manually train Bayesian Filter

[minsk]

Is it possible to manually submit known-spam and known-good messages to Bayesian Filter? Using thresholds is useless - most of the spam that we receive, scores around 5. Plus there is no way to correct a mistake.