Gaulish Stress

This is a re-posting of a comment on Languagehat's blog. In a discussion that somehow moved from a discussion of register differences in Japanese to initial mutations and stress in Celtic (one of the many charms of discussions on that blog), I posted some half-remembered information from Pierre-Yves Lambert, "La Langue Gauloise", Éditions Errance, Paris, 2003 and promised to check and post what I found. So here are (1) the relevant paragraphs in French (from p. 48), (2) an English translation (did it myself, so please point out any mistakes), and (3) a few comments.


(1) L’accent en gaulois
Nous n’avons pas beaucoup d’indices sur l’accent gaulois; quelques formes du latin de Gaule ont un comportement spécial. On a depuis longtemps relevé les deux traitements que présentent les noms des cités gauloises, un accent antépénultième donne Rennes, Bourges, et l’accent pénultième donne Redon, Berry.
Bitúriges > Bourges
Bituríges > Berry
Ainsi Nemausus donne (accent pénultième) Nemours, mais avec accent sur l’antépénultième, Nîmes; Condate donne Condes où Condé, Arelate donee Arles ou Arlet.
Autres exemples de formes accentuées sur l’antépénultième: Caturiges > Chorges, Cambo-ritum (« le gué courbé ») Chambord, Eburovices Evreux, Durocasses Dreux, Bodiocasses Bayeux…    
En fait, les formes avec accent antépénultième ne sont pas celles qui posent problème : on en avait aussi en latin et même en latin tardif (ex. : hóminem > homme). Le problème est de savoir pourquoi certains de ces mots sont devenus accentués sur la pénultième, avec allongement de la voyelle pénultième (Cóndate > Condáte > Condāte > Condé). Il n’est pas sûr que le phénomène remonte vraiment au gaulois : cela peut être dû à des disparités socio-linguistiques dans la société gallo-romaine.

(2) Stress in Gaulish
We don’t have many clues about Gaulish stress; some forms of the Latin of Gaul have a special behavior. One has long noted the two treatments that the names of Gaulish cities present, an antepenultimate stress gives Rennes, Bourges, and the penultimate stress gives Redon, Berry.
Bitúriges > Bourges
Bituríges > Berry
Thus Nemausus gives (penultimate stress) Nemours, but with stress on the antepenultimate, Nîmes; Condate gives Condes or Condé, Arelate gives Arles or Arlet.
Other examples of forms stressed on the antepenultimate: Caturiges > Chorges, Cambo-ritum (“the curved ford”) Chambord, Eburovices Evreux, Durocasses Dreux, Bodiocasses Bayeux…    
Actually, the forms with antepenultimate accent are not those that pose a problem: those were there also in Latin and even in Late Latin (e.g.: hóminem > homme). The problem is to know why some of these words have become accented on the penultimate, with lengthening of the penultimate vowel (Cóndate > Condáte > Condāte > Condé). It is not certain that the phenomenon really goes back to Gaulish: this may be due to socio-linguistic disparities in Gallo-Roman society

(3) Comments:
While it is certainly true that Latin knew stress on the antepenultimate, this was only true for words where the penultimate was short. As length was not normally indicated in Latin writing, we partially have to rely on the stress indicated by the modern forms of the names or on etymology to establish Gaulish vowel length; in this case, relying on the modern stress can become a circular argument. But there ought to be no doubt that names like Nemausus or the names in –casses ought to have penultimate stress in accordance with Latin rules, so the antepenultimate stress indicated by some of the modern French names needs to be explained. I’ve also generally seen the “i” in –riges described as long; again, that would demand penultimate accent according to Latin rules. Of course, as the authors state, the indicated stress may not be the Gaulish stress but due to some differences between Gallo-Roman and Standard Latin; that we’ll probably never know.
I originally posted to respond to the statement that Celtic had word-initial stress - an opinion that has a good pedigree but is supported by surprisingly little evidence - the only Celtic branch that clearly has initial stress is Goidelic. Brythonic had penultimate stress and for Continental  Celtic we only have the clues for Gaulish quoted above, which point to antepenultimate stress. Forms like Némausus and Cóndate could also indicate word-initial stress, but most of the forms with more than three syllables rule that out. The only exception is Arelate, where forms like Arles actually don’t indicate antepenultimate stress, but initial (or ante-antepenultimate) stress, something that Lambert seems not to notice. So perhaps Gaulish didn’t have a fixed stress, or prefixed nouns (prefix are-) behaved differently?