Cross-lingual Spoken Language Understanding with Regularized Representation Alignment