflect an earlier *ъ (ŭ) (c.f.Fasmer 1964-87, III: 61), which agrees perfectly with the Lithuanian form.The German name seems to be related in some way, too.Balaišis (1987) believes that it originated from the Balto-Slavic form by deformation of sonorants.In any case, the German variant is the earliest attested one: No (Greek) Hypanis 810 no 34 French -eu-continues earlier -ŏ-(probably of a Celtic prototype), whereas the Dutch name reflects the Germanic sound shift ŏ > ă (cf.Krahe 1964: 96f.). 35The river originates at a place called Fontibre.It also gave the Iberian peninsula its name.On this river name ⇒ below in section 9. 36 The river originates at a place called Duruelo de la Sierra.On this river name ⇒ below in section 9. 37 Attested only by Ptolemaios.Although he even provides geographical coordinates, the identification of toponyms in the far North is difficult since his coordinates are distorted.I follow the identification by Reichert (2005:  280).The modern name is first attested in the 9th century as Odagra (cf.Eichler 1981: 51; Georgiev 1966: 190).Both Volm's (1958: 8) identification of the river name with the Indo-European 'water'-term (Engl.water) and the connection with Avestan aδu-'channel' envisaged by Krahe (1964: 41) and Udolph (1990: 209f.)seem doubtful. 38I assume that the form Hypanis is related to the modern (Russian) name despite the differing initial consonants.A Greek form Kōphēn, which matches the modern name more neatly, is attested in the 6th century AD (cf.Schramm 1973: 96). 39In the modern name, the Arabic noun wādī 'valley' was added to the original name. 43Including Džurak-Sal. 44Pronounced /sεn/; Old French Saine.We can probably postulate an early Romance form *sékna or *sákna, the development -kC-> -iC-being regular in French.The presence of -qu-in the ancient spelling has always been puzzling because an earlier /kw/ should have developed into *p in the regional Celtic language (Gaulish).There are two ways out, either the assumption that -qu-rendered something else than /kw/ in this name, or that the region did not yet belong to the Gaulish territory when that sound change occurred.Jung (1970) opts for the first alternative. 45On this river name ⇒ below in section 9. On the phonetic development ⇒ Donau. 46First attested in the 12th century as Pripetь (cf.Nepokupnyj et al. 1979: 446f.). 47All older sources already have the spelling Bug (cf.Fasmer 1964-87, I: 227).Cf. ⇒ Južnyj Bug. 48On this river name ⇒ below in section 9. 49 Interpretable as 'she-bear' in Russian. 50Today the upper river is called Werra and the lower river Weser.Both are split developments from a single older German name Wisera(ha).For similar names of other rivers cf.Krahe (1964: 50f.) and ⇒ below in section 9. 51 Genitive Psla.Various Old Russian spellings from the 12th century on (cf.Nepokupnyj et al. 1979: 451f.)  which Fasmer (1964-87, III: 397) normalizes as *Pьsьlъ. 52See Duridanov (1999) and Schramm (1981: 351-353)   54 First attested in the 12th century as Goryn' (cf.Nepokupnyj et al. 1979: 149). 55From Arabic (al)-wādī al-kabīr 'the big valley'. 56The mouth of this river is called Gironde, which is a dialectal variant of the same name. 57In older sources Pado, from which the modern name derives with a dialectal loss of -d-.I assume that the oscillating ancient forms reflect variants of essentially a single name. 58Older spelling Sъžь (cf.Fasmer 1964-1987, III: 706f.). 59On the phonetic development ⇒ Donau.The medieval spelling is still Alt, which is why Schramm (1981:  194f.)ascribes the change to Oin this name to a relatively late influence of Hungarian. 60Standard opinion (cf.Fasmer 1964-87, I: 154 and others) connects the name with Russ.ber'oza 'birch-tree'.Otkupščikov (1974) explains that there are both morphological and botanical obstacles to this etymology, and prefers a derivation from Slavonic *bǔrzǔ (Serbian brz etc.) 'quick', which, however, is not without complications phonetically. 61Can be translated as 'field-river'. 62This name looks like a Latin diminutive (type ampulla) of the name of ⇒ Meuse / Mosa.The Moselle is not a tributary of the Meuse, but both rivers are roughly parallel in their upper parts and approach to a minimal distance of about 10 km.The Moselle in its turn has a tributary called Moselotte, which is thus a double diminutive. 