A trie can track more than just "does this word exist" — it can count. Given a list of words to insert (duplicates allowed — the same word can be inserted more than once) followed by a list of erasures to remove one occurrence each of, answer one of two questions about a single query: exactly how many currently-present words equal it, or how many currently-present words have it as a prefix? The trick is giving every trie node two running totals instead of one flag: how many inserted words end exactly at that node, and how many pass through it at all (ending there or continuing further). Inserting a word increments the "passes through" total all along its path and the "ends here" total at the final node; erasing does the same in reverse, but only when a copy genuinely exists to remove — so an erasure can never push either total below zero. Once built, a query is just one walk down to the relevant node, reading off whichever total the question asked about.