The Etymon: Where Genealogies Coalesce

How do we trace words and morphemes to their source? It’s simple. Take a set of forms that seem to be related (or cognate, in the jargon of historical linguistics), travel back in time and watch out for coalescence effects. The farther backwards you go, the more similar the candidate cognates should grow, until at some points they coalesce completely. If instead they seem to become less and less similar, and their genealogies seem to drift apart rather than converge, you are probably dealing with spurious cognates – words whose similarity is accidental, and not due to common descent.

If you have no TARDIS to actually travel in time, do what scientists do. Take a reliable model of historical tranformations (in this case, of phonetic and morphological changes: linguists have built them for many of the known language families), and run it backwards. It is not always easy to reverse reconstructed language changes, because they are frequently accompanied by an irrevocable loss of information. For example, if a language has been affected by a change consisting in the merger of the vowels *e and *o and *a (let’s imagine that they all become /a/), you may be unable to recover the original quality of the vowel in a given modern word containing /a/. Fortunately, information lost in the genealogy of one particular language may be preserved in related languages. In fact, a merger just like the hypothetical one described above occurred in the prehistory of the Indo-Iranian languages, but data from languages such as Greek, where the original vowel qualities are preserved more faithfully, allow us to disambiguate the Indo-Iranian evidence. Furthermore, even where the three vowels have merged, traces of their old pronunciation may be visible as different phonetic impact on the development of the neighbouring sounds (with information being re-encoded rather than lost). If in a distant past old *e regularly palatalised a preceding *k, which became č [ʧ] as a result, and if this č can only have this kind of origin, you know that a modern sequence like ča must reflect *ke. Modern ka is still ambiguous (*ko or *ka), but at least you can rule out *ke. And guess what? This is also what we really find in Indo-Iranian.

Note that the genealogy of a set of cognates does not have to follow the structure of family relations among the languages that contain those cognates. If, for example, a Germanic word has a cognate in Slavic and another one in Greek, we usually expect the genealogies of the cognates to meet in the common ancestor of Germanic, Slavic, and Greek. But words are often exchanged “horizontally” between languages. They have their own unique histories; they may even survive the death of their original host language and live on somewhere else. It may turn out that their genealogies, traced backwards in time, leave the Indo-European family before they finally coalesce.

It is also possible for a set of related forms to converge only partly by the time we reach the common ancestor of the languages that contain them. For example, OE fȳr, Gothic fon (gen.sg. funins), and Old Norse funi beside fýr and fúrr ‘fire’ are all related and represent the scattered and partly reworked remains of a single ancient paradigm. That rather complicated paradigm was quite likely felt to be somewhat irregular already in Proto-Indo-European. We reconstruct it like this: nom.sg. *páh₂wr̥, gen.sg. *p(h₂)wén-s (the bracketed segment was probably left unpronounced). In Germanic, the *r/n alternation in the stem of the word was no longer conditioned grammatically. We have different Proto-Germanic byforms  of the ‘fire’ word, with either *r or *n generalised accross the board in the historically known languages. To “unify” them, we have to go beyond Proto-Germanic, but even as we reach Proto-Indo-European, the coalescence is not complete. The stem has different variants (“allomorphs”) in different case-forms. We can treat them as “internal” cognates traceable back to a still more remote common ancestor. Such an approach is called internal reconstruction, in this case within PIE. The stem looks as if it had originally consisted of two morphemes, with the word stress alternating between them, roughly along the following lines:

• “strong” cases: *péh₂- + *-weN-
• “weak” cases: *peh₂- + *-wéN-

The symbol *N stands for a hypothetical Pre-PIE segment that became PIE *r in the nom./acc. (where it was word-final) and *n in other case-forms. The transcription used above is provisional and has no claim to strict accuracy: internal reconstruction is inherently insecure, and the comparative method can’t be applied to PIE itself (unless one day we discover hitherto unrecognised cognates lurking in other language families). Can we identify the two morphemes? The *-wr̥/-wén- part is at least a recurrent derivational suffix, even if its function isn’t entirely clear. We find it in several other neuter nouns, usually related to a verb root. In Hittite, where it is particularly frequent, it productively forms verbal substantives. Perhaps, then, the *pah₂- part is the same as the verb root *{peh₂-} ‘protect, guard’ (note the regular change of underlying *e to *a triggered by an adjacent *h₂), and this ‘fire’ word originally mean ‘stuff to be protected’ (like the sacred fire of Vesta)? Possible but speculative. Nevertheless, at least from the formal point of view, we have approached the point of where the genealogies of fire and fon coalesce not only with each other but also with those of Greek pũr, Classical Armenian howr, Hittite pahhur (gen.sg. pahhuenas), etc. – and I mean not only the quotation form of each word, but also all its case-forms. In order to reach that point we have to travel back in time beyond Proto-Germanic and even beyond Proto-Indo-European, into the mostly uncharted territory of internally reconstructed Pre-Proto-IE.

There will be more about different types of coalescence later.