Controllable Text Generation

Generative Models

VAE

1. Extracting and composing robust features with denoising autoencoders Pascal Vincent, Hugo Larochelle, Yoshua Bengio, Pierre-Antoine Manzagol ICML08 [PDF]
2. Auto-Encoding Variational Bayes Diederik P Kingma, Max Welling ICLR14 [PDF]
3. Generating Sentences from a Continuous Space Samuel R. Bowman, Luke Vilnis CONLL16 [PDF]
4. Adversarial Autoencoders Alireza Makhzani, Jonathon Shlens, Navdeep Jaitly, Ian Goodfellow, Brendan Frey ICLR16 [PDF]
5. Learning Structured Output Representation using Deep Conditional Generative Models Kihyuk Sohn, Xinchen Yan, Honglak Lee NIPS15 [PDF]
6. Deep Unsupervised Clustering with Gaussian Mixture Variational Auotoencoders Nat Dilokthanakul, Pedro A. M. Mediano, Marta Garnelo, Matthew C. H. Lee, Hugh Salimbeni, Kai Arulkumaran, Murray Shanahan ICLR17 [PDF]
7. Wasserstein Auto-Encoders Ilya Tolstikhin, Olivier Bousquet, Sylvain Gelly, Bernhard Sch¨olkopf ICLR18 [PDF] [code]
8. Adversarially Regularized Autoencoders Jake Zhao, Yoon Kim, Kelly Zhang, Alexander M. Rush, Yann LeCun ICML18 [PDF] [code]
   model

   
   algorithm

   
9. T-CVAE: Transformer-Based Conditioned Variational Autoencoder for Story Completion Tianming Wang, Xiaojun Wan IJCAI19 [PDF] [code]
   abstract

   Story completion is a very challenging task of generating the missing plot for an incomplete story, which requires not only understanding but also inference of the given contextual clues. In this paper, we present a novel conditional variational autoencoder based on Transformer for missing plot generation. Our model uses shared attention layers for encoder and decoder, which make the most of the contextual clues, and a latent variable for learning the distribution of coherent story plots. Through drawing samples from the learned distribution, diverse reasonable plots can be generated. Both automatic and manual evaluations show that our model generates better story plots than stateof-the-art models in terms of readability, diversity and coherence.
   model

   
10. Transformer-based Conditional Variational Autoencoder for Controllable Story Generation Le Fang, Tao Zeng, Chaochun Liu, Liefeng Bo, Wen Dong, Changyou Chen arXiv [PDF] [code]
    abstract

    We investigate large-scale latent variable models (LVMs) for neural story generation—an under-explored application for open-domain long text—with objectives in two threads: generation effectiveness and controllability. LVMs, especially the variational autoencoder (VAE), have achieved both effective and controllable generation through exploiting flexible distributional latent representations. Recently, Transformers and its variants have achieved remarkable effectiveness without explicit latent representation learning, thus lack satisfying controllability in generation. In this paper, we advocate to revive latent variable modeling, essentially the power of representation learning, in the era of Transformers to enhance controllability without hurting state-of-the-art generation effectiveness. Specifically, we integrate latent representation vectors with a Transformer-based pre-trained architecture to build conditional variational autoencoder (CVAE). Model components such as encoder, decoder and the variational posterior are all built on top of pre-trained language models—GPT2 specifically in this paper. Experiments demonstrate state-of-the-art conditional generation ability of our model, as well as its excellent representation learning capability and controllability.
    model

    
11. APo-VAE: Text Generation in Hyperbolic Space Shuyang Dai, Zhe Gan, Yu Cheng, Chenyang Tao, Lawrence Carin, Jingjing Liu NAACL21 [PDF] [code]

GAN

Normalizing Flow

1. NICE: Non-linear Independent Components Estimation Laurent Dinh, David Krueger, Yoshua Bengio arXiv [PDF]
2. Density estimation using Real NVP Laurent Dinh, Jascha Sohl-Dickstein, Samy Bengio arXiv [PDF]
3. Glow: Generative Flow with Invertible 1×1 Convolutions NeurIPS18 Diederik P. Kingma, Prafulla Dhariwal [PDF] [code]
4. Masked Autoregressive Flow for Density Estimation NeurIPS17 George Papamakarios, Theo Pavlakou, Iain Murray [PDF]
5. Improved Variational Inference with Inverse Autoregressive Flow NeurIPS16 Diederik P. Kingma, Tim Salimans, Rafal Jozefowicz, Xi Chen
6. FlowSeq: Non-Autoregressive Conditional Sequence Generation with Generative Flow EMNLP2019 Xuezhe Ma, Chunting Zhou, Xian Li, Graham Neubig, Eduard Hovy [PDF] [code]
7. Continuous Language Generative Flow ACL21 Zineng Tang, Shiyue Zhang, Hyounghun Kim, Mohit Bansal [PDF] [code]