63The name seems to be, apart from the Slavonic derivation suffix, identical with ⇒ Ingul.The sources of both rivers are at a distance of about 25 km. 64The name is interpretable as 'valley-river'. 65Including Västerdalälven. 66Old Russian form Lovotь (cf.Fasmer 1964-87, II: 508). 67In medieval sources Moyn ~ Meun (German) / Mogus (Latin) (cf.Sperber 1970: 105-109).Modern ai for *oi is a dialectal feature.I believe that a derivative of the river name is found in the name of the city of Mainz / (Latin) Mogontiacum, which is located close to the mouth of the Main.Mogontiacum has traditionally been explained from the name of a Celtic god Mogons, who, however, is not otherwise known to have had any particular relationship to this place. 68The ancient name must evidently have been stressed Mátrona.Nîmes < Némausus vs. Nemours < Nemáusus, Cosne < Cóndate vs. Condé < Condáte.The French and Breton names even of one and the same place may disagree as to their original stress positions, e. g.French Nantes < Námnetes = Breton Naoned < Namnétes, French Rennes < Rédones = Breton Roazhon < Redónes.A convincing explanation of this phenomenon is still lacking (see de Bernardo Stempel 1994: 16-18 for discussion). 69Since there is no known way of reconciling Latin aewith German i-, it has been assumed that ae is here an aberrant spelling for *e.Some authors (e. g.Bichlmeier 2009: 32f.) derive this name from a root *pen-by assuming a transmission through Celtic, where *p was lost. 70On this river name ⇒ below in section 9.The ancient name has been preserved not only in Greek but also in the name of a Bulgarian village Poibrene situated on this river, as well as of a Bulgarian tributary Ibər (cf.Georgiev 1960: 26).The name Marica is first attested in a 12th century Arabic source (cf.Duridanov 1996:  222). 71I do not count the modern Greek name Évros here because, according to Schramm (1981: 290f.), a Greek name Marítza had already become predominant, when the ancient form Évros was artificially restituted in Greece as the name of this river. 72First attested in the 16th century as Tornö älff (cf.Wahlberg 2003: 322). 73The Lithuanian capital Vilnius, situated on this river, took its name from it. 74Cf. ⇒ Tajo (on the phonetic aspects), ⇒ Saône (on the suffix). 75On this river name ⇒ below in section 9.The modern name taken alone would seem comparable to the Bulgarian adjective mórav 'purple'.On the phonetic development ⇒ Donau.Some details of the phonetic evolution remain obscure. 76Velika Morava + Zapadna Morava. 77The Dordogne is formed by a confluence of two torrents named Dore and Dogne.As the ancient form Duranus shows, the name cannot, however, be a compound from Dore + Dogne.Instead, this coincidence is either a result of secondary mutual assimilations of the three names or, as Rohlfs (1960: 11)   Dauzat et al. 1978: 61).Lof the modern name must probably be a glued article, but the river is now called le Lot. 79First attested in the 12th century as Vъskolъ (cf.Nepokupnyj et al. 1979: 404f.).On this river name ⇒ below in section 9. 80 Pronounced /son/; Old French Seonne.The latter part of this name could be (with Dauzat et al. 1978: 7; Jung  1970: 444) the Gaulish word onno 'river' recorded in a glossary from Late Antiquity (De nominibus gallicis, also called "Endlicher's glossary": onno = flumen), so that Saucwould remain as the original name core.This Gaulish noun is probably identical with the element -on(no) described as a recurring suffix in river names by Rohlfs  (1960: 27), found also in ⇒ Júcar and possibly in ⇒ Dordogne. 81Cf. the town of Umeå on the mouth of this river. 82Cf. the town of Luleå on the mouth of this river.The town is first attested in the 14th century as Lulu (Wahlberg 2003: 199). 83Interpretable as 'the clear river'. 84First attested in the 12th century as Vъrъskla (cf.Nepokupnyj et al. 1979: 123).Dobrodomov (1987) thinks that the name contains an Iranian word for 'white' (represented in Ossetic urs ~ wors) as its first component.On this river name ⇒ also below in section 9.
