Deep Tree Transductions - A Short Survey

TL;DR

This survey reviews recent developments in TreeLSTM models for tree-structured transductions, analyzing their biases and effectiveness across various real-world benchmarks.

Contribution

It provides a comprehensive overview of modern TreeLSTM models, discussing their biases and empirical performance on diverse transduction tasks.

Findings

No single TreeLSTM model is universally effective for all transduction problems.

The direction of tree processing significantly influences model bias and performance.

Empirical results highlight the need for diverse approaches depending on the task.

Abstract

The paper surveys recent extensions of the Long-Short Term Memory networks to handle tree structures from the perspective of learning non-trivial forms of isomorph structured transductions. It provides a discussion of modern TreeLSTM models, showing the effect of the bias induced by the direction of tree processing. An empirical analysis is performed on real-world benchmarks, highlighting how there is no single model adequate to effectively approach all transduction problems.