Denoising Diffusion Model

1. Deep Unsupervised Learning using Nonequilibrium Thermodynamics ICML15 Jascha Sohl-Dickstein, Eric Weiss, Niru Maheswaranathan, Surya Ganguli [PDF]
2. Generative Modeling by Estimating Gradients of the Data Distribution NeurIPS19 Yang Song, Stefano Ermon [PDF]
3. Improved Techniques for Training Score-Based Generative Models NeurIPS20 Yang Song, Stefano Ermon [PDF]
4. Denoising Diffusion Probabilistic Models NeurIPS20 Jonathan Ho, Ajay Jain, Pieter Abbeel [PDF] [code]
5. Diffusion Models Beat GANs on Image Synthesis NeurIPS21 Prafulla Dhariwal, Alex Nichol [PDF]
6. GLIDE: Towards Photorealistic Image Generation and Editing with Text-Guided Diffusion Models arXiv Alex Nichol, Prafulla Dhariwal, Aditya Ramesh, Pranav Shyam, Pamela Mishkin, Bob McGrew, Ilya Sutskever, Mark Chen [PDF]
7. Diffusion-LM Improves Controllable Text Generation NeurIPS22 Xiang Lisa Li, John Thickstun, Ishaan Gulrajani, Percy Liang, Tatsunori B. Hashimoto [PDF] [code]

Language Model

• BERT, GPT, T5, BART, ...

Control with Fixed Language Model

Weighted Decoding

1. Plug and Play Language Models: A Simple Approach to Controlled Text Generation Sumanth Dathathri, Andrea Madotto, Janice Lan, Jane Hung, Eric Frank, Piero Molino, Jason Yosinski, Rosanne Liu ICLR20 [PDF] [code]
2. FUDGE: Controlled Text Generation With Future Discriminators Kevin Yang, Dan Klein NAACL21 [PDF] [code]
3. GeDi: Generative Discriminator Guided Sequence Generation Ben Krause, Akhilesh Deepak Gotmare, Bryan McCann, Nitish Shirish Keskar, Shafiq Joty, Richard Socher, Nazneen Fatema Rajani EMNLP21 findings [PDF] [code]
4. DExperts: Decoding-Time Controlled Text Generation with Experts and Anti-Experts Alisa Liu, Maarten Sap, Ximing Lu, Swabha Swayamdipta, Chandra Bhagavatula, Noah A. Smith, Yejin Choi ACL21 [PDF]
5. Plug-and-Blend: A Framework for Controllable Story Generation with Blended Control Codes Zhiyu Lin, Mark Riedl arXiv21 [PDF] [code]
6. Improving Controllable Text Generation with Position-Aware Weighted Decoding ACL22 findings Yuxuan Gu, Xiaocheng Feng, Sicheng Ma, Jiaming Wu, Heng Gong, Bing Qin [PDF]
7. Bridging the Gap Between Training and Inference of Bayesian Controllable Language Models arXiv Han Liu, Bingning Wang, Ting Yao, Haijin Liang, Jianjin Xu, Xiaolin Hu [PDF]
8. Classifiers are Better Experts for Controllable Text Generation arXiv Askhat Sitdikov, Nikita Balagansky, Daniil Gavrilov, Alexander Markov [PDF]
9. Gamma Sampling: Fine-grained Controlling Language Models without Training arXiv Shangda Wu, Maosong Sun [PDF]
10. Controllable Text Generation with Neurally-Decomposed Oracle NeurIPS22 Tao Meng, Sidi Lu, Nanyun Peng, Kai-Wei Chang [PDF]

