Dual Learning for Machine Translation

Abstract

While neural machine translation (NMT) is making good progress in the past two years, tens of millions of bilingual sentence pairs are needed for its training. However, human labeling is very costly. To tackle this training data bottleneck, we develop a dual-learning mechanism, which can enable an NMT system to automatically learn from unlabeled data through a dual-learning game. This mechanism is inspired by the following observation: any machine translation task has a dual task, e.g., English-to-French translation (primal) versus French-to-English translation (dual); the primal and dual tasks can form a closed loop, and generate informative feedback signals to train the translation models, even if without the involvement of a human labeler. In the dual-learning mechanism, we use one agent to represent the model for the primal task and the other agent to represent the model for the dual task, then ask them to teach each other through a reinforcement learning process. Based on the feedback signals generated during this process (e.g., the language-model likelihood of the output of a model, and the reconstruction error of the original sentence after the primal and dual translations), we can iteratively update the two models until convergence (e.g., using the policy gradient methods). We call the corresponding approach to neural machine translation \emph{dual-NMT}. Experiments show that dual-NMT works very well on English$\leftrightarrow$French translation; especially, by learning from monolingual data (with 10% bilingual data for warm start), it achieves a comparable accuracy to NMT trained from the full bilingual data for the French-to-English translation task.

References in corpus (3)

• ADADELTA: An Adaptive Learning Rate Method
• Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation
• On Using Monolingual Corpora in Neural Machine Translation

Cited by in corpus (182)

• Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks
• Deep Visual Domain Adaptation: A Survey
• Deep Reinforcement Learning: An Overview
• An Overview of Deep Semi-Supervised Learning
• Self-training with Noisy Student improves ImageNet classification
• Phrase-Based & Neural Unsupervised Machine Translation
• Unsupervised Machine Translation Using Monolingual Corpora Only
• An Introduction to Image Synthesis with Generative Adversarial Nets
• Survey on reinforcement learning for language processing
• An Effective Approach to Unsupervised Machine Translation
• A Teacher-Student Framework for Zero-Resource Neural Machine Translation
• CoaCor: Code Annotation for Code Retrieval with Reinforcement Learning
• The Missing Ingredient in Zero-Shot Neural Machine Translation
• Tree-to-tree Neural Networks for Program Translation
• COOT: Cooperative Hierarchical Transformer for Video-Text Representation Learning
• Adversarial Neural Machine Translation
• A Probabilistic Formulation of Unsupervised Text Style Transfer
• Unsupervised Translation of Programming Languages
• Unpaired Sentiment-to-Sentiment Translation: A Cycled Reinforcement Learning Approach
• Soft Contextual Data Augmentation for Neural Machine Translation
• Multiple-Attribute Text Style Transfer
• Unsupervised Text Style Transfer using Language Models as Discriminators
• Recommender systems based on graph embedding techniques: A comprehensive review
• A Dual Reinforcement Learning Framework for Unsupervised Text Style Transfer
• Reconstruction Network for Video Captioning
• Towards Binary-Valued Gates for Robust LSTM Training
• Robust Neural Machine Translation with Doubly Adversarial Inputs
• DDTCDR: Deep Dual Transfer Cross Domain Recommendation
• Corpus Augmentation by Sentence Segmentation for Low-Resource Neural Machine Translation
• Closed-loop Matters: Dual Regression Networks for Single Image Super-Resolution
• CycleGT: Unsupervised Graph-to-Text and Text-to-Graph Generation via Cycle Training
• Joint Training for Neural Machine Translation Models with Monolingual Data
• Extracting Parallel Sentences with Bidirectional Recurrent Neural Networks to Improve Machine Translation
• Semi-Supervised QA with Generative Domain-Adaptive Nets
• Meta-Learning for Low-Resource Neural Machine Translation
• Revisiting Low-Resource Neural Machine Translation: A Case Study
• Unsupervised Learnable Sinogram Inpainting Network (SIN) for Limited Angle CT reconstruction
• Unsupervised Speech Recognition
• Zero-Shot Visual Recognition using Semantics-Preserving Adversarial Embedding Networks
• Network Alignment
• Conditional Image-to-Image Translation
• Sequence-to-sequence Pre-training with Data Augmentation for Sentence Rewriting
• Zero-Shot Dual Machine Translation
• WaveCycleGAN2: Time-domain Neural Post-filter for Speech Waveform Generation
• Neural Machine Translation with Monolingual Translation Memory
• Estimating Gradients for Discrete Random Variables by Sampling without Replacement
• Generating Realistic Training Images Based on Tonality-Alignment Generative Adversarial Networks for Hand Pose Estimation
• Non-Adversarial Unsupervised Word Translation
• Learning to Sketch with Shortcut Cycle Consistency
• Explaining and Generalizing Back-Translation through Wake-Sleep
• Triangular Architecture for Rare Language Translation
• Unsupervised Multi-modal Neural Machine Translation
• Dual Metric Learning for Effective and Efficient Cross-Domain Recommendations
• A Simple Baseline to Semi-Supervised Domain Adaptation for Machine Translation
• Scalable Unbalanced Optimal Transport using Generative Adversarial Networks
• CDL: Curriculum Dual Learning for Emotion-Controllable Response Generation
• Synthetic QA Corpora Generation with Roundtrip Consistency
• Multimodal Machine Translation with Reinforcement Learning
• Cross-model Back-translated Distillation for Unsupervised Machine Translation
• Learning Residual Images for Face Attribute Manipulation
• Addressing Semantic Drift in Question Generation for Semi-Supervised Question Answering
• Dual Learning Music Composition and Dance Choreography
• Voice Conversion Using Sequence-to-Sequence Learning of Context Posterior Probabilities
• Retrieving Sequential Information for Non-Autoregressive Neural Machine Translation
• Neural Machine Translation: Challenges, Progress and Future
• DAL: Dual Adversarial Learning for Dialogue Generation
• When and Why is Unsupervised Neural Machine Translation Useless?
• Fork or Fail: Cycle-Consistent Training with Many-to-One Mappings
• Modelling Latent Translations for Cross-Lingual Transfer
• Logician: A Unified End-to-End Neural Approach for Open-Domain Information Extraction
• Duplex Sequence-to-Sequence Learning for Reversible Machine Translation
• Bi-Directional Neural Machine Translation with Synthetic Parallel Data
• Explicit Alignment Objectives for Multilingual Bidirectional Encoders
• Language Style Transfer from Sentences with Arbitrary Unknown Styles
• Disentangling Factors of Variation with Cycle-Consistent Variational Auto-Encoders
• Information Maximizing Visual Question Generation
• Language Graph Distillation for Low-Resource Machine Translation
• Multi-Scale Video Frame-Synthesis Network with Transitive Consistency Loss
• On Synthetic Data for Back Translation
• Neural Machine Translation: A Review of Methods, Resources, and Tools
• Dynamic of Stochastic Gradient Descent with State-Dependent Noise
• Beyond Error Propagation in Neural Machine Translation: Characteristics of Language Also Matter
• Dynamic Data Selection and Weighting for Iterative Back-Translation
• Bi-Mix: Bidirectional Mixing for Domain Adaptive Nighttime Semantic Segmentation
• Language-Driven Image Style Transfer
• Dual Supervised Learning for Natural Language Understanding and Generation
• Greedy Search with Probabilistic N-gram Matching for Neural Machine Translation
• Zero-Resource Neural Machine Translation with Multi-Agent Communication Game
• Knowledge Distillation for Multilingual Unsupervised Neural Machine Translation
• Tag-less Back-Translation
• SJTU-NICT's Supervised and Unsupervised Neural Machine Translation Systems for the WMT20 News Translation Task
• Exploring Explicit Domain Supervision for Latent Space Disentanglement in Unpaired Image-to-Image Translation
• End-to-End Feedback Loss in Speech Chain Framework via Straight-Through Estimator
• Revisiting Iterative Back-Translation from the Perspective of Compositional Generalization
