Recipes for Building an Open-Domain Chatbot
Published 1 January 2021Open access
Stephen Roller, Emily Dinan, Naman Goyal, Da Young Ju, Mary Williamson, Yinhan Liu
Citations171
Generate an AI Snapshot to get a quick, structured summary of this paper.
Study Snapshot
ObjectiveStudy objective
MethodsResearch methodology
PopulationPopulation studied
Sample sizeSample sizes
OutcomesStudy outcomes here
ResultsStudy results comes here
LimitationsResearch study limitations comes here
A concise AI-generated summary of the paper will appear here once you click Generate AI Snapshot.
TL;DR
Human evaluations show the best models outperform existing approaches in multi-turn dialogue on engagingness and humanness measurements, and the limitations of this work are discussed by analyzing failure cases of the models.
Abstract
Stephen Roller, Emily Dinan, Naman Goyal, Da Ju, Mary Williamson, Yinhan Liu, Jing Xu, Myle Ott, Eric Michael Smith, Y-Lan Boureau, Jason Weston. Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume. 2021.
Keywords
Computer Science
UvA-DARE (University of Amsterdam)Adam: A Method for Stochastic Optimization
84,783 Citations2014Diederik P. Kingma, Jimmy Ba
32,525 Citations2019Jacob Devlin, Ming‐Wei Chang +2 more
A new language representation model, BERT, designed to pre-train deep bidirectional representations from unlabeled text by jointly conditioning on both left and right context in all layers, which can be fine-tuned with just one additional output layer to create state-of-the-art models for a wide range of tasks.
DROPS (Schloss Dagstuhl – Leibniz Center for Informatics)HISTORIAE, History of Socio-Cultural Transformation as Linguistic Data Science. A Humanities Use Case
17,334 Citations2019Yinhan Liu, Myle Ott +8 more
This work considers the task of building machine learning models to automatically select the best combination for a problem instance and contributes to the automatic learning of instance features directly from the high-level representation of a problem instance using a transformer encoder.
arXiv (Cornell University)Longformer: The Long-Document Transformer
2,200 Citations2020Iz Beltagy, Matthew E. Peters
Following prior work on long-sequence transformers, the Longformer is evaluated on character-level language modeling and achieves state-of-the-art results on text8 and enwik8 and pretrain Longformer and finetune it on a variety of downstream tasks.
arXiv (Cornell University)Scaling Laws for Neural Language Models
1,467 Citations2020Jared Kaplan, Sam McCandlish +8 more
Larger models are significantly more sample-efficient, such that optimally compute-efficient training involves training very large models on a relatively modest amount of data and stopping significantly before convergence.
arXiv (Cornell University)A Deep Reinforced Model for Abstractive Summarization
1,272 Citations2017Romain Paulus, Caiming Xiong +1 more
BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension
1,222 Citations2020Mike Lewis, Yinhan Liu +6 more
BART is presented, a denoising autoencoder for pretraining sequence-to-sequence models, which matches the performance of RoBERTa on GLUE and SQuAD, and achieves new state-of-the-art results on a range of abstractive dialogue, question answering, and summarization tasks.
Personalizing Dialogue Agents: I have a dog, do you have pets too?
1,146 Citations2018Saizheng Zhang, Emily Dinan +4 more
This work collects data and train models tocondition on their given profile information; and information about the person they are talking to, resulting in improved dialogues, as measured by next utterance prediction.
Hierarchical Neural Story Generation
1,124 Citations2018Angela Fan, Mike Lewis +1 more
This work collects a large dataset of 300K human-written stories paired with writing prompts from an online forum that enables hierarchical story generation, where the model first generates a premise, and then transforms it into a passage of text.
arXiv (Cornell University)GPipe: Efficient Training of Giant Neural Networks using Pipeline\n Parallelism
886 Citations2018Yanping Huang, Youlong Cheng +9 more
arXiv (Cornell University)Mixed Precision Training
861 Citations2017Paulius Micikevicius, Sharan Narang +9 more
This work introduces a technique to train deep neural networks using half precision floating point numbers, and demonstrates that this approach works for a wide variety of models including convolution neural networks, recurrent neural networks and generative adversarial networks.
Proceedings of the International AAAI Conference on Web and Social MediaThe Pushshift Reddit Dataset
825 Citations2020Jason Baumgartner, Savvas Zannettou +3 more
arXiv (Cornell University)Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallelism
818 Citations2019Mohammad Shoeybi, Mostofa Patwary +4 more
A simple, efficient intra-layer model parallel approach that enables training transformer models with billions of parameters and shows that careful attention to the placement of layer normalization in BERT-like models is critical to achieving increased performance as the model size grows.
arXiv (Cornell University)CTRL: A Conditional Transformer Language Model for Controllable Generation
786 Citations2019Nitish Shirish Keskar, Bryan McCann +3 more
CTRL is released, a 1.63 billion-parameter conditional transformer language model, trained to condition on control codes that govern style, content, and task-specific behavior, providing more explicit control over text generation.