85 Pronounced /du/; the spelling with -b-is historicizing.The name has been connected (e. g. by Dauzat et al.  1978: 40) with the common Celtic term for 'black' (Irish dubh, Breton and Welsh du). 86Cf. the village of Kalix on the mouth of this river. 87Older spelling Mъsta ~ Musta (cf.Pospelov 1998: 276f.).Pospelov explains the name from Finnish musta 'black'. 88First attested in the 9th century as (German) Muora (cf.Schramm 1981: 299; Udolph 1983b: 586).On this river name ⇒ below in section 9.There is a tributary Mürz which bears the same name together with a Slavonic diminutive suffix. 89On a possible etymology of this name see Udolph (1983b: 591f.). 90Homonymous with the Russian term for 'crow'. 91First attested in the 14th century as Lusn (cf.Wahlberg 2003: 195).Might seem to be related to Swed.ljus 'light'. 92 Udolph (1990: 264-270) reconstructs the original form as *Sęnъ. 93First attested in the 14th century as Indal (cf.Wahlberg 2003: 150).An original and obscure name core *in was first expanded by dal 'valley' and then again by älv 'river'. 94Literally 'the big one'. 95Medieval spellings Wlitaua ~ Wultha (cf.Schwarz 1961: 48).They suggest that the initial Vof the Czech name is original and the similarity of the German name to other hydronyms (such as ⇒ Mulde or a river Moldova in Romania) is secondary.The latter river may have given its name to the state of Moldavia, unless Mann (1977: xii and 84) is right who proposes an Albanian etymology for the country name. 96Pronounced /alje/.A 9th century attestation Elarius (cf.Dauzat et al. 1978: 18) provides the link between the ancient and the modern forms. 97First attested in the 7th century as (Greek) Ēlibakia (cf.Schramm 1981: 257).Slavonic speakers seem to have reshaped the ending as *-ica, which was transcribed in Romanian as -iţa (as also in ⇒ Bistriţa). 98In older sources Atexis ~ Adizza ~ Adice ~ Etisa etc. (cf.Battisti 1962), whose spellings of the second consonant seem to reflect a spoken /ʃ/ as still used in local Ladin.Modern Italian -g-/ʒ/ is of dialectal origin.There is a town Este (ancient Ateste) not far from this river, whose name is related. 99Cf. the town of Piteå on the mouth of this river.The town is first attested in the 14th century as Pitu (cf.Wahlberg 2003: 249). 100For similar names of other rivers cf.Krahe (1964: 49f.).
No  102 This name is usually (e. g.Schramm 1981: 372) considered to be cognate with nouns such as Engl.stream and Latvian straume 'stream'.On this river name ⇒ also below in section 9.
103 Tiberis was the common name of this river in the Latin of 2000 years ago.The Etruscan name is not directly attested, but allusions in ancient texts suggest that it must have been similar.By contrast, Albula has been reported as the early Latin name of the river (cf.Le Gall 1953: 50-53).I do not account for this latter name in my computation because its time distance from now significantly exceeds 2000 years. 104Pronounced /jɵŋŋan/.First attested in the 15th century as Ognar(-oos) (cf.Wahlberg 2003: 194).The initial Ljof the modern form is a baroque spelling of the preposition i 'in' which was fused with the original name. 105It has been suggested that the name is a derivation from the Slavonic root *som-'catfish' (cf.Georgiev 1961:  93; Duridanov 1996: 222).On the phonetic development ⇒ Donau.On this river name ⇒ also below in section 9.
