6 Episodic memory for NLP

 

This chapter covers

  • Applying strongly supervised end-to-end memory networks to sequential NLP problems
  • Implementing a multi-hop memory network that allows for semi-supervised training
  • Strongly supervised vs. semi-supervised memory networks

In this chapter, essentially, we will attempt to extend the use of episodic memory to an array of NLP problems by rephrasing them as instance of Question Answering problems. For the data we will use in this chapter, strongly supervised memory networks easily produce above-baseline results with very little effort. Semi-supervised memory networks produce better accuracy in some cases, but not consistently.

6.1 Memory networks for sequential NLP

6.2 Data and data processing

6.2.1 PP-attachment data

6.2.2 Dutch diminutive data

6.2.3 Spanish part-of-speech data

6.3 Strongly supervised memory networks: Experiments and results

6.3.1 PP-attachment

6.3.2 Dutch diminutives

6.3.3 Spanish part-of-speech tagging

6.4 Semi-supervised memory networks

6.4.1 Semi-supervised memory networks: Experiments and results

Summary

sitemap