In the Sino-Tibetan family, some languages have complex verbal agreement systems (Rgyalrong, Kiranti), while others (such as Chinese, Lolo-Burmese and Tibetan) seem to show no trace of any relational morphology on the verb. No consensus has yet emerged concerning the antiquity of agreement morphology in Sino-Tibetan: some scholars view it as retention, while others argue it to be the result of independent innovations.

In this article, we propose that the irregular verb za 'to eat' in Tibetan preserves an indirect trace of verbal agreement. The past tense of this verb, zos, presents an -a/-o alternation without equivalent elsewhere in the language, and a similar irregular alternation is found in the cognates of this verb in various Sino-Tibetan languages (including Kiranti and Qiangic). Evidence from Kiranti languages show that this vowel alternation originally reflects the fusion of the stem vowel with a third person patient past tense *-u suff ix. This suggests that Tibetan and other Bodish languages used to have a full-fledge agreement system which disappeared at an early stage, only leaving indirect traces.