106 Including Someşul Mare. 107First attested in the 11th century as Sprewa (cf.Borchers 2006: 65f.). 108I follow the identification by Schramm (1981: 410f.).A derivative of the ancient name still survives in Kysuca, a tributary of the Váh. 109For details and considerations on the original name form see Udolph (1990: 185-193) and Krahe (1964: 28  and 47f.). 110The underlying root seems to be *sinn-(cf.Pokorny 1940: 127f.).Pokorny believes that the -ē-of the Greek transcription is a spelling for *i. 111There is a small tributary Tə ža near the source of the Tundža.Both names are certainly related.Since Bulgarian ə is the regular development of a Proto-Slavonic nasal vowel *õ, Təža seems to be the authentic Bulgarian reflection of the ancient name, whereas Tundža must have been transmitted through a non-Slavonic language, probably Turkish (cf.(Greek) Drilōn 330 136 yes 123 If we take the ancient spelling seriously, we have to assume a sound change /de/ > /dʒe/ (= ge in the modern spelling) which is not regular in Romanian.Ionescu (1971) does accept the identification but nevertheless looks for Indo-European etymologies of the name based on a root *arg-.On this river name ⇒ also below in section 9. 124 On this river name ⇒ below in section 9. 125 For Fasmer (1964-87, II: 131), the name of this and several other Russian rivers is "doubtlessly" of Turkic origin, though he does not decide in favour of a specific etymology.Pospelov (1998: 247) compares the Estonian noun laugas 'pool'. 127Earliest German form Maraha (9th cent.; cf.Wiesinger 1994: 60), which shows that the German name is composed from Mar-(as attested in Latin) + -aha, a frequent element of Germanic river names (still a living word in Icelandic á 'river', also in Gothic as ahwa 'river').On this river name ⇒ below in section 9. On the phonetic development ⇒ also Donau. 128First attested in a 12th century source as Erelь, the same source providing a "Russian" translation as Ugolъ, lit.'corner'.This may be explained as a folk-etymological association based on a Turkic language (cf.Fasmer  1964-87, III: 151). 129First attested in the 10th century as Rъsь (cf.Nepokupnyj et al. 1979: 475). 130Pronounced /'sevə:n/.The change s-> his regular in Welsh.
131 Pronounced /temz/; Old English Temes (cf.Ekwall 1928: 402-405, Förster 1941: 461-604).The modern spelling is artificial and seems to have been influenced by the Latin name. 132The loss of -r-, which became established during the 1st millennium (cf.Deroy/Mulon 1992: 351), cannot be explained from Romance and is perhaps to be ascribed to a peculiarity of the hypothetical local Celtic dialect.The Latin name must here be read Ísara, whereas the Latin name of ⇒ Isère, which had the same spelling, must have been Isára.Cf. ⇒ Marne on this issue.On this river name ⇒ also below in section 9. 133 Cf. the town of Ventspils (Winda in earlier sources) on the mouth of this river.'usa (cf.Pospelov 1998: 333). 181Originates at a place called Prado Tormejón. 182Originates in the town of Teruel, whose name is related. 183For the identification see Schramm (1981: 214f.) and Talbert (2000, vol.I: 289).Medieval Latin sources write Bosina (Schramm).It seems that the ancient -th-transliterated a fricative /θ/ of the local language (Illyrian?), which Slavonic speakers later substituted by /s/.On the phonetic development ⇒ also Donau. 184On the name cf. also ⇒ Segura. 185Medieval spellings include Biusa, Vaiussa and (Greek) Boousa (cf.Schramm 1981: 406).We can probably reconstruct the original name as approximately *Wajōs-os/ā, from which w-, -j-and -s-were regularly lost in Greek.The Albanian name must have been transmitted by a language in which all three consonants remained intact.

Alphabetic index
The following alphabetic index contains all rivers of the above list in alphabetical sorting.

