Joint Learning of Word and Label Embeddings for Sequence Labelling in Spoken Language Understanding
This work addresses slot filling for spoken dialogue systems, offering a more efficient and flexible method compared to existing approaches.
The authors tackled slot filling in spoken language understanding by proposing an architecture that jointly learns word and label embeddings without requiring label embeddings as input, achieving state-of-the-art performance with significantly fewer trainable parameters on established datasets.
We propose an architecture to jointly learn word and label embeddings for slot filling in spoken language understanding. The proposed approach encodes labels using a combination of word embeddings and straightforward word-label association from the training data. Compared to the state-of-the-art methods, our approach does not require label embeddings as part of the input and therefore lends itself nicely to a wide range of model architectures. In addition, our architecture computes contextual distances between words and labels to avoid adding contextual windows, thus reducing memory footprint. We validate the approach on established spoken dialogue datasets and show that it can achieve state-of-the-art performance with much fewer trainable parameters.