Unified multi-task model for NLP

• Multi-task learning is a blocker for general NLP systems
• Unified model can decider how to transfer knowledge

• Unified, multi-task model can:

  • more easily adapt to new tasks
  • make deploying to production X times simpler
  • lower the bar for more people to solve new tasks
  • potentially move towards to continual learning

    Express many NLP tasks in the same framework

    三类任务

• sequence tagging: name entity recognition, aspect specific sentiment

• text classification: dialogue state tricking, sentiment classification
• seq2seq: machine translation, summarization, question answering

  Multi-task Learning as Question Answering (decaNLP)

  
  

  Multi-task Question Answering Network

  | | | | —- | —- |