A few days ago I was asked about my opinion on OIr. odbrann (o, m) ‘ankle, Lat. talus’. I did not have one to start with, but maybe it is possible to make some progress with the etymology of this word after all. First a word about the Irish attestations. eDIL (dil.ie/33493) records a number of variants, such as adbronn, fadbrann, but since odbrann occurs in early sources such as Sg. 48a5 and since the similar-shaped odb ‘knot’ < *ozbo- (cf. W oddf) shows a similar tendency to develop into adb and fadb in later times, it is natural to take odbrann as the earliest form. Occasionally odbran is written with a single n, but this is within the normal limits of OIr. spelling variation and may reflect the neutralisation of the contrast between lenited and unlenited nasals in word-final position due to Mac Néill’s Law. The attestations cited in eDIL are not unambiguous as to the sound value of the <b>; ModIr. fadhbairne ‘lumpy object’ (Ó Dónaill; http://www.teanglann.ie/en/fgb/fadhbairne), which must be its modern descendant, supports /b/. Dinneen’s Dictionary is confusing in this context. It has fadhbhairne with lenited bh, but it also has e.g. fadhbh ‘knot’ where Ó Dónaill (and the modern spoken language) have fadhb. /b/ is finally supported by ScG. aobrann ‘ankle’, where ao is the regular development of aδ before b, cf. ScG. faob ‘knot’ vs. Ir. fadhb.
Odbrann is therefore best taken to represent early /oδbrən/. Looked at in isolation, there are various formally possible explanations:
1. it could be odb ‘knot’ + collective suffix -ar + singulative suffix *-ino-; however, I am not aware of any parallel for this sequence of suffixes in Irish, and singulative *-ino- is rare in Irish in the first place, in contrast to the situation in British;
2. it could be a compound odb ‘knot’ + rann ‘part, share’, but the latter is a feminine ā-stem, whereas odbrann inflects as a masculine o-stem;
3. it could be a compound of odb ‘knot’ + brú, gen. bronn ‘belly’ (n, f) (suggestion Roibeard Ó Maolalaigh), but again this would require a shift in the inflectional class and semantically I don’t regard it as very convincing.
In a wider Celtic perspective, the British languages have W uffarn, Bret. ufern, uvern ‘ankle’; OCorn. lifern must be a copying error for *ufern. The a in W uffarn is a regular development of e before rn in this language, cf. tafarn < Lat. taberna, or llantarn < Engl. lantern. These words stand beside W ffêr ‘ankle’, OCorn. fer ‘shank’ (Bret. fer only fossilised in personal names), which, together with OIr. seir, gen. seired ‘heel’, go back to PC *sφeret-. The PIE root is *√spʰerH- (< *(t)sperH-) ‘to kick with the foot’ (see LIV2, 585 and http://www.martinkuemmel.de/liv2add.html#_Toc392258080 for the further discussion of this root). OBrit. *ufern, which can thus be reconstructed for ‘ankle’, has been explained as a compound PC *oφi-sφernā < *opi-spʰer(H)nā ‘thing (up)on the heel = ankle’ (IEW 324). While this makes sense semantically, compounds with first element PIE *opi ‘on, upon’ are extremely rare in Celtic.
I wonder whether we can actually marry the etymologies for OIr. odbrann and OBrit. *ufern. First of all, they have several things in common, more than meets the eye. They have r and n in the final part of the word, the sound before that is a labial, and since Brit. u must continue an o-diphthong, either PC *ou̯ or *oi̯, both really start with an o. The two elements that make sense semantically are the first part of the Ir. word, odb < *ozbo- ‘knot’, and the second part of the Brit. word -fern ‘thing connected to the heel’ < *sφerno-. If the word for ‘ankle’ was originally a PC compound *ozbo-sφerno- ‘knot(ty)-heel’, this may provide a basis from which the words in both branches can be explained. Note that PC *z is only a phonetically conditioned allophon of *s before voiced stops. In Brit., at a relatively late stage, *ozvosφerno- would have resulted. It is conceivable that zv was dissimilated against the phonetically fairly similar sφ and was irregularly turned into *ou̯(o)sφerno-. This leads to OBrit. *ufern regularly. In Ir., *ozbo-sφerno- would have regularly yielded Archaic Ir. *oδb-fern. In this case, it is necessary to operate with the extra assumption that the highly unusual sequence *δbf was simplified to *δb (I am not aware of a second example to test this hypothesis). For the resulting *oδbern, it has to be assumed that from contexts where an extra syllable was added, e.g. in the dative dual or plural and which after syncope and anaptyxis would have appeared as *oδbₔrnəβ’, a hypercorrect oδbrən = odbrann with metathesis was abstracted, after the analogy of, e.g., ingnad ‘strange’ < *in-gnāθ vs. ingantu ‘stranger’ < *in-gₔn†θu < *in-gnāθu (I can’t think of a good example with r at the moment).