# Multi-Referenced Training for Dialogue Response Generation

arXiv: https://arxiv.org/abs/2009.07117 · HTML: https://mlsanity.com/paper/2009.07117/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 15 Sep 2020.

## Built on by (most cited first, 3 of 3)

- [Synthesizing Adversarial Negative Responses for Robust Response Ranking and Evaluation](https://mlsanity.com/paper/2106.05894/) (2021)
- [There Is No Standard Answer: Knowledge-Grounded Dialogue Generation with Adversarial Activated Multi-Reference Learning](https://mlsanity.com/paper/2210.12459/) (2022)
- [One Reference or Many? Rethinking Multi-Reference Training for Intermediate-Entropy Text Generation](https://mlsanity.com/paper/2609.22805/) (2026)

## Abstract (arXiv)

In open-domain dialogue response generation, a dialogue context can be continued with diverse responses, and the dialogue models should capture such one-to-many relations. In this work, we first analyze the training objective of dialogue models from the view of Kullback-Leibler divergence (KLD) and show that the gap between the real world probability distribution and the single-referenced data's probability distribution prevents the model from learning the one-to-many relations efficiently. Then we explore approaches to multi-referenced training in two aspects. Data-wise, we generate diverse pseudo references from a powerful pretrained model to build multi-referenced data that provides a better approximation of the real-world distribution. Model-wise, we propose to equip variational models with an expressive prior, named linear Gaussian model (LGM). Experimental results of automated evaluation and human evaluation show that the methods yield significant improvements over baselines. We will release our code and data in https://github.com/ZHAOTING/dialog-processing.