Compounds separated by hyphen are initially split into their components, which are then treated separately by ``german_token_to_words''.