(Apparently) transparent river names
Numerous river names are or, better, seem to be semantically transparent.Aside from the fact that they often contain a component simply meaning 'river' (see the next section), I wish to point out three semantic fields that appear to be particularly recurrent in river names:  The names of a lot of rivers in Europe and world-wide are interpretable as colour terms, most frequently in the senses of 'white' ~ 'clear', 'black' ~ 'dark', or 'red'.Among all rivers whose names are (or seem to be) semantically transparent, this type is the most common one.Kirsna (Lithuania, associable to a Balto-Slavic root for 'black'), Lauterbach ~ Lutter (Germany, lauter is an obsolete term for 'clear'), Mavronéri (Greece), Nera (Italy), Río Tinto (Spain), Röthenbach (Germany), Tamnava (Serbia) and Zwarte Water (Netherlands), to cite only a few examples.In cases such as Argentina ('silver', Italy) or Zlatica ('golden', Serbia), it could be disputed whether the name refers to the visual appearance of the river or to prehistorical metal findings in these locations.
The colour motive is very common on other continents as well.Rohlfs (1960: 5) states that the European colonists of America often resorted to colour terms when coining names for newly encountered rivers.We so find numerous rivers in Latin America bearing the Spanish names Río Blanco or Río Negro, as well as several Rio Verde in Brazil.Well-known rivers in the USA include the Colorado River, Green River, Orange River197 , Red River and Yellowstone River.Two of the largest Turkish rivers are called Kızılırmak 'red river' and Yeşilırmak 'green river', there is also a Bozçay 'gray torrent', and two Anatolian rivers had the name Xanthos 'yellow' in Ancient Greek.Two major rivers of China are the Huáng Hé 'yellow river' and the Hóng Hé 'red river'.One of the largest rivers of Africa is the Niger198 , from which the states of Niger and Nigeria took their names.See further the lists under the entries "White River", "Black River", "Red River" etc. in the English Wikipedia.
It may be surprising to find that terms for 'blue' seem to be somewhat less popular for the purpose of river naming, but examples with this colour term are attested as well: Blau (Germany, near the town of Blaubeuren), Blue Nile, several Blue Rivers in the USA, Glaúkos (Greece), and possibly the Italian river Piave (ancient Plavis), if this name should be related to Croatian plav 'blue' and/or German blau 'blue'.
 Names translatable as 'big / great / large river' also appear repeatedly, such as in ⇒ Dnestr, ⇒ Guadalquivir, ⇒ Velikaja, or (not in my list) Lielupe (Latvia) and Storå (Denmark).One of the major rivers of Ireland combines both possibilities as it is called Blackwater in English and An Abhainn Mhór 'the great river' in Irish.
 Quite a few river names seem to be interpretable as animal terms, cf. in our list ⇒ More examples of rivers homonymous with animal names are cited by Rohlfs (1960:  14-18) and Dauzat et al. (1978: 1f.).For a reason that escapes me, associations with the wolf seem to be particularly popular: Ancient Greek sources cite several rivers named Lykos 'wolf', as there are some German torrents called Wolfsbach, a Loup in France, a Río Lobos in Spain, a Vilka in Lithuania, and a Vuka (ancient Ulca) 199 in Croatia.Several Wolf Rivers also exist in North America, one of which flows into a Fox River.
 Among the rivers considered in this paper, there are a few others whose meanings seem to be transparent in various ways: ⇒ Berezina, ⇒ Dalälven, ⇒ Kemijoki, ⇒ Luga, ⇒ Meža.
