Unsupervised Inference of Data-Driven Discourse Structures using a Tree Auto-Encoder

TL;DR

This paper introduces an unsupervised method using a tree auto-encoder to infer discourse structures, enabling the creation of large, diverse discourse treebanks without relying on annotated data.

Contribution

It extends latent tree induction with auto-encoding to generate task-agnostic discourse trees in an unsupervised manner, addressing the scarcity of annotated discourse data.

Findings

Enables unsupervised inference of discourse trees

Produces larger, more diverse discourse treebanks

Applicable to various tree-structured objectives

Abstract

With a growing need for robust and general discourse structures in many downstream tasks and real-world applications, the current lack of high-quality, high-quantity discourse trees poses a severe shortcoming. In order the alleviate this limitation, we propose a new strategy to generate tree structures in a task-agnostic, unsupervised fashion by extending a latent tree induction framework with an auto-encoding objective. The proposed approach can be applied to any tree-structured objective, such as syntactic parsing, discourse parsing and others. However, due to the especially difficult annotation process to generate discourse trees, we initially develop such method to complement task-specific models in generating much larger and more diverse discourse treebanks.