Retrieve and Refine: Improved Sequence Generation Models For Dialogue

By Jason Aaron Edward Weston, Emily Dinan, Alexander H. Miller. 2018

209 citations recorded by the source corpus.

Sequence generation models for dialogue are known to have several problems: they tend to produce short, generic sentences that are uninformative and unengaging. Retrieval models on the other hand can surface interesting responses, but are restricted to the given retrieval set leading to erroneous replies that cannot be tuned to the specific context. In this work we develop a model that combines the two approaches to avoid both their deficiencies: first retrieve a response and then refine it – the final sequence generator treating the retrieval as additional context. We show on the recent ConvAI2 challenge task our approach produces responses superior to both standard retrieval and generation models in human evaluations.

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

Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (2020, background): “Such tasks include open-domain question answering [5, 25], fact checking [50], fact completion [42], long-form question answering [12], Wikipedia article generation [32], dialogue [36, 59, 9, 13], translation [16], and language modeling [17, 23].”
“I’m sorry to hear that”: Finding New Biases in Language Models with a Holistic Descriptor Dataset (2022, methodology): “We tag each response with a binary label indicating its level of bias, and we then perform style-controlled generation on those labels so that the model can be prompted to generate responses containing lower amounts of bias (Weston et al., 2018; Smith et al., 2020a).”
Open-Domain Conversational Agents: Current Progress, Open Problems, and Future Directions (2020, methodology): “Until recently, their performance was below that of retrieval models (Dinan et al., 2019b; Li, Weston, and Roller, 2019; Rashkin et al., 2019; Shuster et al., 2019; Zhang et al., 2018), unless they relied on refining retrieved utterances (Weston, Dinan, and Miller, 2018).”
Simple and Effective Retrieve-Edit-Rerank Text Generation (2020, methodology): “Retrieve-and-edit methods have been developed for summarization (Cao et al., 2018), machine translation (Wu et al., 2019), language modeling (Guu et al., 2018), and conversation generation (Weston et al., 2018).”
Neural Text Generation with Unlikelihood Training (2019, methodology): “Some other lines of work of note are the use of generative control (Kikuchi et al., 2016; Fan et al., 2017; Peng et al., 2018; See et al., 2019) and the retrieve-and-reﬁne setting (Weston et al., 2018; Guu et al., 2018).”
Recipes for Building an Open-Domain Chatbot (2020, background): “the human utterances they were trained on (Weston et al., 2018).”
The Second Conversational Intelligence Challenge (ConvAI2) (2019, neutral): “A retrieve and reﬁne approach (combining generative and retrieval methods) is another possibility that was not explored in the competition [10].”
Importance of Search and Evaluation Strategies in Neural Dialogue Modeling (2018, background): “We concentrate on a second approach for evaluation where a neural dialogue model has a multiturn conversation with a human partner (or annotator) (Zhang et al., 2018; Zemlyanskiy and Sha, 2018; Weston et al., 2018; Dinan et al., 2019).”
Controlling Style in Generated Dialogue (2020, methodology): “Based on results in Weston et al. (2018) showing a superiority of retrieve-and-refine models over either retrieval or generation methods for dialogue, we considered an intuitive approach combining retrieval and style-controlled generation.”
Challenges in Building Intelligent Open-domain Dialog Systems (2019, methodology): “roduce novel responses. But they often generate undesirable responses that are either ungrammatical or irrelevant. Hybrid methods combine the strengths of both and usually adopt a two-stage procedure [153, 165, 192]. In the first stage, some relevant conversations, known as prototype responses in [1”
Non-Parametric Adaptation for Neural Machine Translation (2019, background): “Beyond NMT, there have been a few other attempts to incorporate non-parametric approaches into neural generative models (Guu et al., 2018; Hayati et al., 2018; Weston et al., 2018).”
Justifying Recommendations using Distantly-Labeled Reviews and Fine-Grained Aspects (2019, background): “In dialogue systems, several works have studied frameworks to extract templates from historical responses, which are then edited to form new responses (Weston et al., 2018; Wu et al., 2018).”
Linguistic calibration through metacognition: aligning dialogue agent responses with expected correctness (2020, background): “Retrieve-and-reﬁne generation (Weston et al., 2018; Roller et al., 2020) conditions on a possible answer, but does not control the style of the response.”
Controllable Semantic Parsing via Retrieval Augmentation (2021, methodology): “Some methods, like CASPER, encode the retrievals alongside the query and let the model decides how to use them (Guu et al., 2018; Hashimoto et al., 2018; He et al., 2020; Weston et al., 2018; Pandey et al., 2018; Lewis et al., 2020).”
Controllable Paraphrase Generation with a Syntactic Exemplar (2019, background): “Another related line of work builds generation upon sentential exemplars (Guu et al., 2018; Weston et al., 2018; Pandey et al., 2018; Cao et al., 2018; Peng et al., 2019) in order to improve the quality of the generation itself, rather than to allow for control over syntactic structures.”