Adversarial Learning on the Latent Space for Diverse Dialog Generation
This work addresses the problem of improving dialog generation for conversational AI systems, though it appears incremental as it builds on existing adversarial and autoencoding techniques.
The paper tackled the challenge of generating relevant and diverse dialog responses by proposing a two-step adversarial learning framework on the latent space, resulting in more fluent, relevant, and diverse responses compared to state-of-the-art methods.
Generating relevant responses in a dialog is challenging, and requires not only proper modeling of context in the conversation but also being able to generate fluent sentences during inference. In this paper, we propose a two-step framework based on generative adversarial nets for generating conditioned responses. Our model first learns a meaningful representation of sentences by autoencoding and then learns to map an input query to the response representation, which is in turn decoded as a response sentence. Both quantitative and qualitative evaluations show that our model generates more fluent, relevant, and diverse responses than existing state-of-the-art methods.