• toon2real: Translating Cartoon Images to Realistic Images
• A Closed-Loop Perception, Decision-Making and Reasoning Mechanism for Human-Like Navigation
• Reconstruct and Represent Video Contents for Captioning via Reinforcement Learning
• Semi-supervised Multimodal Representation Learning through a Global Workspace
• Reciprocal Learning Networks for Human Trajectory Prediction
• Towards Unsupervised Language Understanding and Generation by Joint Dual Learning
• Unsupervised Multimodal Neural Machine Translation with Pseudo Visual Pivoting
• Reference Language based Unsupervised Neural Machine Translation
• Generative Dual Adversarial Network for Generalized Zero-shot Learning
• A Survey of Deep Facial Attribute Analysis
• Learning to Transfer: Unsupervised Meta Domain Translation
• Constructing Emotion Consensus and Utilizing Unpaired Data for Empathetic Dialogue Generation
• Extract and Edit: An Alternative to Back-Translation for Unsupervised Neural Machine Translation
• Injecting Text in Self-Supervised Speech Pretraining
• $\mathcal{G}$-SGD: Optimizing ReLU Neural Networks in its Positively Scale-Invariant Space
• Multilingual Neural Machine Translation for Zero-Resource Languages
• Jointly Learning Semantic Parser and Natural Language Generator via Dual Information Maximization
• Deep Learning and the Global Workspace Theory
• Knowledge Transfer via Dense Cross-Layer Mutual-Distillation
• Playing 20 Question Game with Policy-Based Reinforcement Learning
• Parallel Data Augmentation for Formality Style Transfer
• AR: Auto-Repair the Synthetic Data for Neural Machine Translation
• Deformation-aware Unpaired Image Translation for Pose Estimation on Laboratory Animals
• Multi-task Learning for Multilingual Neural Machine Translation
• Semi-Supervised Wide-Angle Portraits Correction by Multi-Scale Transformer
• The TALP-UPC System for the WMT Similar Language Task: Statistical vs Neural Machine Translation
• Bridging the Imitation Gap by Adaptive Insubordination
• Decoupled Dialogue Modeling and Semantic Parsing for Multi-Turn Text-to-SQL
• Cross-lingual Supervision Improves Unsupervised Neural Machine Translation
• Hierarchical Photo-Scene Encoder for Album Storytelling
• Mimicking the In-Camera Color Pipeline for Camera-Aware Object Compositing
• Using Monolingual Data in Neural Machine Translation: a Systematic Study
• Semantic Parsing with Dual Learning
• Grounded and Controllable Image Completion by Incorporating Lexical Semantics
• Iterative Domain-Repaired Back-Translation
• Dual Learning: Theoretical Study and an Algorithmic Extension
• Exploiting Method Names to Improve Code Summarization: A Deliberation Multi-Task Learning Approach
• Meta Back-translation
• Neural Data-to-Text Generation with LM-based Text Augmentation
• Improving Neural Machine Translation by Bidirectional Training
• Neural-Symbolic Solver for Math Word Problems with Auxiliary Tasks
• Sequence-Level Training for Non-Autoregressive Neural Machine Translation
• Dual Reconstruction: a Unifying Objective for Semi-Supervised Neural Machine Translation
• Large-Scale Optimal Transport via Adversarial Training with Cycle-Consistency
• Neural Machine Translation: A Review and Survey
• Cycle-consistency training for end-to-end speech recognition
• Risk Bounds for Unsupervised Cross-Domain Mapping with IPMs
• Learning How to Self-Learn: Enhancing Self-Training Using Neural Reinforcement Learning
• Guiding the One-to-one Mapping in CycleGAN via Optimal Transport
• Handling Syntactic Divergence in Low-resource Machine Translation
• Duality Regularization for Unsupervised Bilingual Lexicon Induction
• Iterative Batch Back-Translation for Neural Machine Translation: A Conceptual Model
• Uncertainty-Aware Semantic Augmentation for Neural Machine Translation
• Dual Inference for Improving Language Understanding and Generation
• WikiTableT: A Large-Scale Data-to-Text Dataset for Generating Wikipedia Article Sections
• DeepQAMVS: Query-Aware Hierarchical Pointer Networks for Multi-Video Summarization
• Alternated Training with Synthetic and Authentic Data for Neural Machine Translation
• Depth Growing for Neural Machine Translation
• Time-Aware Ancient Chinese Text Translation and Inference
• Augmenting Images for ASR and TTS through Single-loop and Dual-loop Multimodal Chain Framework
• Evaluating a Generative Adversarial Framework for Information Retrieval
• Semi-supervised Text Style Transfer: Cross Projection in Latent Space
• Transformation of Dense and Sparse Text Representations
• Language-Independent Representor for Neural Machine Translation
• Learning from Multiple Sources for Data-to-Text and Text-to-Data
• On the Generative Utility of Cyclic Conditionals
• Dual Attentive Sequential Learning for Cross-Domain Click-Through Rate Prediction
• Learning Dual Retrieval Module for Semi-supervised Relation Extraction
• Bi-Directional Differentiable Input Reconstruction for Low-Resource Neural Machine Translation
• Iterative Dual Domain Adaptation for Neural Machine Translation
• More Robust Dense Retrieval with Contrastive Dual Learning
• Active Learning for Massively Parallel Translation of Constrained Text into Low Resource Languages
• Unsupervised Domain Adaptive Object Detection using Forward-Backward Cyclic Adaptation
• Listening while Speaking and Visualizing: Improving ASR through Multimodal Chain
• Reciprocal Supervised Learning Improves Neural Machine Translation
• Approximate Distribution Matching for Sequence-to-Sequence Learning
• Dynamic DNN Decomposition for Lossless Synergistic Inference
• Learning to Generate Questions with Adaptive Copying Neural Networks
• Jointly Improving Language Understanding and Generation with Quality-Weighted Weak Supervision of Automatic Labeling
• Cycle-SUM: Cycle-consistent Adversarial LSTM Networks for Unsupervised Video Summarization
• Do all Roads Lead to Rome? Understanding the Role of Initialization in Iterative Back-Translation
• Improving Zero-shot Neural Machine Translation on Language-specific Encoders-Decoders
• Multi-Party Dual Learning
• Langevin Cooling for Domain Translation
• Weakly Supervised Recovery of Semantic Attributes
• Improving Multi-Turn Response Selection Models with Complementary Last-Utterance Selection by Instance Weighting
• One to Multiple Mapping Dual Learning: Learning Multiple Sources from One Mixed Signal
• LOGEN: Few-shot Logical Knowledge-Conditioned Text Generation with Self-training