Joint Learning of Word and Label Embeddings for Sequence Labelling in Spoken Language Understanding