This lets you set some advanced features and settings of the Bayesian Analyzer.
Significant Words - By default, the Bayesian Analyzer uses the 15 most significant words (by their probability) to determine whether the e-mail is spam or good.
Prune dictionary after training - determines whether the dictionary will be purged of non-significant words before saving. The non-significant words are words that have not appeared enough times in your spam or good e-mails to be considered in the calculations. This is not selected by default because if you train incrementally, the word counts need to be retained because one additional e-mail might cause the word to be significant. If you Prune the dictionary after training, the dictionary file will be smaller, but the word counts for non-significant words will not be retained for future training sessions.