Frontiers of Information Technology & Electronic EngineeringFrom Eliza to XiaoIce: challenges and opportunities with social chatbots
720 Citations2018Heung‐Yeung Shum, Xiaodong He +1 more
The success metric for social chatbots is defined as conversation-turns per session (CPS), and it is shown how XiaoIce can dynamically recognize emotion and engage the user throughout long conversations with appropriate interpersonal responses.
arXiv (Cornell University)Bootstrapping a Data-Set and Model for Question-Answering in Portuguese (Short Paper)
586 Citations2021Zihang Dai, Zhilin Yang +1 more
An exploratory work carried out to come up with a state-of-the-art model for question-answering tasks, for the Portuguese language, based on deep neural networks, achieves 50% accuracy in the testing data-set.
arXiv (Cornell University)The Curious Case of Neural Text Degeneration
527 Citations2020Ari Holtzman, Jan Buys +3 more
By sampling text from the dynamic nucleus of the probability distribution, which allows for diversity while effectively truncating the less reliable tail of the distribution, the resulting text better demonstrates the quality of human text, yielding enhanced diversity without sacrificing fluency and coherence.
The Springer series on challenges in machine learningThe Second Conversational Intelligence Challenge (ConvAI2)
362 Citations2019Emily Dinan, Varvara Logacheva +15 more
To improve performance on multi-turn conversations with humans, future systems must go beyond single word metrics like perplexity to measure the performance across sequences of utterances (conversations)—in terms of repetition, consistency and balance of dialogue acts.
arXiv (Cornell University)TransferTransfo: A Transfer Learning Approach for Neural Network Based Conversational Agents
282 Citations2019Thomas Wolf, Victor Sanh +2 more
A new approach to generative data-driven dialogue systems (e.g. chatbots) called TransferTransfo is introduced which is a combination of a Transfer learning based training scheme and a high-capacity Transformer model which shows strong improvements over the current state-of-the-art end-to-end conversational models.
arXiv (Cornell University)Towards a Human-like Open-Domain Chatbot
266 Citations2020Daniel Adiwardana
Meena, a multi-turn open-domain chatbot trained end-to-end on data mined and filtered from public domain social media conversations, is presented and a human evaluation metric called Sensibleness and Specificity Average (SSA) is proposed, which captures key elements of a human-like multi- turn conversation.
arXiv (Cornell University)Wizard of Wikipedia: Knowledge-Powered Conversational Agents
245 Citations2018Emily Dinan, Stephen Roller +4 more
The best performing dialogue models are able to conduct knowledgeable discussions on open-domain topics as evaluated by automatic metrics and human evaluations, while a new benchmark allows for measuring further improvements in this important research direction.
Training Millions of Personalized Dialogue Agents
215 Citations2018Pierre-Emmanuel Mazaré, Samuel Humeau +2 more
A new dataset providing 5 million personas and 700 million persona-based dialogues is introduced and it is shown that, at this scale, training using personas still improves the performance of end-to-end systems.
arXiv (Cornell University)A Persona-Based Neural Conversation Model
181 Citations2016Jiwei Li, Michel Galley +4 more
Retrieve and Refine: Improved Sequence Generation Models For Dialogue
176 Citations2018Jason Weston, Emily Dinan +1 more
This work develops 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.
International Conference on Learning RepresentationsPoly-encoders: Architectures and Pre-training Strategies for Fast and Accurate Multi-sentence Scoring
170 Citations2020Samuel Humeau, Kurt Shuster +2 more
This work develops a new transformer architecture, the Poly-encoder, that learns global rather than token level self-attention features, and shows that the models achieve state-of-the-art results on four tasks.
arXiv (Cornell University)fairseq: A Fast, Extensible Toolkit for Sequence Modeling
163 Citations2019Myle Ott, Sergey Edunov +6 more
arXiv (Cornell University)Adafactor: Adaptive Learning Rates with Sublinear Memory Cost
163 Citations2018Noam Shazeer, Mitchell Stern
arXiv (Cornell University)Building a Conversational Agent Overnight with Dialogue Self-Play
161 Citations2018Pararth Shah, Dilek Hakkani‐Tür +5 more
A new corpus of 3,000 dialogues spanning 2 domains collected with M2M is proposed, and comparisons with popular dialogue datasets on the quality and diversity of the surface forms and dialogue flows are presented.
Learning Semantic Textual Similarity from Conversations
155 Citations2018Yinfei Yang, Steve Yuan +8 more
A novel approach to learn representations for sentence-level semantic similarity using conversational data and achieves the best performance among all neural models on the STS Benchmark and is competitive with the state-of-the-art feature engineered and mixed systems for both tasks.
International Conference on Learning RepresentationsNeural Text Generation With Unlikelihood Training
130 Citations2020Sean Welleck, Ilia Kulikov +4 more
It is shown that the likelihood objective itself is at fault, resulting in a model that assigns too much probability to sequences containing repeats and frequent words, unlike those from the human training distribution, thus providing a strong alternative to existing techniques.
