We used the UMBC corpus to automatically generate rules. We randomly select sentences, and then randomly select spans. We construct rules by randomly deciding which type to use (tag, word, lemma). For variation, we also insert wildcards.