Interpretable NLG for Task-oriented Dialogue Systems with Heterogeneous Rendering Machines

End-to-end neural networks have achieved promising performances in natural language generation (NLG). However, they are treated as black boxes and lack interpretability. To address this problem, we propose a novel framework, heterogeneous rendering machines (HRM), that interprets how generators render an input dialogue act (DA) into utterance. HRM consists of renderer set mode switcher. The contains multiple decoders vary both structure functionality. For every step, the switcher selects appropriate decoder from to generate item (a word or phrase). verify effectiveness our method, conducted extensive experiments on 5 benchmark datasets. In terms automatic metrics (e.g., BLEU), model is competitive with current state-of-the-art method. qualitative analysis shows can interpret process well. Human evaluation also confirms interpretability proposed approach.