For each testing phase some number of words should form a bucket where word's probability increases/decreases based on the user's guesses