The Abkh. form presupposes PAT *baʕa (*bǝʕa), where -ʕ- < PWC *-q:I-. In Kab. one would rather expect -q: and not -q (a regular PAK match for PAT *baʕa would be *baq:V). This form must have indeed existed and is reflected in the Abaz. loan from Kabardian - baq̇ 'cattle-shed' (where q̇ is a regular Abaz. substitute for Kab. q: in loanwords). The reason for weakening -q: > -q in modern Kabardian is not quite clear. From the semantic point of view the correlation "cattle-shed" : "fortress" is not surprising, because it is known that towers/fortresses were used for keeping cattle. Armenian bak 'sheep-fold, enclosure' is probably of Caucasian origin, see Acharyan 1, 907 (although it may be borrowed also from Georg. baḳi 'fence, enclosure' - which, for phonetic reasons, can not be compared with WC forms - despite Kuipers 1960, 83, Shagirov 1, 73).