What I suggest here is that the semantic fields cited above are not only common motives for the initial coinage of river names, but also, perhaps more typically, favourite semantic targets for folk-etymological reinterpretations of existing names.I assume particularly for the animal names that the apparent meanings are not normally the original ones but came into being by the inclination of speakers to assimilate uninterpretable names to common words of their language. 200The same may also be true for many of the apparent colour names.It would be a worth-wile endeavour to investigate whether any geophysical justification for the colour names can be found, e. g. in the case of the 'white' / 'black' couples.As far as I see, no justification of that kind is evident.I wish to note in passing that colour and animal terms are common naming motives, or common targets for folk-etymologies, also with other geographical objects such as countries 201 , seas 202 and mountains. 203

Head and core elements of proper names
A common pattern of change to be observed in names, as in all kinds of words, is the recomposition with another element.Recomposition is an antagonist to phonetic erosion which would otherwise continually shorten all words.In toponyms, recomposition typically means that a core element that is no longer understood is expanded by a head element that is semantically transparent at the time of composition.I am adopting here the principle expressed by Vennemann (2003: 39) who states that head and core elements of toponyms usually derive from different chronological layers, the core element being the older one.Examples of toponyms composed in this way are very easy to find.A few illustrations: 199 Cf.Schramm (1981: 409f.), who assumes a folk-etymological association for this name. 200I do not share the view expressed by Schwarz (1961: 98f.) who believes that people of ancient times were still apt to give rivers the names of animals, a faculty that, as he thinks, was lost later.etc.), where the Lithuanian noun upė 'river' was composed with an older core element. Several Latin or Celtic (and later intransparent) names of towns in Germany were later expanded by the element Burg 'castle, borough', such as Augusta > Augs-burg, Beda > Bit-burg, Gontia > Günz-burg, Lopodunum > Laden-burg, Regino > Regens-burg. Numerous British place names are composed of an incomprehensible core element and a head element that is still (more or less) transparent, such as -ton ('town'), -ham ('home') or -borough / -burgh. Several English country names were expanded by the noun land, such as Ireland (< Irish Éire + land) and similarly England, Finland, Poland, etc.

Recurring elements in river names
We can assume that compounding as described in the preceding section has occurred in the European river names at all times.This should result in recurring prefixes or suffixes originating from name heads, the most recently added elements. 204Such recurring affixes can indeed be found.The following are among the most convincing examples:  The most striking recurring element is a prefix *d˘n-which appears particularly in the names of some of the largest rivers of Europe (⇒ Dnepr, ⇒ Dnestr, ⇒ Don, ⇒ Donau, ⇒ Donec, ⇒ Dunajec, ⇒ Zapadnaja Dvina; less probably also as a noninitial element in ⇒ Rhône < Rhodanus).It is generally agreed upon that this prefix represents a term for 'river' of some Indo-European language.It is most commonly considered as specifically Iranian (cf.Ossetic don 'river', Avestan dānuš 'river'), but an explanation from Dacian has also been suggested (Schramm 1973:  58-60), and the existence of an (unattested) Celtic noun *dān (don) 'river' has been suspected from the existence of rivers with this name in Britain (Förster 1941:  141-148).It should also be noted that the constituent order head -dependent as evidenced by the d˘n-names is not typical for any known Iranian language.Be that as it may, it is very probable that names such as Dnepr, Dnestr, Donau etc. are compounds with an Indo-European head element d˘n-, the remaining parts such as (˘)pr-, (˘)str-, (u)vibeing the original name cores.In agreement with what was said above, the core elements need not be explainable from the same language as the head element, as has often been believed, 205 but are, in fact, likely to be more ancient. 204Conversely, shared elements of river names are best explained as being fused name heads.A fundamentally different and, in my view, bizarre proposal was made by Schramm (1998) who believes that the name inventors deliberately constructed a network of name references among the river names of Europe, so that, e.g., the name of the ⇒ Dnepr would have been coined as a compound of ⇒ Donau and ⇒ Prut, because the name inventors conceived it as a Prut-like kind of Danube (p.29; and similarly for various other names). 205It has been a common approach to explain all these names as purely Iranian.

