The present disclosure relates generally to determining intent based upon speech input using a dialog system. More particularly, techniques are described using matching-based machine learning techniques to identify an intent corresponding to speech input in a dialog system. These procedures do not require training when intents are added or removed from the set of possible intents.