DCN+: Mixed Objective and Deep Residual Coattention for Question Answering

Abstract

Traditional models for question answering optimize using cross entropy loss, which encourages exact answers at the cost of penalizing nearby or overlapping answers that are sometimes equally accurate. We propose a mixed objective that combines cross entropy loss with self-critical policy learning. The objective uses rewards derived from word overlap to solve the misalignment between evaluation metric and optimization objective. In addition to the mixed objective, we improve dynamic coattention networks (DCN) with a deep residual coattention encoder that is inspired by recent work in deep self-attention and residual networks. Our proposals improve model performance across question types and input lengths, especially for long questions that requires the ability to capture long-term dependencies. On the Stanford Question Answering Dataset, our model achieves state-of-the-art results with 75.1% exact match accuracy and 83.1% F1, while the ensemble obtains 78.9% exact match accuracy and 86.0% F1.

arXiv comment10 pages, 6 figures

References in corpus (10)

• Adam: A Method for Stochastic Optimization
• Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation
• A Deep Reinforced Model for Abstractive Summarization
• Hierarchical Question-Image Co-Attention for Visual Question Answering
• SQuAD: 100,000+ Questions for Machine Comprehension of Text
• Multi-Task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics
• Machine Comprehension Using Match-LSTM and Answer Pointer
• Deep Reinforcement Learning for Dialogue Generation
• ReasoNet: Learning to Stop Reading in Machine Comprehension
• Making Neural QA as Simple as Possible but not Simpler

Cited by in corpus (27)

• The Natural Language Decathlon: Multitask Learning as Question Answering
• HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering
• Global-Locally Self-Attentive Dialogue State Tracker
• Multilingual Extractive Reading Comprehension by Runtime Machine Translation
• A Unified MRC Framework for Named Entity Recognition
• Entity-Relation Extraction as Multi-Turn Question Answering
• Reinforced Mnemonic Reader for Machine Reading Comprehension
• Stochastic Answer Networks for Machine Reading Comprehension
• Description Based Text Classification with Reinforcement Learning
• Neural Machine Reading Comprehension: Methods and Trends
• Efficient and Robust Question Answering from Minimal Context over Documents
• The Effect of Natural Distribution Shift on Question Answering Models
• Weaver: Deep Co-Encoding of Questions and Documents for Machine Reading
• A Discrete Hard EM Approach for Weakly Supervised Question Answering
• Attention-Guided Answer Distillation for Machine Reading Comprehension
• Orchestrating the Development Lifecycle of Machine Learning-Based IoT Applications: A Taxonomy and Survey
• Multi-Passage Machine Reading Comprehension with Cross-Passage Answer Verification
• Do Multi-hop Readers Dream of Reasoning Chains?
• HEAD-QA: A Healthcare Dataset for Complex Reasoning
• Making Neural Machine Reading Comprehension Faster
• Look at the First Sentence: Position Bias in Question Answering
• Cross-lingual Machine Reading Comprehension with Language Branch Knowledge Distillation
• An Empirical Study of Content Understanding in Conversational Question Answering
• Relation Module for Non-answerable Prediction on Question Answering
• A Study of the Tasks and Models in Machine Reading Comprehension
• Query-Based Named Entity Recognition
• Knowledge-guided Open Attribute Value Extraction with Reinforcement Learning