Traditional views on the genetic affiliation of river names
Most scholars have believed that the river names of Europe are predominantly of Indo-European origin.A classic and still very influential work on the topic is Krahe's (1964), whose ideas have been closely followed not only by his German school of hydronymists (e. g.Schmid and Udolph) but also beyond (e.g.Kitson 1996).First, Krahe discovered that a number of roots recur repeatedly among the European river names.Second, he believed that these roots can be equated with items of Indo-European appellative vocabulary.The etymologies proposed by Krahe involve a large number of Indo-European terms signifying water or qualities of water, in fact a larger number of terms than survive in any single Indo-European daughter language.This made him suppose that these names derived from a layer close to the proto-language, in which the sum of all water-related terms found in the later Indo-European daughter languages were still in existence. 207He called this layer "alteuropäisch" ('Old-European').eorgiev (1966)   'Sumpf', 'Marsch' u. dgl.bedeuten" (Georgiev 1966: 191).There has, naturally, been some criticism of the Old-European approach.Vennemann is the most prominent opponent.He (Vennemann 2003) suggested that many hydronyms do not derive from Indo-European, but rather from a prehistoric language family he calls "Vasconic" which predates the Indo-European invasion and whose last survivor is modern Basque.Blong (2003) has searched a compromise between both approaches by proposing that European hydronyms are in part of Indo-European and in part of Vasconic origin.I will not endeavour to discuss the numerous individual arguments brought forward by both parties.Instead, I wish to introduce an entirely new argument into the discussion, namely a glottochronological one.I argue that, from a glottochronological point of view, most of the river names are unlikely to be Indo-European.While I am not making any positive statement concerning a possible relationship to Basque, my conclusions are in any case less compatible with Krahe's than with Vennemann's theory.
The lexemes compared with the European hydronyms are often taken from geographically remote branches of Indo-European: "[...] many, perhaps most, of these watery elements survived as lexical items only in eastern Indo-European languages not in European ones.It follows that the naming-system was in operation since before the eastern languages separated from the western continuum" (Kitson 1996: 86). 208Schmid and others of Krahe's adherents have often stated that "Old-European" hydronyms are scarce in (presumed) fringe areas of the Indo-European world, such as India, Anatolia, or the Mediterranean.To the best of my knowledge, it has never been clarified whether there is really a lack of river names in these areas for which Indo-European etymologies could be suggested, or only a lack of investigation of the river names of these areas. 209There is even an article "Vasconic substratum theory" devoted to the topic in the English Wikipedia.

Computation and conclusion
Out of the 210 rivers under study, 92 are attested in classical Greek or Latin sources.I consider only these rivers in the following, assuming that the names that happen not to be attested from antiquity would show the same statistical properties as the attested ones. 210he ancient name has survived in 78 instances and been replaced in 10 instances.The four instances of rivers with alternate ancient names only one of which has survived (Donau, Kuban', Miño, Saône) are counted half, so we get 78 + 4/2 = 80. 211This means that 80/92 = 87% of the ancient names have been preserved through 2000 years.Under Swadesh's hypothesis, namely the assumption that the preservation rate was essentially the same in the prehistoric periods, we can estimate the distribution of name ages.The result is an exponential distribution which, however, as was explained in section 2, provides no exact points in time but only lower limits for the age estimates.This is why I add "at least" to the statements that follow.We thus find that:  80/92 = 87% of the river names have existed for 2000 years (based on direct observation). (80/92) 2 = 76% of the river names are likely to have existed for at least 4000 years. (80/92) 3 = 66% of the river names are likely to have existed for at least 6000 years. 57% of the river names are likely to have existed for at least 8000 years. 50% of the river names are likely to have existed for at least 10000 years. 43% of the river names are likely to have existed for at least 12000 years. 38% of the river names are likely to have existed for at least 14000 years. 33% of the river names are likely to have existed for at least 16000 years. 28% of the river names are likely to have existed for at least 18000 years. 25% of the river names are likely to have existed for at least 20000 years.
Put differently, 50% of the present names are likely to have existed for at least 2000 * log 0.5 / log (80/92) = 9900 years = since 7900 BC.It should be emphasized that these figures presuppose a regular transmission of the names that was not disrupted by extraordinary events.Assume, for the sake of the argument, that some charismatic personality had rebaptized all European river names in about 2000 BC.This fact would not become evident from my method, and my computations would be misleading.Such 210 There is a methodological problem that should be mentioned here.The probability for an ancient river to have been identified is higher if its name was not replaced because name similarity can be one of the arguments to support an identification.For example, the identification of modern ⇒ Bosna with the ancient Bathinus heavily relies on the name form, and this identification would seem much less certain, had the ancient name been a different one.In that case, I would not have counted this river as "no match" rather than "match", but I would rather have excluded it from my statistics, which is what I do for all river names without (evident) ancient attestation.This effect should lead, in theory, to a systematic overestimation of the name preservation rate.The impact of this problem is hard to quantify.I hope that the effect is on the whole insignificant and that it will be largely cancelled out by the fact that my dates are not meant as exact points in time but only as lower limits for the age estimations. 211Instead of asking how many of the ancient names have survived, it could be asked how many of the modern names are ancient.By that logic, we would count a full match when one of two alternate old names has survived (as in ⇒ Donau) but only a half match when one of two alternate modern names is ancient (as in ⇒ Vardar).The global result should be essentially the same.I preferred the other way of counting because the modern names are known more completely than the ancient names, so that we are able to judge the fate of a given ancient name with more precision than to judge the age of a given modern name.