Multi-Objective Optimization

1. Controlled Text Generation as Continuous Optimization with Multiple Constraints NeurIPS21 Sachin Kumar, Eric Malmi, Aliaksei Severyn, Yulia Tsvetkov [PDF]
2. Mix and Match: Learning-free Controllable Text Generation using Energy Language Models ACL22 Fatemehsadat Mireshghallah, Kartik Goyal, Taylor Berg-Kirkpatrick [PDF]
3. COLD Decoding: Energy-based Constrained Text Generation with Langevin Dynamics arXiv Lianhui Qin, Sean Welleck, Daniel Khashabi, Yejin Choi [PDF]
4. Constrained Sampling from Language Models via Langevin Dynamics in Embedding Spaces arXiv Sachin Kumar, Biswajit Paria, Yulia Tsvetkov [PDF]
5. Composable Text Controls in Latent Space with ODEs arXiv Guangyi Liu, Zeyu Feng, Yuan Gao, Zichao Yang, Xiaodan Liang, Junwei Bao, Xiaodong He, Shuguang Cui, Zhen Li, Zhiting Hu [PDF]

Prefix-Tuning

1. Attribute Alignment: Controlling Text Generation from Pre-trained Language Models EMNLP21 findings Dian Yu, Zhou Yu, Kenji Sagae [PDF]
2. Controllable Natural Language Generation with Contrastive Prefixes ACL22 findings Jing Qian, Li Dong, Yelong Shen, Furu Wei, Weizhu Chen [PDF]
3. Fine-Grained Controllable Text Generation Using Non-Residual Prompting ACL22 Fredrik Carlsson, Joey Öhman, Fangyu Liu, Severine Verlinden, Joakim Nivre, Magnus Sahlgren [PDF]
4. Tailor: A Prompt-Based Approach to Attribute-Based Controlled Text Generation arXiv Kexin Yang, Dayiheng Liu, Wenqiang Lei, Baosong Yang, Mingfeng Xue, Boxing Chen, Jun Xie [PDF]
5. Quark: Controllable Text Generation with Reinforced [Un]learning NeurIPS22 Ximing Lu, Sean Welleck, Liwei Jiang, Jack Hessel, Lianhui Qin, Peter West, Prithviraj Ammanabrolu, Yejin Choi [PDF]

Controlling information

Text as exemplar

Unparallel text

1. Disentangled Representation Learning for Non-Parallel Text Style Transfer Vineet John, Lili Mou, Hareesh Bahuleyan, Olga Vechtomova ACL19 [PDF] [code]

   motivation

   To tackle the problem of disentangling the latent representations of style and content in language models.
   model

   

2. Pre-train and Plug-in: Flexible Conditional Text Generation with Variational Auto-Encoders Yu Duan, Canwen Xu, Jiaxin Pei, Jialong Han, Chenliang Li ACL20 [PDF] [code]

   motivation

   Flexible when new conditions added to a well trained VAE which requires no more retraining.
   abstract

   Conditional Text Generation has drawn much attention as a topic of Natural Language Generation (NLG) which provides the possibility for humans to control the properties of generated contents. Current conditional generation models cannot handle emerging conditions due to their joint end-to-end learning fashion. When a new condition added, these techniques require full retraining. In this paper, we present a new framework named Pre-train and Plug-in Variational Auto-Encoder (PPVAE) towards flexible conditional text generation. PPVAE decouples the text generation module from the condition representation module to allow “one-to-many” conditional generation. When a fresh condition emerges, only a lightweight network needs to be trained and works as a plug-in for PPVAE, which is efficient and desirable for real-world applications. Extensive experiments demonstrate the superiority of PPVAE against the existing alternatives with better conditionality and diversity but less training effort.
   model

   

3. Plug and Play Autoencoders for Conditional Text Generation Florian Mai, Nikolaos Pappas, Ivan Montero, Noah A. Smith, James Henderson EMNLP20[PDF] [code]

   motivation

   Reduce the need for labeled training data for the task and makes the training procedure more efficient by learning a mapping within the autoencoder’s embedding space.
   
   model

   

4. Stylized Dialogue Response Generation Using Stylized Unpaired Texts Yinhe Zheng, Zikai Chen, Rongsheng Zhang, Shilei Huang, Xiaoxi Mao, Minlie Huang AAAI 21 [PDF] [code]

   model

   
   algorithm

   

5. Plug and Play Language Models: A Simple Approach to Controlled Text Generation Sumanth Dathathri, Andrea Madotto, Janice Lan, Jane Hung, Eric Frank, Piero Molino, Jason Yosinski, Rosanne Liu ICLR20 [PDF] [code]

