Know Yourself and Know Your Neighbour : A Syntactically Informed Self-Supervised Compositional Sentence Representation Learning Framework using a Recursive Hypernetwork

TMLR Paper5251 Authors

Abstract: In this work, we propose a syntactically informed self-supervised learning framework for generating sentence representations. In our framework, we train a recursive hypernetwork to compose sentence representation from any word-level representation by using a set of newly proposed self-supervised tasks. We verify that the newly proposed framework can generate sentence representations that encode more linguistic information than state-of-the-art sentence representations and verify the stability and adaptability of our model.

Submission Length: Regular submission (no more than 12 pages of main content)

Assigned Action Editor: ~Eleni_Triantafillou1

Submission Number: 5251