Don Ringe's guest post "The Linguistic Diversity of Aboriginal Europe" (1/6/2009) led to a series of additional posts in response to readers' questions: "Horse and wheel in the early history of Indo-European", "More on IE wheels and horses", "Inheritance versus lexical borrowing: a case with decisive sound-change evidence", "The linguistic history of horses, gods, and wheeled vehicles". This morning brings another in the series.
As in Don's earlier posts, there's quite a bit of technical detail; but it seems that many readers are happy to follow Don through the historical undergrowth. Even if you find some of the terminology and typography puzzling, these posts may give you a picture of how historical linguists think and reason.
And as before, because the typographical details are very important in this sort of discussion, and I'm not certain that I've rendered them correctly in html, a pdf of the post is here.
[Guest post by Don Ringe.]
This is the last of the posts I promised, showing what loanwords, including Wanderwörter, look like from the standpoint of mainstream historical linguistic methodology. After this I’m going to have to stop posting for a while; term is starting, with a full load of teaching and advising, and unfortunately I’m also chair of my department at the moment.
The first three words discussed here are possible or probable loanwords into subgroups of Indo-European; whether you want to call them Wanderwörter depends on how much wandering you want to see demonstrated before you apply the term. The last word fits the definition of a Wanderwort unproblematically.
One notable fact about the reconstructable Proto-Indo-European lexicon is that it lacks a word for ‘ax’, though the PIE speech community (whoever they were, exactly) must have had axes. No doubt that’s a historical accident: whatever the PIE word was, it’s been edged out by more recent words in every daughter (or every daughter but one). But there are two or three words that seem to be attested in more than one first-order subgroup of the family, and all exhibit peculiarities which make it fairly likely that they are loanwords.
The most straightforward case is Sanskrit paraśús and Greek πέλεκυς /pélekus/. The forms match perfectly, except for the position of the accent (which could have been shifted in one language or the other—there are other similar cases, though the process is not well understood), and it’s not hard to reconstruct a preform *peleḱus. Strictly speaking it can’t be labelled “PIE”, because reasonable cladistic trees suggest that the last common ancestor of Greek and Indo-Iranian was probably not also the ancestor of Anatolian, Tocharian, Italic, or Celtic, but there are plenty of other words about which the same has to be said simply because of the distribution of their surviving reflexes. There are also Iranian cognates (though the usual words for ‘ax’ in Avestan and Old Persian happen to be different). The Proto-Iranian form must have been *paratsuš, becoming *parasuš in some dialects and *paraθuš in others. It eventually became færæt in Ossetic; a similar form was borrowed into pre-Proto-Tocharian as *peretə, which developed into TB peret and TA porat. The borrowing must have occurred after the Tocharian palatalization of consonants by front vowels had run its course, because PT *e could have become TA o only if the preceding labial was not palatalized; that means that the borrowing must have been fairly late in the prehistory of Tocharian, which is not surprising.
So what is wrong with this picture? The structure of the reconstructable word is anomalous, from a PIE point of view. The root can’t be *pel-, because *-eḱu- is not a known suffix, and it can’t be *pele-, because PIE roots can’t end in vowels; it has to be *peleḱ-, with *-u- as the (completely unremarkable) suffix. But PIE roots don’t surface with two nonhigh vowels in them; one vowel or the other is normally dropped. (PIE words can have more than one nonhigh vowel in them; cf. e.g. Hittite nēpisas ‘of the sky’ and Homeric Gk. νέφεος /népheos/ ‘of a cloud’ < *nébhesos, or Latin deus and Skt. devás ‘god’ < *deywós. But those reconstructable words are securely analyzable as *nébh-es-os and *deyw-ó-s — in each case, root + suffix + ending—with only one nonhigh vowel in the root.) So this word is probably a loanword into the last common ancestor of Greek and Indo-Iranian (or one of that language’s ancestors, conceivably even PIE). Whether it has anything to do with Akkadian pilakku or Sumerian balak seems very doubtful, since those words actually seem to mean ‘spindle’; but in any case this ‘ax’ word is pretty clearly a loanword.
