Detecting Acoustic Morphemes in Lattices for Spoken Language Understanding