Generating High-Quality and Informative Conversation Responses with Sequence-to-Sequence Models

By Yuanlong Shao, Stephan Gouws, Denny Britz, Anna Goldie, Brian Strope, R. Kurzweil. 2017

218 citations recorded by the source corpus.

Sequence-to-sequence models have been applied to the conversation response generation problem where the source sequence is the conversation history and the target sequence is the response. Unlike translation, conversation responding is inherently creative. The generation of long, informative, coherent, and diverse responses remains a hard task. In this work, we focus on the single turn setting. We add self-attention to the decoder to maintain coherence in longer responses, and we propose a practical approach, called the glimpse-model, for scaling to large datasets. We introduce a stochastic beam-search algorithm with segment-by-segment reranking which lets us inject diversity earlier in the generation process. We trained on a combined data set of over 2.3B conversation messages mined from the web. In human evaluation studies, our method produces longer responses overall, with a higher proportion rated as acceptable and excellent as length increases, compared to baseline sequence-to-sequence models with explicit length-promotion. A back-off strategy produces better responses overall, in the full spectrum of lengths.

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

GLTR: Statistical Detection and Visualization of Generated Text (2019, background): “…sample from the head of the distribution, e.g., through max sampling (Gu et al., 2017), k-max sampling (Fan et al., 2018), beam search (Chorowski and Jaitly, 2016; Shao et al., 2017), temperature-modulated sampling (Dagan and Engelson, 1995), or even implicitly with rule-based templated approaches.”
Towards a Human-like Open-Domain Chatbot (2020, methodology): “…approach to mitigating this problem is to use more sophisticated decoding algorithms, for instance with different forms of re-ranking (Li et al., 2016a; Shao et al., 2017) or conditioning on proﬁles, topics, and styles (Li et al., 2016b; Wang et al., 2017; Xing et al., 2017; Zhang et al., 2018b).”
A Retrieve-and-Edit Framework for Predicting Structured Outputs (2018, background): “In prediction tasks with complex outputs, generating well-formed outputs is challenging, as is wellknown in natural language generation [20, 28].”
Generating Sentences by Editing Prototypes (2017, background): “At the same time, naive strategies to increase diversity have been shown to compromise grammaticality (Shao et al., 2017), suggesting that current NLMs may lack the inductive bias to faithfully represent the full diversity of complex utterances.”
Towards Coherent and Engaging Spoken Dialog Response Generation Using Automatic Conversation Evaluators (2019, methodology): “Our method extends the strategy employed by Shao et al. (2017) utilizing a trained model as the reranking function.”
Learning to Write with Cooperative Discriminators (2018, background): “Alternative Decoding Objectives A number of papers have proposed alternative decoding objectives for generation (Shao et al., 2017).”
Neural Approaches to Conversational AI (2018, positive): “models for dialogue have been shown to be useful (Yao et al., 2015; Mei et al., 2017; Shao et al., 2017), e.”
A Discourse-Aware Attention Model for Abstractive Summarization of Long Documents (2018, background): “Seq2seq models tend to struggle with longer sequences because at each decoding step, the decoder needs to learn to construct a context vector capturing relevant information from all the tokens in the source sequence (Shao et al., 2017).”
Look Before you Speak: Visually Contextualized Utterances (2020, background): “Approaches include hand-crafted rules [17, 80], examplebased agents [39, 44] and modern neural networks [70, 72], and aim to generate realistic responses for goal-oriented dialog systems or chatbots.”
Generating More Interesting Responses in Neural Conversation Models with Distributional Constraints (2018, methodology): “As a starting point for our approach we leverage the Seq2Seq model (Sutskever et al., 2014; Bah-danau et al., 2014) which has been used as a basis for a broad range of recent work on neural conversation (Kannan et al., 2016; Li et al., 2016a; Serban et al., 2016; Shao et al., 2017).”
Better Automatic Evaluation of Open-Domain Dialogue Systems with Contextualized Embeddings (2019, background): “Due to the lack of strong automatic evaluation metrics, many researchers resort primarily to human evaluation for assessing their dialogue systems performances (Shang et al., 2015; Sordoni et al., 2015; Shao et al., 2017).”
Wasserstein Barycenter Model Ensembling (2019, methodology): “barycenter via the entropic regualrization parameter ε, we also decode using randomized Beam search of (Shao et al., 2017), where instead of maintaining the top k values, we sample D candidates in each beam.”
Sounding Board: A User-Centric and Content-Driven Social Chatbot (2018, background): “General conversational systems, called chatbots, have constrained social interaction capabilities but have difficulty generating conversations with long-term coherence (Serban et al., 2017; Sato et al., 2017; Shao et al., 2017; Tian et al., 2017; Ghazvininejad et al., 2018).”
Polite Dialogue Generation Without Parallel Data (2018, background): “Lastly, there have also been style transfer models that rely on the latent representation of text and use variational auto-encoders or cross-alignment to disentangle the representation of content and style in text (Hu et al., 2017; Shen et al., 2017; Zhao et al., 2017; Fu et al., 2018).”
Multi-Task Learning for Speaker-Role Adaptation in Neural Conversation Models (2017, background): “The system presented in this paper better represents the speaker role (support person), domain of expertise (technical), and speaking style (courteous). sponses that are often commonplace and uninter-esting (Li et al., 2016a; Shao et al., 2017).”
LSDSCC: a Large Scale Domain-Specific Conversational Corpus for Response Generation with Diversity Oriented Evaluation Metrics (2018, background): “…metrics, that can be taken to measure both candidate response’s diversity and its relevance to the given query, to effectively guide the training of NRG models towards the state of promoting meaningful and diverse responses (Li et al., 2016a; Shao et al., 2017; Freitag and Al-Onaizan, 2017).”