The second item is more problematic. There is a well-attested Germanic family of feminine words meaning ‘ax’: Gothic aqizi, Old Norse øx, Old English æcs, æces (the former is the West Saxon form, the latter the Mercian form), Old High German acchus. The North and West Germanic forms reconstruct to *akusi or *akusiz, except that (1) we don’t understand why the second vowel of the West Saxon OE form was syncopated, and (2) OHG cch (variously spelled, but clearly a geminate) must reflect *kk < *kkw < *kw. The first problem is apparently just a minor glitsch in one OE dialect, but the second has further consequences (see below). The Gothic form clearly reflects *akwizī. The first question is how to reconcile those two preforms.
It seems clear that PGmc. *ī was shortened to *i word-finally in North and West Germanic, so that works out; and the consonant cluster *kw that underlies the OHG geminate was the normal West Germanic outcome of PGmc. labiovelar *kw, so that works out. But the difference between *s and *z is clearly an example of the widespread “Verner’s Law alternation”; apparently the alternation occurred within the paradigm of this word. What caused the VL alternation was an alternation in the position of the accent in pre-PGmc.; that alternation, in turn, was usually associated with alternations between different vowels (“ablaut”), which must have given rise to the different vocalism of the second syllables of the two forms. Finally, the Gothic form shows clearly that this was a PGmc. feminine stem with nom. sg. *-ī, oblique *-jō-; and that class of nouns is known to reflect a PIE class that exhibited a “proterokinetic” accent and ablaut alternation.
Putting all that together, we can reconstruct the following paradigm:
PGmc. *akwisī ~ *akuzjō- < pre-PGmc. *agwésī ~ *agusyā́- < post-PIE *agwés-ih2 ~ *agus-yéh2– (bearing in mind that the first consonant could just as well be *ǵ as *g, and that the vowel preceding it could just as well have been *h2e or *(H)o, i.e. *o preceded by any laryngeal or by none).
This requires us to posit extensive levelling in the daughter languages, but it’s the only reconstruction I can come up with that accounts for all the facts. (I’m aware that I’ve posited a change from an inherited sequence of velar + *w into a labiovelar in PGmc., then back into a sequence of velar + *w in the West Germanic languages. That sort of thing is fairly commonplace and not at all problematic, because linguistic changes are very localized in time and space—they take no account of what things were like a thousand years before, or what is going on in the next valley over. The consonant cluster is securely established for West Germanic by the OHG geminate, and for some pre-Germanic stage by the ablaut; for arguments that PGmc. had labiovelars instead of velar-plus-*w sequences see Ringe 2008:90-3.)
The reconstruction that I’ve labelled “post-PIE” above looks like a PIE word; but once again, an attempt to analyze it yields puzzles rather than answers. In the first place, is *-wés- ~ *-us- a suffix? It looks exactly like the suffix of perfect participles; but if that’s what it is, why does the root to its left show no trace of reduplication, which is a normal marker of perfect stems? And which root is it? The only obvious candidate is *h2eǵ- ‘drive’, but it’s not easy to see what axes have to do with driving. Moreover, PIE perfect stems were statives, not at all like a Modern English (or Latin) perfect. If there was a perfect stem formed to ‘drive’, it must have meant something like ‘be a driver’ or ‘be drivable’. So far as I can see, none of this makes any sense at all as applied to axes. Finally, the initial nonhigh vowel is at least slightly problematic, given that the accent must have fallen further to the right in every form of the word. It looks like we might have another loanword here.
