Leveraging Natural Supervision for Language Representation Learning: Background Summary

Table of Links

• Abstract

• Acknowledgements

• 1 INTRODUCTION

• 1.1 Overview

• 1.2 Contributions

• 2 BACKGROUND

• 2.1 Self-Supervised Language Pretraining

• 2.2 Naturally-Occurring Data Structures

• 2.3 Sentence Variational Autoencoder

• 2.4 Summary

• 3 IMPROVING SELF-SUPERVISION FOR LANGUAGE PRETRAINING

• 3.1 Improving Language Representation Learning via Sentence Ordering Prediction

• 3.2 Improving In-Context Few-Shot Learning via Self-Supervised Training

• 3.3 Summary

• 4 LEARNING SEMANTIC KNOWLEDGE FROM WIKIPEDIA

• 4.1 Learning Entity Representations from Hyperlinks

• 4.2 Learning Discourse-Aware Sentence Representations from Document Structures

• 4.3 Learning Concept Hierarchies from Document Categories

• 4.4 Summary

• 5 DISENTANGLING LATENT REPRESENTATIONS FOR INTERPRETABILITY AND CONTROLLABILITY

• 5.1 Disentangling Semantics and Syntax in Sentence Representations

• 5.2 Controllable Paraphrase Generation with a Syntactic Exemplar

• 5.3 Summary

• 6 TAILORING TEXTUAL RESOURCES FOR EVALUATION TASKS

• 6.1 Long-Form Data-to-Text Generation

• 6.2 Long-Form Text Summarization

• 6.3 Story Generation with Constraints

• 6.4 Summary

• 7 CONCLUSION

• APPENDIX A - APPENDIX TO CHAPTER 3

• APPENDIX B - APPENDIX TO CHAPTER 6

• BIBLIOGRAPHY

2.4 Summary

In this chapter, we describe the background materials needed for the remainder of this thesis. In Chapter 3, we present our contributions to improving self-supervised training objectives for language model pretraining. The new training objectives help enhance the quality of general language representations and model performance on few-shot learning. Chapter 4 presents our contributions to exploit naturallyoccurring data structures on Wikipedia for entity and sentence representations and textual entailment. Chapter 5 presents our contributions on leveraging freelyavailable parallel corpora for disentangling semantic and syntactic representations. Then we apply the technique to controlling the syntax of generated sentences using a sentential exemplar. Chapter 6 presents our contributed datasets for data-to-text generation, abstractive summarization, and story generation. They are tailored from naturally-occurring textual resources and have unique challenges in their respective task settings.