Learning Discourse-level Diversity for Neural Dialog Models using Conditional Variational Autoencoders

By Tiancheng Zhao, Ran Zhao, Maxine Eskénazi. 2017

786 citations recorded by the source corpus.

While recent neural encoder-decoder models have shown great promise in modeling open-domain conversations, they often generate dull and generic responses. Unlike past work that has focused on diversifying the output of the decoder from word-level to alleviate this problem, we present a novel framework based on conditional variational autoencoders that capture the discourse-level diversity in the encoder. Our model uses latent variables to learn a distribution over potential conversational intents and generates diverse responses using only greedy decoders. We have further developed a novel variant that is integrated with linguistic prior knowledge for better performance. Finally, the training procedure is improved through introducing a bag-of-word loss. Our proposed models have been validated to generate significantly more diverse responses than baseline approaches and exhibit competence of discourse-level decision-making.

Subfields: Artificial Intelligence, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

Citations in the corpus

Breaking the Softmax Bottleneck: A High-Rank RNN Language Model (2017, methodology): “For evaluation, we include both the perplexity and the precision/recall of Smoothed Sentence-level BLEU, as suggested by Zhao et al. (2017).”
Towards a Human-like Open-Domain Chatbot (2020, background): “Recent works also explore new frameworks such as adversarial learning (Li et al., 2017; Zhang et al., 2018c), variational autoencod-ing (Zhao et al., 2017; Gu et al., 2019), or both (Gao et al., 2019b) at the cost of added complexity and less scalability.”
Incremental Learning from Scratch for Task-Oriented Dialogue Systems (2019, methodology): “In the existing work, this approach was used to generate diverse machine responses in both open domain dialogue systems (Zhao et al., 2017; Serban et al., 2016) and task-oriented dialogue systems (Wen et al.”
Approximating Interactive Human Evaluation with Self-Play for Open-Domain Dialog Systems (2019, background): “Hierarchical recurrent neural networks (RNNs) have been proposed as a way to improve long-term tracking of the conversation, through maintaining both a word- and utterance-level RNN [3–5, 21, 22].”
Importance of Search and Evaluation Strategies in Neural Dialogue Modeling (2018, background): “Greedy search Greedy search has been the search algorithm of choice among the recent papers on neural dialogue modeling (Gu et al., 2018; Zhao et al., 2017; Xu et al., 2017; Weston et al., 2018; Zhang et al., 2018).”
Challenges in Building Intelligent Open-domain Dialog Systems (2019, background): “Other frameworks, including conditional variational autoencoder (CVAE) [26, 56, 121, 124, 184, 185] and generative adversarial network (GAN) [64, 161], are also applied to dialog generation.”
USR: An Unsupervised and Reference Free Evaluation Metric for Dialog Generation (2020, background): “, BLEU, F-1, METEOR, ROUGE) have several shortcomings that make them unsuitable for dialog evaluation: (1) The one-to-many nature of dialog (Zhao et al., 2017) makes wordoverlap metrics ineffective for scoring valid system output that deviates from the ground-truth response (Liu et al.”
Hello, It’s GPT-2 - How Can I Help You? Towards the Use of Pretrained Language Models for Task-Oriented Dialogue Systems (2019, background): “This is in opposition to the standard task-oriented dialogue architectures (Wen et al., 2017; Zhao et al., 2017) where the belief state and the database state are encoded in a numerical form.”
Task-Oriented Dialog Systems that Consider Multiple Appropriate Responses under the Same Context (2019, background): “Previous studies model this one-to-many conversation property to improve utterance-level diversity in open-domain dialog generation (Zhao, Zhao, and Eskenazi 2017; I am so hungry - can you find me a place to eat in the city center?”
Topic-based Evaluation for Conversational Bots (2018, methodology): “Dialog researchers have used metrics of word overlap between a bot response and a reference valid response [19, 10, 23, 24], such as BLEU [17], METEOR [2], and embedding based metrics [24].”
Z-Forcing: Training Stochastic Recurrent Networks (2017, background): “Several efforts have been made in order to bias the learning process towards parameter solutions for which the latent variables are used (Bowman et al., 2015; Karl et al., 2016; Kingma et al., 2016; Chen et al., 2017; Zhao et al., 2017).”
DialogWAE: Multimodal Response Generation with Conditional Wasserstein Auto-Encoder (2018, methodology): “We define n-gram precision and n-gram recall as the average and the maximum score respectively (Zhao et al., 2017).”
Policy-Driven Neural Response Generation for Knowledge-Grounded Dialog Systems (2020, background): “Previous works proposed conditioning response generators on latent (Serban et al., 2016, 2017; Shen et al., 2017; Zhao et al., 2017) or discrete attributes, including dialog acts (Sankar and Ravi, 2019), sentiment (Sankar and Ravi, 2019), speaker identiﬁers (Li et al., 2016a), lexical features (See…”
Think Before You Speak: Explicitly Generating Implicit Commonsense Knowledge for Response Generation (2021, methodology): “For retrieval, we follow most common approaches in knowledge-selection (Zhao et al., 2017; Wolf et al., 2020; Eric et al., 2021) and train RoBERTa (Liu et al., 2019) to classify triples using our knowledge-aligned data (matched or not matched), and use it to label candidate triples during testing (…”
PLATO-2: Towards Building an Open-Domain Chatbot via Curriculum Learning (2020, methodology): “To facilitate the training process of discrete latent variables, the bag-of-words (BOW) loss (Zhao et al., 2017) is also employed: where V refers to the whole vocabulary.”
Knowledge-Grounded Dialogue Generation with Pre-trained Language Models (2020, background): “Later, the vanilla encoder-decoder architecture is widely extended to improve diversity of responses (Li et al., 2015; Xing et al., 2017a; Zhao et al., 2017; Tao et al., 2018); to model the structure of conversation contexts (Ser-ban et al., 2016, 2017; Xing et al., 2017b; Zhang et al., 2019a); to…”
An Introduction to Variational Autoencoders (2019, background): “Another prominent application of VAEs is modeling of text and or sequential data (Bayer and Osendorfer, 2014; Bowman et al., 2015; Serban et al., 2017; Johnson et al., 2016; Karl et al., 2017; Fraccaro et al., 2016; Miao et al., 2016; Semeniuta et al., 2017; Zhao et al., 2017; Yang et al., 2017; Hu ”