There is a seductively similar word in Greek, ἀξῖνη /aksí:nę:/; its first and second syllables seem to match the *ag- and *-sī of the nom. sg. of the pre-PGmc. form reconstructed above. But where did the vocalics in the middle go? If we really have to, we can cook up a story that will explain their disappearance, as follows. An inherited alternation *agwes- ~ *agus- would probably have yielded *agwes- ~ *agus- in Greek in the first place, since there is some slight evidence that velar + *w sequences developed into labiovelars in Greek just as they did in Germanic. (‘Horse’ shows such a development, but that word is so messed up that it can’t really be used as evidence. But ‘fingernail’ seems to require the same development—without gemination—so we can probably posit such a development here too.) The result would be two alternations which must have been anomalous at that stage of the language, namely *gw ~ *g and *e ~ *u. We need to suppose that *g was levelled through the paradigm, yielding an alternant *ages-, and that a new zero-grade *ags- was formed to the latter, replacing the (now anomalous) *agus-. (There can’t have been a zero-grade *agws-, because that would have yielded “aps-” in Greek.) But we’re still not out of the woods, because the *-ī of the pre-PGmc. form has to reflect *-ih2, a known feminine suffix, and all the certain examples of that suffix show up in Proto-Greek as *-ia or *-ya. So either we need to argue that the suffix actually contained one of the other laryngeals, and it was replaced by the productive feminine suffix in pre-PGmc., or else we need to somehow get rid of the *h2 and lengthen the i-vowel in Greek. Either alternative involves further speculation.
I don’t know whether to believe the scenario sketched in the preceding paragraph, and I’m certainly not going to urge you to believe it. It’s much more speculative than the Germanic development posited above, because we have only one input and one output to work with and we can’t connect them by regular sound changes alone; in the Germanic case we had four outputs, and it was by comparing them that we were able to make progress—the usual situation in historical linguistics. It’s also less successful: we don’t actually get the Greek form to come out the other end without special pleading. But of course we have a better alternative: we can suggest that similar loanwords — ultimately connected, but not necessarily identical — were borrowed into pre-Germanic and pre-Greek. If that’s what happened, this could be called a Wanderwort.
The third example is a very long shot. One of the Hittite words for ‘ax’ is ates. If it was inherited from Proto-Anatolian (and that’s a big if), it must have been accented on the second syllable, since an unstressed *e in a closed syllable would have become i (Melchert 1994:139-40); the single intervocalic t must reflect Proto-Anatolian *d, which would reflect still earlier *d or *dh. So if the word is old, we’re looking at a preform *adés or *adhés. (Actually the initial vowel is indeterminate; it could also have been *o, and either *a or *o could have been preceded by the first laryngeal, which is completely unrecoverable in this environment. The second or third laryngeal, though, would have survived as ḫ in Hittite.) Astonishingly, a similar word shows up in OE, namely adesa ‘adze’ (once adosa; Mercian eadesa). It has no Germanic cognates, in fact no possible cognates at all except for the Hittite word. If it was inherited from pre-PGmc., the accent must have originally fallen on the second syllable (otherwise the *s would have become PGmc. *z, yielding OE “r”); that matches one of our inferences about the Hittite word. Given that that’s where the accent must have been, the OE d < PGmc. *d can reflect Pre-PGmc. *dh or *t; again that matches one of our inferences about the Hittite word. The n-stem suffix that follows the *s can be a Germanic innovation. Unfortunately the way the first vowel of the OE forms is spelled shows that the second vowel must have been a back vowel in prehistoric OE, and at that point in the language’s development the only possible candidates are long *ō and short *u, reflecting the same PGmc. vowels. The initial vowel is etymologically indeterminate, like that of the Hittite word, though there are a few more possibilities in this case.
So we seem to be stuck with a pre-PAnat. *adhés- and a pre-PGmc. *adhṓs- or *adhús-; in both forms the initial vowel could actually have been *o, and in any case it could have been preceded by the first laryngeal. At first I thought this was the end of the line, but eventually it occurred to me that you could posit a PIE etymology if you’re willing to reconstruct an “amphikinetic” paradigm: nom. sg. *(h1)ódhōs, direct case stem *(h1)ódhos-, oblique case stem *(h1)dhs-´, locative *(h1)dhés(-i). You can suggest that in both subgroups the accent was levelled onto the suffixal syllable in all forms of the paradigm—there are good parallels for that; in Germanic you remodel the whole paradigm on the nom. sg. (for which at least ‘foot’ is a good parallel), while in Anatolian you level the initial vowel into the locative and remodel the whole paradigm on that form.
