Parsers and taggers perform great when there is enough data to train them. But often there isn't. We crowdsource Czech school kids to obtain a lot of analyses; using weighted voting to get the required quality and rule based transformations to get the required annotation schema.