A Hierarchical Latent Variable Encoder-Decoder Model for Generating Dialogues

TL;DR

This paper introduces a hierarchical latent variable encoder-decoder model designed for dialogue generation, effectively capturing complex dependencies and improving the quality and coherence of generated responses.

Contribution

It presents a novel neural architecture with latent stochastic variables spanning multiple time steps, enhancing dialogue response generation over existing models.

Findings

Improved response quality and coherence in dialogue generation

Latent variables help generate longer, contextually consistent outputs

Model outperforms recent neural network architectures in evaluations

Abstract

Sequential data often possesses a hierarchical structure with complex dependencies between subsequences, such as found between the utterances in a dialogue. In an effort to model this kind of generative process, we propose a neural network-based generative architecture, with latent stochastic variables that span a variable number of time steps. We apply the proposed model to the task of dialogue response generation and compare it with recent neural network architectures. We evaluate the model performance through automatic evaluation metrics and by carrying out a human evaluation. The experiments demonstrate that our model improves upon recently proposed models and that the latent variables facilitate the generation of long outputs and maintain the context.