But what are the odds that this is correct? After all, we have only two attested forms, very widely separated in space and time. Their initial vowels are etymologically ambiguous (though in much the same way, as it happens), and so are the consonants that follow (though there is enough overlap to permit a connection by positing regular sound changes). The second-syllable vowels don’t match at all; we have to resort to one of four possible PIE noun ablaut paradigms to explain that away. The only things that match perfectly are *s, the syllable structure, and the apparent position of the accent. It’s not clear that positing a cognation is a better hypothesis than positing independent borrowing from some third source—in which case we could have another Wanderwort meaning ‘ax’.
But in this case there’s a further possibility that might be even more plausible: the words resemble one another by sheer chance. That should always be considered, because such chance resemblances are common among languages whose histories are well known (so that we can say for certain that particular resemblances can only be the result of chance).
That there might be two or even three discoverable Wanderwörter meaning ‘ax’ is not surprising, given that axes seem to have been luxury trade items in the Black Sea area during the Neolithic and the Bronze Age. Somewhat unexpectedly, there is also a northern European Wanderwort meaning ‘silver’.
In many older IE languages the word for ‘silver’ is clearly native, derived from the root *h2erǵ- ‘white’. The most widespread derivative is *h2r̥ǵn̥tóm; its direct descendants include Avestan ərəzatəm, Old Persian ardatam, Old Irish airget, and Welsh arian. Latin argentum belongs here too, but its first syllable has probably been remodelled, since it doesn’t exhibit the same outcome as the isolated word ursus ‘bear’ < *h2ŕ̥ḱtsos; but that’s not problematic, since full-grade *h2erǵ-, which survives in arguere ‘to make clear’ and argilla ‘white clay’, should have been an available source for analogical remodelling for a long time. The first syllable of Sanskrit rajatám has also been remodelled in some way that is not well understood. Albanian ergjënd is a Latin loanword. Armenian arcath was borrowed from Iranian *r̥dzatam at some point after inherited intervocalic *t had undergone change but before the more systematic Armenian consnant shift (Jochem Schindler, p.c. ca. 1991); if the word were native it would end in “-and”. Greek ἄργυρος /árguros/ is a different derivative of the same root.
But there is also a different word for ‘silver’ attested in Germanic and Balto-Slavic—or rather there are several very similar-looking words; a single protoform is not reconstructable. Here are the data:
Gothic silubr, ON silfr, OE siolfor (gen. sg. still siolufres in early West Saxon), OS siluƀar, OHG silabar, silbar < PGmc. *silubrą;
Lithuanian sidãbras, Latvian sudrabs (with metathesis and unexpected u), Old Prussian siraplis (in the Elbing vocabulary), sirablan (accusative);
Old Church Slavonic sĭrebro.
The languages actually all agree about the gender of this noun: it’s neuter in OCS and throughout Germanic; in Baltic it’s masculine, but in that subgroup the neuter gender merged with the masculine. But if we try to reconstruct a common preform, the best we can do is *siCVbhrom, where *C is some voiced coronal and *V is some short vowel. I reconstruct *r before the ending because all the languages except Old Prussian agree on that detail; it’s reasonable to suppose that OPr. sirabl- reflects metathesis of *silabr-, with consonants matching those of Germanic. (The -p- of the Elbing form can’t be taken too seriously; all OPr. documents were written by native speakers of German, and a substitution of p for b is not unexpected on the part of a mediaeval German speaker.) We could even claim that OCS r…r reflects dissimilation of *l…r, bringing that form too into closer conformity with the Germanic word. But that won’t solve all the problems; East Baltic still has an unexpected d after the first vowel, and the three subfamilies have three etymologically incompatible vowels in the second syllable. At this point a sensible linguist should admit defeat and conclude that we have a Wanderwort.