One of the first steps in the utterance interpretation pipeline of many task-oriented conversational ai systems is to identify user intents and the corresponding slots. Neural sequence labeling models have achieved very high accuracy on these tasks when trained on large amounts of trai