6. Hooks in the Headline: Learning to Generate Headlines with Controlled Styles Di Jin, Zhijing Jin, Joey Tianyi Zhou, Lisa Orii, Peter Szolovits ACL20 [PDF] [code]

   motivation

   Enrich headlines with controlled style options.
   
   model

   

Parallel text

Text signals

1. CTRL: A Conditional Transformer Language Model for Controllable Generation Nitish Shirish Keskar, Bryan McCann, Lav R. Varshney, Caiming Xiong, Richard Socher arXiv [PDF] [code]
   control code type

   Style by domain

   More complex control codes

   Triggering specific tasks

   Zero-shot code-mixing

2. A Controllable Model of Grounded Response Generation Zeqiu Wu, Michel Galley, Chris Brockett, Yizhe Zhang, Xiang Gao, Chris Quirk, Rik Koncel-Kedziorski, Jianfeng Gao, Hannaneh Hajishirzi, Mari Ostendorf, Bill Dolan AAAI21 [PDF] [code]
   model

   
3. Generate Your Counterfactuals: Towards Controlled Counterfactual Generation for Text Nishtha Madaan, Inkit Padhi, Naveen Panwar, Diptikalyan Saha AAAI21 [PDF] [code]
4. GEDI: GENERATIVE DISCRIMINATOR GUIDED SEQUENCE GENERATION Ben Krause, Akhilesh Deepak Gotmare, Bryan McCann, Nitish Shirish Keskar, Shafiq Joty, Richard Socher, Nazneen Fatema Rajani arXiv [PDF] [code]
5. MEGATRON-CNTRL: Controllable Story Generation with External Knowledge Using Large-Scale Language Models Peng Xu, Mostofa Patwary, Mohammad Shoeybi, Raul Puri, Pascale Fung, Anima Anandkumar, Bryan Catanzaro EMNLP20 [PDF] [code]

Syntactic Guidance

1. Syntax-guided Controlled Generation of Paraphrases Ashutosh Kumar, Kabir Ahuja, Raghuram Vadapalli, Partha Talukdar TACL20 [PDF] [code]
   model

   
2. Transformer-Based Neural Text Generation with Syntactic Guidance Yinghao Li, Rui Feng, Isaac Rehg, Chao Zhang arXiv [PDF] [code]
   model

   

Multi-signals

1. Plug and Play Language Models: A Simple Approach to Controlled Text Generation Sumanth Dathathri, Andrea Madotto, Janice Lan, Jane Hung, Eric Frank, Piero Molino, Jason Yosinski, Rosanne Liu ICLR20 [PDF] [code]
   model

   
2. A Distributional Approach To Controlled Text Generation Muhammad Khalifa, Hady Elsahar, Marc Dymetman ICLR21 [PDF] [code]
   model

   
   algorithm

   
3. Plug-and-Blend: A Framework for Controllable Story Generation with Blended Control Codes Zhiyu Lin, Mark O. Riedl arXiv [PDF] [code]