Bootstrapping a Neural Conversational Agent with Dialogue Self-Play, Crowdsourcing and On-Line Reinforcement Learning
127 Citations2018Pararth Shah, Dilek Hakkani‐Tür +2 more
This paper discusses the advantages of this approach for industry applications of conversational agents, wherein an agent can be rapidly bootstrapped to deploy in front of users and further optimized via interactive learning from actual users of the system.
ParlAI: A Dialog Research Software Platform
109 Citations2017Alexander Miller, Will Feng +6 more
ParlAI (pronounced “par-lay”), an open-source software platform for dialog research implemented in Python, is introduced, to provide a unified framework for sharing, training and testing dialog models; integration of Amazon Mechanical Turk for data collection, human evaluation, and online/reinforcement learning.
arXiv (Cornell University)DialoGPT: Large-Scale Generative Pre-training for Conversational Response Generation
104 Citations2019Yizhe Zhang, Siqi Sun +7 more
arXiv (Cornell University)ACUTE-EVAL: Improved Dialogue Evaluation with Optimized Questions and Multi-turn Comparisons
79 Citations2019Margaret Li, Jason Weston +1 more
A novel procedure involving comparing two full dialogues, where a human judge is asked to pay attention to only one speaker within each, and make a pairwise judgment, resulting in better tests.
Experience Grounds Language
74 Citations2020Yonatan Bisk, Ari Holtzman +10 more
It is posited that the present success of representation learning approaches trained on large text corpora can be deeply enriched from the parallel tradition of research on the contextual and social nature of language.
DSpace@MIT (Massachusetts Institute of Technology)Approximating Interactive Human Evaluation with Self-Play for Open-Domain Dialog Systems
33 Citations2019Asma Ghandeharioun, Judy Hanwen Shen +5 more
It is shown that this metric is capable of capturing the human-rated quality of a dialog model better than any automated metric known to-date, achieving a significant Pearson correlation (r>.7, p<.05).
arXiv (Cornell University)Dialog-based Language Learning
27 Citations2016Jason Weston
This work studies dialog-based language learning, where supervision is given naturally and implicitly in the response of the dialog partner during the conversation, and shows that a novel model incorporating predictive lookahead is a promising approach for learning from a teacher's response.
arXiv (Cornell University)Don't Say That! Making Inconsistent Dialogue Unlikely with Unlikelihood Training
20 Citations2019Margaret Li, Stephen Roller +5 more
arXiv (Cornell University)Learning Robust Dialog Policies in Noisy Environments
19 Citations2017Maryam Fazel-Zarandi, Shang-Wen Li +5 more
A user simulator is developed which interacts with the assistant through voice commands in realistic scenarios with noisy audio, and is used to learn dialog policies through deep reinforcement learning, showing that dialogs generated by the simulator are indistinguishable from human generated dialogs, as determined by human evaluators.
Queens are Powerful too: Mitigating Gender Bias in Dialogue Generation
14 Citations2020Emily Dinan, Angela Fan +4 more
This work measures gender bias in dialogue data, and examines how this bias is actually amplified in subsequent generative chit-chat dialogue models, and considers three techniques to mitigate gender bias: counterfactual data augmentation, targeted data collection, and bias controlled training.
arXiv (Cornell University)Analyzing the Forgetting Problem in the Pretrain-Finetuning of Dialogue Response Models
12 Citations2019Tianxing He, Jun Liu +5 more
This work focuses on the transformer encoder-decoder model for the open-domain dialogue response generation task and proposes an intuitive fine-tuning strategy named "mix-review", which finds that mix-review effectively regularize the fine- Tuning process, and the forgetting problem is largely alleviated.
The Dialogue Dodecathlon: Open-Domain Knowledge and Image Grounded Conversational Agents
9 Citations2020Kurt Shuster, Da Young Ju +4 more
D dodecaDialogue is introduced, a set of 12 tasks that measures if a conversational agent can communicate engagingly with personality and empathy, and that the multi-tasking in general provides gains to both text and image-based tasks using several metrics in both the fine-tune and task transfer settings.
Sequence Effects in Crowdsourced Annotations
9 Citations2017Nitika Mathur, Timothy Baldwin +1 more
This paper explores sequence effects where annotations of an item are affected by the preceding items, and recommends a simple way to minimise sequence effects.
arXiv (Cornell University)I love your chain mail! Making knights smile in a fantasy game world: Open-domain goal-oriented dialogue agents
8 Citations2020Shrimai Prabhumoye, Margaret Li +5 more
A Goal-oriented Neural Conversation Model by Self-Play
1 Citations2018Wei Wei, Quoc V. Le +2 more
This paper proposes a new setting in goal-oriented dialogue system to tighten the gap between these two aspects by enforcing model level information isolation on individual models between two agents.