Semantically Conditioned Dialog Response Generation via Hierarchical Disentangled Self-Attention

TL;DR

This paper introduces a hierarchical disentangled self-attention model that leverages dialog act graph structures to improve multi-domain neural response generation, addressing scalability issues in complex semantic scenarios.

Contribution

It proposes a novel hierarchical graph-based structure and disentangled self-attention mechanism for scalable, semantically controlled dialog response generation across multiple domains.

Findings

Significant improvement over baselines on Multi-Domain-WOZ dataset

Effective modeling of complex dialog act semantics

Enhanced automatic and human evaluation performance

Abstract

Semantically controlled neural response generation on limited-domain has achieved great performance. However, moving towards multi-domain large-scale scenarios are shown to be difficult because the possible combinations of semantic inputs grow exponentially with the number of domains. To alleviate such scalability issue, we exploit the structure of dialog acts to build a multi-layer hierarchical graph, where each act is represented as a root-to-leaf route on the graph. Then, we incorporate such graph structure prior as an inductive bias to build a hierarchical disentangled self-attention network, where we disentangle attention heads to model designated nodes on the dialog act graph. By activating different (disentangled) heads at each layer, combinatorially many dialog act semantics can be modeled to control the neural response generation. On the large-scale Multi-Domain-WOZ dataset, our model can yield a significant improvement over the baselines on various automatic and human evaluation metrics.