摘要
In many Tibeto-Burman languages we find that there are a number of forms that are clearly related though differ in one segment. In some cases these variations may be due to regular or common alternations, such as in Tibetan, where you have dental suffixes that can nominalize a verb (e.g. rkun-po 'thief', from rku 'steal'). In other cases we cannot find any morphological reason for the variation, even though the variation may involve the same segments, as in Tibetan bka, skad 'speech'. When we reconstruct the Proto-Tibeto-Burman provenience of these cognates, we sometimes have no way of knowing which form is older, so we must reconstruct two forms that are clearly related, that are what James A. Matisoff has dubbed 'allofams'. On the Chinese side of Sino-Tibetan we find similar alternations among cognate forms, as in 亡 *mja!1/4, 無 *mjag 'negative/not have'; 往 *gwja!1/4, 于 *gwjag 'go'.
This paper concentrates on variable finals, and argues that just as we find a certain amount of both rule-governed and non-rule governed variation in modern languages, in reconstructing Proto-Sino-Tibetan we should recognize the possibility of these types of variation. Second, the variation we find in PST and its immediate daughters is not as symmetrical and orderly as has been assumed. Third, the causes of the variation are complex and multifarious. Fourth, reconstructing a complex, typologically unlikely system to 'explain' the variation, such as the voiced stop finals reconstructed for Old Chinese, may also prevent us from attempting to find out the real causes of the variation. Fifth, the concept of word families is an important one, but we should not be unnecessarily constrained in our search for cognate sets by artifacts of our reconstructed system.