The key to this problem is recursion and avoiding quadratic complexities, so you need to convert your text into words and then compose those words into a trie (not tree) of indexed characters. You then traverse that trie, recursing as necessary, to produce a list of matching words and their indices into the original text. Note: I made the whole search corpus into a trie, and I shouldn't have for maximum points. I should have only trie-d the first three or four characters and used linear matching for the rest of the words into order to save on memory.