Towards Efficiently Diversifying Dialogue Generation via Embedding Augmentation

TL;DR

This paper introduces a novel embedding augmentation method with soft labels to enhance diversity in dialogue generation models, achieving more varied responses without sacrificing response quality.

Contribution

The paper proposes a new soft embedding augmentation technique combined with soft labels to improve diversity in neural dialogue generation models.

Findings

Generated responses are more diverse than baseline models.

The method maintains similar n-gram accuracy, ensuring response quality.

Experimental results on two datasets validate the effectiveness of the approach.

Abstract

Dialogue generation models face the challenge of producing generic and repetitive responses. Unlike previous augmentation methods that mostly focus on token manipulation and ignore the essential variety within a single sample using hard labels, we propose to promote the generation diversity of the neural dialogue models via soft embedding augmentation along with soft labels in this paper. Particularly, we select some key input tokens and fuse their embeddings together with embeddings from their semantic-neighbor tokens. The new embeddings serve as the input of the model to replace the original one. Besides, soft labels are used in loss calculation, resulting in multi-target supervision for a given input. Our experimental results on two datasets illustrate that our proposed method is capable of generating more diverse responses than raw models while remains a similar n-gram accuracy that ensures the quality of generated responses.