>I hope you don't naïvely assume that because French (not "France" as somebody said) has a Gaulish substrate, its phonological system was also modelled after Celtic. Rather on the contrary, French is a Romance language with a *Germanic* phonology from its Frankish superstrate.
> Some French words derived from Greek also have anlaut-voicing. The only example I can think of offhand is <galoche> 'galosh' from Middle Greek *kalo'poda (acc.) 'wooden shoe'. The best explanation of *grassu- is a Vulgar Latin cross between <crassus> and <grossus>, yielding *grassus which did not spread to the NW extremity of the Empire, hence dialects of NW France reflect *crassu-.
> A more serious objection to Tavi's simple explanation is that N Italy also had a Celtic substratum; Mediolanum/Milan is a Gaulish name, but Milanese has borrowed Fr. <cabaret>, <cabriolet> as <gabare'>, <gabriole'>.Precisely. In languages with an original tense/lax contrast on stops, an unaspirated stop will be assimilated to the lax grade.
> Also, as I have already discussed with Tavi, there is stratification of Latin loanwords in Basque; we have <katu> 'cat' beside <gaztelu>, for example.The voicing of stops at word-initial in loanwords is regular in Paleo-Basque, as shown by Mitxelena. His explanation is that Paleo-Basque had a tense/lax contrast on stops as well as a phonotactic constraint which prevented tense stops from appearing at word-initial.