Rethinking Supervised Learning and Reinforcement Learning in Task-Oriented Dialogue Systems.
Ziming LiJulia KiselevaMaarten de RijkePublished in: EMNLP (Findings) (2020)
Keyphrases
- dialogue system
- reinforcement learning
- supervised learning
- human computer
- dialogue management
- natural language
- tutorial dialogue
- natural language generation
- mixed initiative
- temporal difference
- spoken dialogue systems
- function approximation
- kernel based learning
- learning problems
- learning algorithm
- machine learning
- unsupervised learning
- markov decision processes
- learning tasks
- labeled data
- human users
- model free
- state space
- training data
- active learning
- optimal policy
- user model
- reinforcement learning algorithms
- transfer learning
- multi agent
- language understanding
- learning process
- partially observable
- domain independent
- learning agent
- training set
- cooperative
- dynamic programming