End-to-End Synthetic Data Generation for Domain Adaptation of Question Answering Systems

Abstract

We propose an end-to-end approach for synthetic QA data generation. Our model comprises a single transformer-based encoder-decoder network that is trained end-to-end to generate both answers and questions. In a nutshell, we feed a passage to the encoder and ask the decoder to generate a question and an answer token-by-token. The likelihood produced in the generation process is used as a filtering score, which avoids the need for a separate filtering model. Our generator is trained by fine-tuning a pretrained LM using maximum likelihood estimation. The experimental results indicate significant improvements in the domain adaptation of QA models outperforming current state-of-the-art methods.

EMNLP 2020

References in corpus (3)

• PyTorch: An Imperative Style, High-Performance Deep Learning Library
• Question Answering and Question Generation as Dual Tasks
• Learning to Answer by Learning to Ask: Getting the Best of GPT-2 and BERT Worlds

Cited by in corpus (6)

• Zero-shot Generalization in Dialog State Tracking through Generative Question Answering
• Improving Factual Consistency of Abstractive Summarization via Question Answering
• Answering Ambiguous Questions through Generative Evidence Fusion and Round-Trip Prediction
• Towards Robust Neural Retrieval Models with Synthetic Pre-Training
• Generating Self-Contained and Summary-Centric Question Answer Pairs via Differentiable Reward Imitation Learning
• Contrastive Domain Adaptation for Question Answering using Limited Text Corpora