Latent Variable Dialogue Models and their Diversity

TL;DR

This paper introduces a dialogue generation model that effectively captures response variability, producing more diverse and consistently acceptable outputs compared to traditional deterministic models.

Contribution

The paper proposes a novel latent variable-based dialogue model that enhances response diversity and consistency over existing deterministic approaches.

Findings

Generated outputs are more diverse than baseline models.

The model produces more consistently acceptable responses.

Experiments demonstrate improved diversity and quality.

Abstract

We present a dialogue generation model that directly captures the variability in possible responses to a given input, which reduces the `boring output' issue of deterministic dialogue models. Experiments show that our model generates more diverse outputs than baseline models, and also generates more consistently acceptable output than sampling from a deterministic encoder-decoder model.