A Joint and Domain-Adaptive Approach to Spoken Language Understanding