Languages display dependencies between pronouns and noun phrases, with co-reference allowed in phrases such as “John ate dinner while he watched television”. Some other constructions are ruled out by the supposedly universal and innate principle-C constraint, making co-reference impossible in “he ate dinner while John watched television.” But there also exist language-specific restrictions on co-reference such as the Russian poka-constraint, which only acts on sentences with fronted adjunct ‘while’-clauses, disallowing co-reference in sentences such as “while he was eating dinner, John watched television” (which is fine in English). This constraint has been shown not to be active in 3 year old Russian kids, but comes online by the time they are 5-6. Yet to learn the constraint directly kids would have to be learning from implicit negative evidence, and not much evidence at that. In this talk, I examine the background on the constraint, consider a theory of Bayesian inference using the size principle that could enable learning from limited implicit negative evidence, and propose a corpus study to investigate the availability of cues to possibly learn such a constraint.