Plug-in and Play Framework

1. Fast Lexically Constrained Decoding with Dynamic Beam Allocation for Neural Machine Translation Matt Post, David Vilar NAACL18 [PDF] [code]
2. CGMH: Constrained Sentence Generation by Metropolis-Hastings Sampling Ning Miao, Hao Zhou, Lili Mou, Rui Yan, Lei Li AAAI19 [PDF] [code]
3. Mask and Infill: Applying Masked Language Model to Sentiment Transfer Xing Wu, Tao Zhang, Liangjun Zang, Jizhong Han, Songlin Hu IJCAI19 [PDF] [code]
4. Unsupervised Text Style Transfer with Padded Masked Language Models Eric Malmi, Aliaksei Severyn, Sascha Rothe EMNLP20 [PDF] [code]
5. LEWIS: Levenshtein Editing for Unsupervised Text Style Transfer Machel Reid, Victor Zhong ACL21 findings [PDF] [code]
6. Plug and Play Autoencoders for Conditional Text Generation Florian Mai, Nikolaos Pappas, Ivan Montero, Noah A. Smith, James Henderson EMNLP20 [PDF] [code]
7. Pre-train and Plug-in: Flexible Conditional Text Generation with Variational Auto-Encoders Yu Duan, Canwen Xu, Jiaxin Pei, Jialong Han, Chenliang Li ACL20 [PDF] [code]
8. Plug and Play Language Models: A Simple Approach to Controlled Text Generation Sumanth Dathathri, Andrea Madotto, Janice Lan, Jane Hung, Eric Frank, Piero Molino, Jason Yosinski, Rosanne Liu ICLR20 [PDF] [code]
9. GeDi: Generative Discriminator Guided Sequence Generation Ben Krause, Akhilesh Deepak Gotmare, Bryan McCann, Nitish Shirish Keskar, Shafiq Joty, Richard Socher, Nazneen Fatema Rajani arXiv20 [PDF] [code]
10. Directed Beam Search: Plug-and-Play Lexically Constrained Language Generation Damian Pascual, Beni Egressy, Florian Bolli, Roger Wattenhofer arXiv20 [PDF] [code]
11. Plug-and-Blend: A Framework for Controllable Story Generation with Blended Control Codes Zhiyu Lin, Mark Riedl arXiv21 [PDF] [code]
12. Plug-and-Play Conversational Models Andrea Madotto, Etsuko Ishii, Zhaojiang Lin, Sumanth Dathathri, Pascale Fung EMNLP20 findings [PDF] [code]
13. FUDGE: Controlled Text Generation With Future Discriminators Kevin Yang, Dan Klein NAACL21 [PDF] [code]
14. DExperts: Decoding-Time Controlled Text Generation with Experts and Anti-Experts Alisa Liu, Maarten Sap, Ximing Lu, Swabha Swayamdipta, Chandra Bhagavatula, Noah A. Smith, Yejin Choi ACL21 [PDF] [code]
15. Zero-Shot Controlled Generation with Encoder-Decoder Transformers Devamanyu Hazarika, Mahdi Namazifar, Dilek Hakkani-Tur arXiv21 [PDF] [code]

Constrained Text Generation

1. Lexically Constrained Decoding for Sequence Generation Using Grid Beam Search Chris Hokamp, Qun Liu ACL17 [PDF] [code]
2. Gradient-guided Unsupervised Lexically Constrained Text Generation Lei Sha EMNLP20 [PDF] [code]
3. CGMH: Constrained Sentence Generation by Metropolis-Hastings Sampling Ning Miao, Hao Zhou, Lili Mou, Rui Yan, Lei Li AAAI19 [PDF] [code]
4. Fast Lexically Constrained Decoding with Dynamic Beam Allocation for Neural Machine Translation Matt Post, David Vilar NAACL18 [PDF] [code]
5. Improved Lexically Constrained Decoding for Translation and Monolingual Rewriting J Edward Hu, Huda Khayrallah, Ryan Culkin, Patrick Xia, Tongfei Chen, Matt Post, Benjamin Van Durme NAACL19 [PDF] [code]
6. Directed Beam Search: Plug-and-Play Lexically Constrained Language Generation Damian Pascual, Beni Egressy, Florian Bolli, Roger Wattenhofer arXiv20 [PDF] [code]
7. Language Generation via Combinatorial Constraint Satisfaction: A Tree Search Enhanced Monte-Carlo Approach Maosen Zhang, Nan Jiang, Lei Li, and Yexiang Xue EMNLP20 findings [PDF] [code]
8. POINTER: Constrained Progressive Text Generation via Insertion-based Generative Pre-training Yizhe Zhang, Guoyin Wang, Chunyuan Li, Zhe Gan, Chris Brockett, Bill Dolan EMNLP20 [PDF] [code]

Evaluation

1. Evaluating Style Transfer for Text Remi Mir, Bjarke Felbo, Nick Obradovich, Iyad Rahwan NAACL19 [PDF] [code]
2. A Review of Human Evaluation for Style Transfer Eleftheria Briakou, Sweta Agrawal, Ke Zhang, Joel Tetreault, Marine Carpuat GEM20201 [PDF] [code]

Benchmark

1. Expertise Style Transfer: A New Task Towards Better Communication between Experts and Laymen Yixin Cao, Ruihao Shui, Liangming Pan, Min-Yen Kan, Zhiyuan Liu, Tat-Seng Chua ACL20 [PDF] [code]
2. StylePTB: A Compositional Benchmark for Fine-grained Controllable Text Style Transfer Yiwei Lyu, Paul Pu Liang, Hai Pham, Eduard Hovy, Barnabás Póczos, Ruslan Salakhutdinov, Louis-Philippe Morency NAACL21 [PDF] [code]
   DataSet