DIORA: An unsupervised tree parser and LM

Subendhu Rongali, Andrew Drozdov, Mohit Iyyer, Andrew McCallum
Last updated on Apr 22, 2022

In the past, I worked work with Prof. Andrew McCallum and Prof. Mohit Iyyer on improving DIORA (Deep Inside Outside Representations using Autoencoders), a fully unsupervised method for inducing tree-structure in english sentences based on dynamic programming. I worked on approximating the dynamic program using representating learning and knowledge distillation.

I also worked on a variant of DIORA called S-DIORA, where the soft dynamic program is replaced by a more exact, hard version. This work was in collaboration with Andrew Drozdov and it was published at EMNLP 2020.

Publications

Unsupervised Parsing with S-DIORA: Single Tree Encoding for Deep Inside-Outside Recursive Autoencoders

The deep inside-outside recursive autoencoder (DIORA) is a self-supervised neural model that learns to induce syntactic tree structures …
Project