an event is of course unlikely, at least as unlikely as would be a deliberate change of word forms by an individual which would have interrupted the natural phonetic change of languages.But there is another extraordinary event that was real.This is the last ice age which lasted (subject to definition) until about 8000 BC.During that time, the Northern part of Europe was uninhabited.This implies that, even though my formula mechanically predicts some of the river names to be 20000 years or older, the repopulation of parts of Europe after the ice age is an obvious limit to the real age of river names.
What are the conclusions concerning a possible Indo-European origin of the river names?In the northern part of Europe, the names cannot predate the hard limit given by the repopulation after the ice age, but I assume that a significant number of them derive from exactly that time.By contrast, most traditional datings fix the Indo-European proto-language at only 3000 or 4000 BC.An earlier dating was suggested by Renfrew (2000) and by Dixon (1997: 84), both of whom linked the spread of Indo-European to the introduction of neolithic culture into Europe, to be dated about 7000 BC.I am personally more inclined towards Renfrew's and Dixon's than the majorities view for reasons not to be discussed in this place.But even under that assumption, it appears that the names of most larger European rivers are too old to be of Indo-European origin.This is particularly so as the Indo-European territory could not yet have covered the whole of Europe at the time when the proto-language was spoken.Therefore, most of the river names are likely to derive from Pre-Indo-European languages.These names are our best, and perhaps our only witnesses of the extinct paleolithic languages that must have been spoken in Europe before the massive spread of the Indo-European languages within (or into) this continent took place.

No. Modern name(s) Ancient name(s) Length in km Ancient name sur- vived?
When the vowel ъ was dropped in Slavonic, o of a preceding syllable underwent a compensatory lengthening which lead to ó (nowadays spoken u) in Polish and to i (via ü) in modern Standard Ukrainian.Traditional dialects of Northern Ukraine are closer to Polish in this respect and have uo or u here (cf.Shevelov 1979: 318).The river name was affected by this sound change and ended up with the dialectal reflex u rather than the Standard Ukrainian reflex i. SeeUdolph (1983a)for further discussion and a list of early attestations.Udolph considers the name as related to the noun represented in German Bach = Engl.dial.
40Old French Rosne.The German form is only in local use in the region of Switzerland where the river originates.On this river name ⇒ below in section 9. On the spelling with Rhcf.⇒Rhein.41Firstattested in the 11th century as Vurta (cf.Udolph 1990: 282).42OldRussianBogъ;Byzantine Greek Bogou.beck.The earliest possible attestation of the modern name is Latin Vagosola ~ Bagossola from the 6th/7th centuries (cf.Nepokupnyj et al. 1979: 422f.).If we accept this identification, it shows that medieval o developed from an earlier a (cf.⇒ Donau on this issue).The name came to be homonymous with that of the ⇒ Zapadnyj Bug which is at a distance of ca. 100 km.This motivated the addition of discriminating adjectives on both names.A tributary of the Južnyj Bug is called Bužok, which is a diminutive formation.
pursued a similar approach as I do here by examining the names of 26 European rivers that exceed 500 km in length.He, as did Krahe, believed to find Indo-European explanations for most of them and concluded: "Die Herkunft aller dieser Flussnamen ist also klar: es stecken darin altertümliche i[n]d[o]e[uropäische] Wörter, die 'Wasser', 'Fluss', 'Strom',