DialoGPS: Dialogue Path Sampling in Continuous Semantic Space for Data Augmentation in Multi-Turn Conversations

TL;DR

DialoGPS introduces a novel many-to-many data augmentation method for multi-turn dialogue generation by sampling coherent dialogue paths in a continuous semantic space, improving model generalization.

Contribution

It is the first to propose a many-to-many augmentation approach for multi-turn dialogues using continuous semantic space sampling with Brownian Bridge.

Findings

Enhanced dialogue diversity in training data.

Improved model generalization and response quality.

Validated by both automatic and human evaluations.

Abstract

In open-domain dialogue generation tasks, contexts and responses in most datasets are one-to-one mapped, violating an important many-to-many characteristic: a context leads to various responses, and a response answers multiple contexts. Without such patterns, models poorly generalize and prefer responding safely. Many attempts have been made in either multi-turn settings from a one-to-many perspective or in a many-to-many perspective but limited to single-turn settings. The major challenge to many-to-many augment multi-turn dialogues is that discretely replacing each turn with semantic similarity breaks fragile context coherence. In this paper, we propose DialoGue Path Sampling (DialoGPS) method in continuous semantic space, the first many-to-many augmentation method for multi-turn dialogues. Specifically, we map a dialogue to our extended Brownian Bridge, a special Gaussian process. We sample latent variables to form coherent dialogue paths in the continuous space. A dialogue path corresponds to a new multi-turn dialogue and is used as augmented training data. We show the effect of DialoGPS with both automatic and human evaluation.