💭 somehow possible to find span of words that are semantically coherent and otherwise meet some requirement but are very rarely found in general text datasets as a way to find new ideas